Research to Develop a Community G¶rGOŠGOŠ BOOKMOBIÿÿÿ’ t î +î ;î Kî [î kî {î ‹î ›î «î »î Ëî Ûî ëî ûî î î +î ;î Kî [î kî {î ‹î ›î «î »î Ëî Ûî ëî ûî î î +î ;î Kî [î kî {î ‹î ›î «î »î Ëî Ûî ëî ûî î î +î ;î Kî [î kî {î ‹î ›î «î »î Ëî Ûî ëî ûî î î +î ;î Kî [î kî {î ‹î ›î «î »î Ëî Ûî ëî ûî î î +î ;î Kî [î kî {î ‹î ›î «î »î Ëî Ûî ëî ûî î î +î ;î Kî [î kî {î ‹î ›î «î »î Ëî Ûî ëî ûî î î +î ;î Kî ?ª s MOBI ýé]¥
US Department of Housing and Urban Development Office of Policy Development and Research A Research to Develop Community Needs Index
Visit PD&R’s Web Site www.huduser.org to find this report and others sponsored by HUD’s Office of Policy Development and Research (PD&R). Other services of HUD USER, PD&R’s Research Information Service, include listservs; special interest reports, bimonthly publications (best practices, significant studies from other sources); access to public use databases; hotline 1-800-245-2691 for help accessing the information you need.
Research to Develop A Community Needs Index Prepared for U.S. Department of Housing and Urban Development Office of Policy Development and Research Prepared by Econometrica, Inc. Bethesda, MD Frederick J. Eggers, Ph.D. June 2007
Acknowledgements Econometrica, Inc. conducted this research under contract to the U.S. Department of Housing and Urban Development (HUD). Fred Eggers was the Principal Investigator; Alex Thackeray and Fouad Moumen compiled and analyzed the data; and Larry Campbell edited the report. The Econometrica Team benefited from the assistance provided by many persons. James Broughman, George Galster, Garth Taylor, and Christopher Walker served as consultants to the project and contributed many helpful ideas and thoughtful criticisms. Kurt Usowski, Craig Davis, and Ian Keith at HUD provided data and suggestions on how to combine data from different sources. Harold Bunce, John Carruthers, Stanley Gimont, Michael Hollar, Derek Hyra, and Kevin Neary of HUD contributed useful comments on the first two drafts of the report; the final report is much better because of their insights. The Team also appreciates the thoughts and opinions provided by the participants at the Orientation Meeting and at the Expert Panel Meeting. Special thanks are due to two people at HUD who made extraordinary contributions to the success of this research. Marina Myhre served as the Government Technical Representative. Her guidance, comments, and attentive management were crucial to the success of the research. Todd Richardson shared his knowledge and experience generously with the Team and also furnished the 2000 census data used in the analysis. The U.S. Postal Service vacancy series used as a proxy for abandonment were created by Todd Richardson.
FOREWORD In order to understand the most efficient and fair way to allocate Community Development Block Grant funds, HUD staff since 1976 have worked on developing measures of community needs. This study, Research to Develop a Community Needs Index, marks a further advance by developing an index that not only shows current needs but also can be used to demonstrate changing community conditions. The study draws on a number of public databases, including the American Community Survey (ACS). It tests the feasibility of relying on the ACS for annual information about community needs, and it devises a method to compare those needs over time. Specifically this study used 2005 ACS data and other readily-available sources in order to create an index of community needs. It then applied that index to measure changes in community needs since 2000. This study also develops and implements an innovative index of real fiscal capacity, which measures the extent to which communities are capable of dealing with their problems without federal assistance. To construct this index, the study compares the ability of cities to raise revenue from various sources. The “real fiscal capacity” index ranks 234 cities on the real resources that they could have used in order to solve their community needs in 2005. The study finds that it is possible to combine a “needs” index and a “fiscal capacity” index for the purpose of measuring relative need for CDBG and other federal support. This study was limited to exploratory and methodological issues. Darlene F. Williams Assistant Secretary for Policy Development and Research
Contents Executive Summary ........................................................................................................ viii Identifying and Measuring Community Needs........................................................... ix Finding Common Patterns among the Needs Indicators ............................................ xi Comparing Community Needs across Time.............................................................xiii Measuring Fiscal Capacity ........................................................................................ xv Implications for Future Analysis .............................................................................. xvi Recommendations for Future Work .......................................................................xviii 1. Introduction................................................................................................................... 1 1.1. HUD’s Mission and an Index of Community Needs........................................... 1 1.2. Policy Context ..................................................................................................... 2 1.3. History of Research into Community Needs....................................................... 3 1.4. Project Goals........................................................................................................ 4 1.5. Overview of Methodology and Organization of the Report................................ 5 2. Indicators of Community Needs.................................................................................. 9 2.1. Data Issues in Building an Index of Community Needs...................................... 9 2.1.1. Type and Size of Community................................................................ 9 2.1.2. Sources of Data Used to Measure City Needs..................................... 10 2.1.3. Time Consistence of Needs Indicators ................................................ 11 2.2. Selection of Needs Indicators ............................................................................ 11 2.2.1. How the Needs Indicators Were Selected ........................................... 11 2.2.2. Needs Indicators .................................................................................. 12 2.2.3. Other Indicators Considered But Not Used......................................... 18 2.3. Review of Indicators Prior to Index Building.................................................... 18 2.3.1. Correlations Among the Needs Indicators........................................... 18 2.3.2. Analysis of Means ............................................................................... 21 2.3.3. Analysis of the Crime Variables.......................................................... 22 2.4. Assessment of Needs Indicators ........................................................................ 23 3. Factor Analysis, Dimensions of Need, and a Community Needs Index ................. 25 3.1. A Brief Introduction to Factor Analysis ............................................................ 27 3.2. Application of Factor Analysis to Needs Indicators.......................................... 29 3.2.1 Anticipated Results from Factor Analysis............................................ 29 3.2.2. Initial Results from Factor Analysis.................................................... 30 3.2.3. Testing the Robustness of the Factor Analysis.................................... 31 3.2.4. Culling the Needs Indicators ............................................................... 33 3.2.5. Derivation of the Final Factors Used in this Report............................ 33 3.2.6. Calculating Factor Scores.................................................................... 36 iii
Contents (continued) 3.3. Comparing Needs at Different Times................................................................ 37 3.4. Creating a Single-Valued Index of Community Needs ..................................... 39 3.4.1. Alternative Indices............................................................................... 39 3.4.2. Comparisons of Scores on Alternative Indices.................................... 41 3.4.3. Transformation of the Factor Score Functions into Functions of Needs Indicators............................................................................................. 44 3.5. Summary of Factor Analysis ............................................................................. 45 4. Community Needs in 2000 and 2005 ......................................................................... 47 4.1. Changes in Community Needs for Cities with Populations of 65,000 or More 47 4.2. Comparison of Scores in 2000 and 2005 on Factor 1........................................ 49 4.3 Comparison of Scores in 2000 and 2005 on Factor 2........................................ 53 4.4. Comparison of Scores in 2000 and 2005 on Factor 3........................................ 56 4.5. Comparison of Scores in 2000 and 2005 on the Equal Weight Index............... 59 4.6. Summary............................................................................................................ 62 5. Measuring Fiscal Capacity......................................................................................... 65 5.1. General Approach.............................................................................................. 65 5.2. Variables Used to Measure Fiscal Capacity ...................................................... 67 5.3. Variations in Fiscal Capacity............................................................................. 70 5.4. Combining Need and Fiscal Capacity ............................................................... 73 5.4.1. Background.......................................................................................... 73 5.4.2. Simple Options for Combining the Measures ..................................... 74 5.5. An Index of Needs Adjusted for Capacity......................................................... 76 5.6. Summary............................................................................................................ 78 6. Implications ................................................................................................................. 79 6.1. ACS Data and Community Needs ..................................................................... 79 6.2. A Single-Valued Index of Community Needs................................................... 81 6.3. Intertemporal Comparisons of Needs ................................................................ 82 6.4. Factor Analysis Involving Different Geographies............................................. 83 6.5. Measuring Progress at the Tract Level .............................................................. 84 6.6. Measuring Fiscal Capacity ................................................................................ 86 6.7. Changing City Boundaries and Cost-of-Living Differences ............................. 87 6.8. Areas for Future Work....................................................................................... 88 iv
Contents (continued) Appendix A—Supplemental Tables ............................................................................ A-1 Appendix B—Using Regression Analysis to Develop a Community Needs Index.. B-1 B.1. General Concept, Model Specification, and Difficulties ................................ B-1 B.1.1. General Concept...................... .......................................................... B-1 B.1.2. Model Specification .......................................................................... B-3 B.1.3. Difficulties ........................................................................................ B-4 B.2. Implementing the “Hedonic” Approach for City-Level Needs Indicators ..... B-4 B.2.1. Choice of Database ........................................................................... B-4 B.2.2. The Tract-Level Variables ................................................................ B-5 B.2.3. Variation Across Core-Based Statistical Areas ................................ B-8 B.2.4. City-Level Needs Indicators ............................................................. B-8 B.3. Modified Specification of the Hedonic Model ............................................. B-10 B.4. Interpreting the Coefficients of City-Level Needs Indicators....................... B-12 B.5. Use of the “Hedonic” Approach for Assessing Importance of Factors ........ B-13 B.6. Use of Information on City-Level Variables to Create Factor Weights ....... B-15 B.6.1. The Relationship between Beta Coefficients and Standardized Scoring Coefficients..................................................................................... B-15 B.6.2. Extension of Experimental Use of Beta Coefficients to Guide Selection of Weights....................................................................................... B-16 B.7. Insights Gained from the Hedonic-type Analysis......................................... B-17 Appendix C—Comparison of Factor Analysis for 2000 with Richardson Factor Analysis for 2000........................................................................................................... C-1 C.1. Comparison of Index Structure....................................................................... C-1 C.2. Comparison of Need Scores............................................................................ C-3 C.3. Conclusion ...................................................................................................... C-5 References......................................................................................................................D-1 v
Tables Table 1. Needs Indicators for Developing an Index of Community Needs..................... 13 Table 2. Comparison of Means for the Needs Indicators ................................................ 21 Table 3. Rotated Factor Loadings for Final Factor Analysis, Each Factor Sorted by Loading .............................................................................................................. 35 Table 4. Correlations among Factor Scores..................................................................... 37 Table 5. Alternative Single-Valued Community Needs Indices...................................... 40 Table 6. Basic Statistics on the Alternative Indices......................................................... 41 Table 7. Correlations Among the Alternative Indices ..................................................... 42 Table 8. Comparison of Scores between Index 1 and Indices 2, 3, and 4 for 370 Cities 43 Table 9. Transformation of Factor-Scoring Coefficients into Scoring Coefficients for Needs Indicators................................................................................................. 45 Table 10. Average Factor Scores and Average Equal Weight Index Scores in 2000 and 2005................................................................................................................. 48 Table 11. Changes in Factor 1 Scores between 2000 and 2005, by Region and Population ....................................................................................................... 49 Table 12. Forty Cities with the Largest Increases in Factor 1 Scores, 2000-2005 .......... 51 Table 13. Forty Cities with the Largest Decr eases in Factor 1 Scores, 2000-2005......... 52 Table 14. Changes in Factor 2 Scores between 2000 and 2005, by Region and Population ....................................................................................................................... 53 Table 15. Forty Cities with the Largest Increases in Factor 2 Scores, 2000-2005 .......... 54 Table 16. Forty Cities with the Largest Decreases in Factor 2 Scores, 2000-2005......... 55 Table 17. Changes in Factor 3 Scores between 2000 and 2005, by Region and Population ........................................................................................................ 56 Table 18. Forty Cities with the Largest Increases in Factor 3 Scores, 2000-2005 .......... 57 Table 19. Forty Cities with the Largest Decreases in Factor 3 Scores, 2000-2005......... 58 Table 20. Changes in Equal Weight Index Scores between 2000 and 2005, by Region and Population ................................................................................................ 59 Table 21. Forty Cities with the Largest Increases in Equal Weight Index Scores, 2000- 2005................................................................................................................. 60 Table 22. Forty Cities with the Largest Decreases in Equal Weight Index Scores, 2000- 2005................................................................................................................. 61 Table 23. Variables Used to Measure Fiscal Capacity .................................................... 69 Table 24. Twenty-five Cities with Greatest Real Fiscal Capacity................................... 71 Table 25. Twenty-five Cities with the Least Real Fiscal Capacity.................................. 71 Table 26. Real Fiscal Capacity for 25 Largest Cities with Data...................................... 73 Table 27. Impact of Combining Needs and Fiscal Capacity for 50 Cities with the Highest Community Needs............................................................................. 77 vi
Tables (continued) Appendix A Table A.1. Correlation Matrix for Needs Indicators...................................................... A-2 Table A.2. Initial Factor Analysis, 2005 Data: Factor Loading for Unrotated Factors A-4 Table A.3. Initial Factor Analysis, 2005 Data: Factor Loading for Varimax Orthogonal Rotated Factors .......................................................................................... A-5 Table A.4. Factor Analysis Using 2000 Data: Unrotated and Rotated Factor Loading A-6 Table A.5. Factor Analysis Using 2005 Data, Rotated Factor Loadings for Large Cities and Small Cities ......................................................................................... A-7 Table A.6. 2005 Factor Analysis Using VIOLCRIME Instead of PT1CRIME, VARIMAX Rotated Factors ...................................................................... A-8 Table A.7. Standardized Scoring Coefficients, Based on 2005 Data Without Crime Variables .................................................................................................... A-9 Table A.8. Alternative Index Scores............................................................................ A-10 Table A.9. Factor and Equal Weight Index Scores in 2005 and 2000 for 370 Cities.. A-23 Table A.10. Adjusted Needs Index for 234 Cities....................................................... A-37 Appendix B Table B.1. Tract-Level Variables Used in Regressions................................................. B-6 Table B.2. Parameter Estimates for Tract-Level Variables ........................................... B-7 Table B.3. Estimated Coefficients for City-Level Needs Indicators ............................. B-9 Table B.4. Parameter Estimates for Tract-Level Variables with Modified Specification.. ............................................................................................................... B- 11 Table B.5. Estimated Coefficients for City-Level Needs Indicators with Modified Specification ............................................................................................ B-12 Table B.6. Percentage Decline in Median House Prices Caused by a One-Standard- Deviation Worsening in a Need Indicator .................................................. B-13 Table B.7. Estimated Coefficients for the Three Factors............................................. B-14 Table B.8. Percentage Change in Median House Prices Caused by a One-Standard- Deviation Worsening in a Factor Score................................................... B-14 Table B.9. Beta Coefficients and Standardized Scoring Coefficients for Needs Indicators with Valid Hedonic Coefficients ............................................................. B-16 Table B.10. Regressions Using All Tracts with Median House Values ...................... B-19 Table B.11. Regressions on Tracts with Less Dispersion in Median Prices................ B-25 Table B.12. Regressions on Tracts with Less Dispersion in Median Prices Using Factor Scores....................................................................................................... B-31 Appendix C Table C.1. Identical or Similar Indicators in Richardson and This Study ..................... C-3 Table C.2. Comparison of Richardson Index Scores and the Equal Weight Index Scores......................................................................................................... C-4 vii
This page left blank intentionally. viii
Executive Summary The U.S. Department of Housing and Urban Development (HUD) funded this research for the purpose of developing an index of community needs. Such an index would take information from various public databases on different types of community problems and produce an overall assessment of the “neediness” of a community. As far back as 1976, HUD devoted its own staff resources to studying community needs and devising ways to synthesize various types of needs into an overall index of needs. HUD’s efforts have been sporadic because the primary source of data on community needs has been the decennial censuses, and thus new information on needs has been available only at 10-year intervals. Now, the American Community Survey (ACS) will provide every year the information that previously was available only from the decennial censuses. The annual availability of information through the ACS makes a community needs index much more valuable for HUD. From HUD’s perspective, this research would provide the foundation for its future analysis of community needs by: • Testing the feasibility of relying on the ACS for annual information about community needs. • Devising a methodology to compare conditions in communities over time. In the early stages of the research, HUD expanded the goals to include developing an index of fiscal capacity and investigating an alternative methodology for constructing an index. Identifying and Measuring Community Needs The first step in the research was to define the range of problems to be grouped together as “community needs” and to identify “indicators” for each of the problems. In this research, “community” means city and “needs” means the problems, experienced by cities, that are relevant to HUD’s urban mission. The “indicators” are quantitative measures available on a consistent basis for all or most of the cities studied. HUD views its mission as including the support of community development. In its Strategic Plan, HUD declares its concern about a wide variety of problems related to strengthening communities; included among HUD’s concerns are housing conditions, physical conditions, the quality of life, and economic opportunities. With this in mind, the research t eam formulated a preliminary list of indicators that covered a broad array of community ills. In selecting the indicators, the team reviewed measures used in previous studies and data available from a variety of databases that contain information at the city level or that could be manipulated to produce city-level measures. In a November 21, 2006, meeting involving the authors, HUD experts, consultants with previous experience ix
in comparing conditions at the local level, representatives from the U.S. Government Accountability Office, and representatives from the District of Columbia government, the strengths and weaknesses of a variety of indicators were discussed and, in additional consultation with HUD, a final list of 26 indicators was selected. Eight indicators identify population groups that may have needs for city services beyond those of the typical citizen. These include: 1. Poverty population. 2. Children living in poverty. 3. Persons over age 74 living in poverty. 4. Low-income population (excludes poverty population). 5. Single-parent families. 6. Adults without a high school diploma. 7. Working-age persons without a college degree. 8. Recent immigrants. Four indicators identify problems with housing, housing markets, or housing finance. These include: 9. Lack of affordable rental housing. 10. Overcrowded housing. 11. Older rental housing occupied by poor persons. 12. Mortgage-loan denial rate. Three indicators identify the extent to which cities have seriously troubled neighborhoods. These include: 13. Population living in high-poverty census tracts. 14. Population living in moderate-poverty census tracts. 15. Abandoned buildings. Four indicators identify social and economic problems at the city level. These include: 16. Rate of violent crimes. 17. Rate of nonviolent crimes. 18. School-age population living in poverty. 19. Unemployment rate. Four indicators identify conditions that might complicate a city’s efforts to deal with its problems. These include: 20. Linguistic isolation. 21. City-metropolitan differences in minority population. 22. City-metropolitan differences in poverty rate. 23. City-metropolitan differences in median family income. x
xi Three indicators identify detrimental long-term trends. These include: 24. Excess infrastructure/loss of households. 25. Change in employment base. 26. Change in concentration of low-income families. Table 1 in Chapter 2 defines each of these indicators more precisely and explains why each was included. All the indicators are defined in percentage or ratio terms so that their magnitude does not depend on city size. Also, all indicators are defined so that the more serious the condition, the larger the value of the indicator. Input from HUD and outside experts was used to choose the indicators. The list includes several innovative indicators. The abandoned building indicator was developed by HUD staff using a combination of census data and vacancy rates compiled by the United States Postal Service (USPS). The lack of affordable rental housing indicator (#9) uses a technique similar to that used by HUD to identify difficult development areas for the low- income housing tax credit program. The mortgage denial rate is used for the first time in this study. Variants of the city-metropolitan difference variables (#21 to #23) have been used in previous studies, but this research uses a simplified definition that makes it easier to calculate these indicators. All three long-trend indicators (#24 to #26) are new formulations for this research. 1 The indicators require data from the ACS, the decennial censuses, the economic censuses, USPS vacancy surveys, Home Mortgage Disclosure Act records, the FBI Uniform Crime Report, and the Bureau of Labor Statistics Local Area Unemployment Statistics. The 2005 ACS reported information on 473 cities (not inc luding Puerto Rico) with populations of 65,000 or more. The research attempted to calculate the 26 indicators for each city. Information needed for individual indicators was missing for a number of cities. The most serious missing data problem involved the crime data, which were not available for 107 cities. Finding Common Patterns among the Needs Indicators The next step in the analysis was to determine the extent to which the needs indicators can be distilled into a small (more manageable) number of underlying common “themes” or components. The report uses factor analysis to search for common themes and to produce a simpler way to observe how needs vary across communities. Previous HUD research used factor analysis for this purpose. 1 A 27 th indicator based on Housing Mortgage Discrimination Act data was identified—poor housing appreciation in high-poverty neighborhoods—but it could not be implemented within the scope of the project.
The research applied standard factor analysis techniques to the 26 indicators and identified three dimensions of community needs. These include: • Needs associated with poverty and structural problems (Factor 1). • Needs associated with immigration and lack of affordable housing (Factor 2). • Needs arising from limited economic prospects (Factor 3). The robustness of the factor analysis was tested in several ways. First, factor analysis was applied to the same indicators using 2000 data. The 2000 and 2005 analyses identified factors that were nearly identical; this result confirmed that factors developed using 2005 data could be applied to 2000 data on needs indicators. Second, the sample of cities was split into those with populations of 200,000 or more and those with populations of less than 200,000. Factor analysis applied separately to the two samples produced results that were similar enough to suggest that the same pattern of needs apply across different size classes of cities. Third, a different measure for violent crimes—one based on occurrences rather than arrests—was substituted for the measure used in the initial analysis. The results of the factor analysis did not seem to vary significantly when the alternative measure was used. Fourth, the needs indicators were examined to see where problems with missing data caused a large number of cities to drop out of the analysis. Based on this examination, the two crime indicators were dropped. When factor analysis was applied to the smaller set of needs indicators, the same factors were found as were found with the full set of indicators. Eliminating the two crime variables increased the number of cities examined from 292 to 370. The factor analysis based on 24 needs indicators is the one used for all the analyses in the remainder of the report. Factor scores were computed for each city on each factor by multiplying a set of standardized scoring coefficients derived from the factor analysis by the standardized value of the needs indicators for the city and summing the products. A standardized value for a needs indicator is obtained by subtracting a mean value from the value of the indicator for that city and dividing the difference by a standard deviation. The explicit goal of this project was to develop a single-valued index of community needs. The report compared six alternative single-valued indices constructed by using various linear combinations of the scores on the three factors. The report was unable to find any statistical, programmatic, or logical reasons that made a compelling case for choosing one index over any of the others. Statistically, an equal weight index—an index formed by giving each of the three factors a weight of 1/3—produces results that are very similar to the results from the other indices that vary the weights given to the factors. For this reason, the report uses the equal weight index in all the analyses involving a single-valued index. High correlation across all cities does not mean that the ranking of some cities is not substantially different depending upon the index used. If HUD were to use one of these indices to allocate funds to cities, the choice of index would be of great concern to individual cities. But, if HUD is interested primarily in analyzing the variation in needs xii
across cities and over time, then the results from the equal weight index will be similar to those from any index that applies reasonable weights to the factor scores. Comparing Community Needs across Time This research developed a methodology for applying factor analysis to data on needs at two points in time and successfully implemented the methodology. There are two keys to carrying out intertemporal comparisons correctly. • First, the dimensions of need identified in the base year must still be relevant in the comparison year. o The comparison of factor analyses using 2000 and 2005 data confirmed that the same factors applied in 2000 and 2005. • Second, needs must be measured relative to conditions in the year in which the factor analysis is performed—that is, the means and standard deviations from the year used to derive the standardized scoring coefficients must be used to standardize the needs indicators in both years. o Since the report uses 2005 data to identify the factors, the report uses the means and standard deviations calculated on data for the 24 needs indicators in 2005 to standardize the values of the needs indicators in both 2000 and 2005. Using this technique, the report compares conditions in cities in 2000 with conditions in 2005 using each of the factors and the equal weight index. The scores are positively related to needs—that is, for each factor and for the equal weight index, an increase in the score means that a city is worse off in 2005 than in 2000. Between 2000 and 2005, cities—on average—became worse off with respect to poverty and structural problems as well as immigration and housing affordability problems, but improved with respect to the limited economic prospects factor. • Regional differences appeared on the individual factors, such as: o The Northeast had the highest average scores on the poverty and structural problems factor in both 2000 and 2005 and the largest increase in average scores between the two years. The West had the lowest average scores on this factor in both years and the smallest increase between the two years. o For the immigration and housing affordability factor, the average scores of cities in the Northeast and West were higher than the national average xiii
in both 2000 and 2005. Cities in the Northeast had the highest average change between 2000 and 2005. o Cities in the Northeast had the lowest scores on the limited economic prospects factor in 2000 and showed the greatest improvement between 2000 and 2005. • Differences by class size of cities were less common. For example: o There appeared to be a systematic relationship between the scores on the poverty and structural problems factor and city size. The average score declined by size class in both 2000 and 2005. The change in scores was approximately the same for all the size classes, except for cities with populations between 500,000 and one million, which had a slightly higher increase in average scores. o With the exception of cities with over one million residents, there appeared to be little relationship between population size and the prevalence of problems related to immigration and housing affordability. The largest cities had an average score of 0.70 or more in both 2000 and 2005; the national average was 0.00 in 2005. • There were also some interesting patterns in the lists of cities with the biggest increases in scores (becoming worse off) and the lists of cities with the biggest decreases in scores (becoming better off). o Some of the worse off cities on the poverty and structural problems factor experienced big increases on this factor between 2000 and 2005; the cities were Camden, Detroit, Cleveland, Rochester, Reading, and Syracuse. o Compared with other states, California had the most cities—95—among the 370 scored. Still, California cities appeared in higher than expected proportions on the list of the 40 biggest losers and gainers. One would expect, proportionally, 10 cities from California on each list. Instead: − Twenty-four of the 40 cities with the biggest improvements on the poverty and structural problems factor were California cities. − Fifteen of the 40 cities with the worse changes on the immigration and housing affordability factor were California cities. − The five cities with the largest improvements on the immigration and housing affordability factor, and 18 of the top 40, were California cities. The equal weight index showed that, on average, community needs decreased slightly between 2000 and 2005. According to the index, conditions were stable or got better in 202 of 370 cities. However, the report notes that the observed improvement appears to xiv
be related strongly to the substantial increase in the proportion of adults with a high school diploma between 2000 and 2005, a fact that was questioned when the report reviewed data on each of the indicators. Measuring Fiscal Capacity The federal government, in general, and HUD in particular, are interested in developing an index of community needs because they want to know the extent to which communities require federal assistance. But a needs index answers only one-half of this question; the federal government also needs to know the extent to which communities are capable of dealing with their problems without federal assistance. The report develops and implements an index of real fiscal capacity. To construct the index, the report compares cities on their ability to raise revenue from various sources, including assistance from state governments. Then, the report translates the potential revenue into real terms by dividing total potential revenue by the average annual wage for government employees calculated at the metropolitan-area level. Using real capacity adjusts for differences across cities in the costs of responding to community needs. The real fiscal capacity index ranks 266 cities on the real resources that they can potentially use to solve community needs in 2005. The report also develops a technique for combining the equal weight index of community needs with the index of real fiscal capacity to obtain an adjusted needs index that looks at both needs and capacity. The report calculates adjusted needs index scores for 234 cities in 2005. The most important findings from the research on fiscal capacity are: • It is possible to construct an index of real fiscal capacity, which is a very important advancement in analyzing the need for federal assistance. • The index of real fiscal capacity is sensitive to both income and wage rates. Places with high income or lower government wages are more likely to have high real fiscal capacity. High-income places can generally afford more government services because they can raise more tax revenue; places with low government wages can generally afford more government services because every tax dollar goes further in providing services. • The index is negatively correlated with the equal weight index of community needs. Cities with high community needs are more likely to have low real fiscal capacity. xv
• It is possible to combine a needs index and a fiscal capacity index. The adjusted needs index developed in this chapter produced different rankings from the equal weight index of community needs. But, in general, the change in rankings was not great, probably because of the negative correlation between the two component indices. Implications for Future Analysis One objective of this research was to test whether the ACS data would support the same type of analysis that HUD had conducted using long-form data. The answer to this question is “yes.” In the future, HUD can depend on the ACS to monitor conditions in cities and counties. The report successfully uses ACS data to construct useful measures of community needs using factor analysis. Of the 24 needs indicators used in the final factor analysis, 16 used ACS data, one used ACS data combined with long-form data, and four used long-form data. All five indicators that used either long-form data or a combination of ACS and long-form data should be available in the future from the ACS. The following are some issues and open questions that HUD will have to keep in mind in future work using the ACS. • The reporting rules used in the ACS are similar to those used for the long form of the decennial census. But, because the ACS sample size is smaller, the rules can result in more frequent suppression of data. • The Census Bureau has established, as a general policy, releasing for the ACS all tabulations prepared for the 2000 long-form data. However, some special tabulations of long-form data have not yet been released. HUD should probably contact the Census Bureau to make sure that these tabulations are not forgotten. • The ACS has not released data on persons in group quarters yet. So, there has been no experience with the usefulness of the tabulations or the reliability of the data. • The Census Bureau will make revisions to the ACS questionnaire, and revisions always create the possibility of discontinuities in the data. An explicit goal of this project was to develop a single-valued index of community needs. The research achieved this objective, but the outcome was only a qualified success. The report was unable to find any statistical, programmatic, or logical reasons that make a compelling case for choosing one index over any of the others. At HUD’s request, the report examined the use of regression analysis to provide definitive guidance in weighting the factors or the needs indicators. However, the prominent role of housing affordability in Factor 2 and in two or three of the needs indicators undermined attempts to apply the regression results directly. The regression approach did provide some xvi
insights on deriving weights, but the report could not explore the full implication of these insights. HUD indicated early on in the project that it was interested in the lessons from this research that could be applied to measuring needs at the tract level. The Administration has proposed creating a special fund within the CDBG program to award communities for making progress in reducing neighborhood distress. Such a proposal would require a community needs measure at the neighborhood level. Since ACS data will be available at the census-tract level beginning in 2010, it was hoped that the experience gained here in constructing a city-level index using ACS data would be useful to HUD in developing a neighborhood-level index. This research laid the ground work for a measure of progress at the census-tract level in three important ways: the identification of needs indicators, the successful application of factor analysis to the needs indicators, and the development of a methodology for making intertemporal comparisons of needs. Despite these useful insights, HUD will need to do a lot of conceptual and empirical work to develop a technique capable of measuring progress at the local level. The obstacles include: • Several of the needs indicators used at the city level would not be applicable at the tract level because of the absence of data at the tract level or because the concepts behind the indicators are more applicable at the city level than at the tract level. • Because of the substantial change in the number and type of indicators, a new factor analysis would have to be performed at the tract level. This factor analysis is likely to identify different dimensions of need than the three identified at the city level in this report. • The ACS has lower sampling rates than the long-form survey in the decennial censuses. This raises concerns about data suppression at the tract level and about measurement errors. • On the conceptual side, a clear distinction needs to be made between measuring a change in needs and measuring how local government actions have reduced community needs. Conceptually, one would like to control for outside influences so that cities would not benefit from favorable external conditions or suffer from unfavorable external conditions. In this respect, measuring needs is simpler than measuring progress. • At the tract level, gentrification can give the appearance of progress in reducing needs, but progress is not really achieved because many people with needs are forced to relocate with their needs still unmet. xvii
xviii Recommendations for Future Work The most important area for future work is to expand and improve upon the list of needs indicators. This report uses a well-conceived, broad-based, and carefully defined set of needs indicators that provide the basis for a useful factor analysis. However, the greatest payoff for understanding community needs is likely to come from improving these indicators and filling in some missing gaps. Future work should concentrate on getting good measures of education and health needs and, most important, on getting better measures of the impact of long-term economic forces on cities.
Page 1 1. Introduction 1.1. HUD’s Mission and an Index of Community Needs The U.S. Department of Housing and Urban Development (HUD) funded the research reported in this document for the purpose of developing an index of community needs. Such an index would take data from various sources on different types of community problems and produce an overall assessment of the “neediness” of a community. Consistent with the “Urban Development” portion of its name, HUD views its mission as including the support of community development. In its latest Strategic Plan, HUD identifies the following five objectives under the goal of strengthening communities: 2 • Assist disaster recovery in the Gulf Coast region. • Enhance sustainability of communities by expanding economic opportunities. • Foster a suitable living environment in communities by improving physical conditions and quality of life. • End chronic homelessness and move homeless families and individuals to permanent housing. • Address housing conditions that threaten health. These objectives, particularly the last four, indicate HUD’s concern with a wide variety of problems that confront local governments. An accurate and reliable index of community needs would help HUD carry out its responsibilities in several ways. These include: • An index would enable HUD to rank communities by the extent of their needs. • Such a ranking would help HUD develop equitable formulas for distributing funds to communities. • An index would also enable HUD to track whether a community’s needs are improving or getting worse over time. • Information on the components that enter into the calculation of an index score would help HUD diagnose the type of problems facing communities in general and individual communities. 2 HUD Strategic Plan FY 2006 – FY 2011, U.S. Departmen t of Housing and Urban Development, March 31, 2006.
Page 2 • Construction of an index would help HUD understand how the various kinds of community problems relate to one another and the extent to which they represent the same or different types of need. As far back as 1976, HUD devoted its own staff resources to studying community needs and devising ways to synthesize various types of needs into an overall index of needs. HUD’s efforts have been sporadic because the primary source of data on community needs has been the decennial censuses, and thus new information on needs has been available only at 10-year intervals. Now the American Community Survey (ACS) will provide every year the information that previously was available only from the decennial censuses. 3 The annual availability of information through the ACS makes a community needs index much more valuable for HUD. For this reason, HUD contracted with Econometrica, Inc. to build upon HUD’s previous research to develop a community needs index that could be implemented with ACS and other contemporary data to provide yearly information on community needs. 1.2. Policy Context In February 2005, HUD issued a report (Richardson 2005) that measured community needs and analyzed how well the current Community Development Block Grant (CDBG) formula distributes funds with respect to community needs. This report also presented alternative formulas that would distribute CDBG funds more equitably with respect to community needs. This was the sixth in a series of reports on the CDBG formula, but it was the first report that HUD prepared without being requested to do so by Congress. The Administration subsequently proposed changes in the CDBG allocation mechanism. In April 2005, the U.S. Government Accountability Office (GAO) presented to Congress the results of its study of the CDBG formula and testified that the allocation mechanism could be improved. In June 2006, GAO officials testified on the Administration’s proposal and explained how GAO planned to respond to a request from Congress to assess the CDBG formula. The GAO created an expert panel using its National Academy of Sciences connection. The panel was asked to examine: • HUD’s construction of a needs index as a criterion for measuring community needs, including HUD’s factor analysis and the specific indicators of need included in its index. 3 The ACS revolutionizes the way the federal government collects demographic data. The ACS collects virtually the same information annually that the long form of the decennial census collected at 10-year intervals, but the ACS has a lower sampling rate than the long form. In 2006, the Census Bureau released data from the 2005 ACS for most places with populations of 65,000 or more and, thereafter, plans to release ACS data every year for those places. Beginning in 2008, it will release 3-year moving average data for all places with populations of 20,000 or more. Thereafter, it plans to release 3-year moving average data every year for these places. Beginning in 2010, it will release 5-year moving average data for all places, including census tracts and block groups, and, thereafter, plans to release 5-year moving average data every year for all places.
• The development of an evaluation criterion for GAO to use that accounts for the potential mismatch between a jurisdiction’s community needs and its economic and fiscal capacity to meet that need. As of the date of this report, GAO has not completed work on its study. A community needs index based on ACS data would be valuable in assessing proposed changes to the CDBG formula arising from the Administration or GAO. The Administration’s proposal also contained a provision that would create a special fund within the CDBG program to award communities for makin g progress in reducing neighborhood distress. Such a proposal would require a community-needs measure at the neighborhood level. Since ACS data will be available at the census-tract level beginning in 2010, the experience gained here in constructing a city-level index using ACS data should be useful to HUD in developing a neighborhood-level index. 1.3. History of Research into Community Needs Between 1976 and 2005, HUD personnel conducted five studies of community needs: • 1976: An Evaluation of the Community Development Block Grant Formula, prepared by Harold L. Bunce. • 1979: City Need and Community Development Funding, prepared by Harold L. Bunce and Robert L. Goldberg. • 1983: Effects of the 1980 Census on Community Development Funding, prepared by Harold L. Bunce, Sue G. Neal, and John L. Gardner. • 1995: Effect of the 1990 Census on CDBG Program Funding, prepared by Kevin Neary and Todd Richardson. • 2005: CDBG Formula Targeting to Community Needs, prepared by Todd Richardson. These five studies had three common characteristics: First, each study focused on whether the formula used to distribute CDBG funds was doing so effectively and equitably. Second, each study gathered data from a variety of sources on conditions in communities receiving CDBG funding. Variables were selected to measure problems that communities are allowed to use CDBG funds to treat. Third, each study used factor analysis to search out underlying patterns among the need variables and to simplify the data for calculating an index. Page 3
1.4. Project Goals The primary goal of the project is to use 2005 ACS data and other data to create an index of community needs that has the following properties: • HUD can use the index to evaluate the needs of cities with populations of 65,000 or more as of 2005. • In 2007 and every year thereafter, when the Census Bureau releases new ACS data, HUD can enter the new data into the index and update its assessment of city needs. • HUD can rely on the index to track changes in the needs of individual cities over time. • The index, with minor modifications as may be required, can be used to evaluate the needs of smaller cities and urban counties when more detailed ACS data become available in 2008. Using an index to compare needs at two different points in time extends previous work with community needs indices and requires a revised methodology. Each new wave of ACS data will provide new information on individual community needs and the opportunity to construct a new community needs index. Since each index creates its own frame of reference, a single frame of reference has to be selected and criteria developed to ensure the validity of the chosen frame. HUD intended Econometrica, Inc. to build the index using the factor analysis approach employed in HUD’s previous work on need indices. However, the project has some secondary research goals, which are to: • Examine whether previous approaches should be modified to take into account community boundary changes and cost-of-living differences across communities. • Explore an alternative approach for creating a needs index based on hedonic-like regression models. • Explore the development of an index of the capacity of communities to deal with problems. As noted in Section 1.2, the Administration has proposed a special fund within the CDBG program to award communities for making progress in reducing neighborhood distress. HUD hopes to use the lessons from this project to provide insights into developing a means to measure progress at the neighborhood level. In the concluding chapter, the report discusses how the experience from this research at the city level could be applied in creating an index of neighborhood distress at the census tract level. Page 4
1.5. Overview of Met hodology and Organization of the Report The process of developing a community needs index involves a number of steps, each of which has its own conceptual issues that must be resolved. The principal steps that an analyst must undertake are: A. The analyst must establish what concepts should be included in the notion of “community needs.” For the purpose of this research, “community” means city or county. The residents of cities and counties experience a wide range of problems. The first step in developing a community needs index is to identify the subset of these problems that cities and counties have the responsibility of alleviating and that are consistent with the “urban development” mission of HUD. B. The analyst must create valid measures of these concepts. For each concept in Step A, the analyst must find data that adequately represent the problem, that are reliable, and that are available for all the communities being studied. Care must be taken to avoid conceptual errors such as measuring the consequences of not dealing with problems instead of measuring the problems themselves. C. The ability of a city or county to deal with community needs depends upon the resources available to the city or county, that is, on its fiscal capacity, and on conditions, such as long-run economic decline or racial segregation, that may make problems more difficult to resolve. The analyst must be able to identify complicating conditions, determine how to measure them, and figure out how to relate them to direct measures of needs. In addition, the analyst must determine whether it is feasible to measure a community’s capacity to meet its needs and, if so, how to relate capacity to needs. D. The outcome of Steps A, B, and C should be a set of variables that measures needs and complicating conditions for the universe of communities being studied. Next, the analyst must determine the extent to which these measures can be distilled into a small (more manageable) number of underlying common “themes” or components. The report uses factor analysis to search for common themes and to produce a simpler way to observe how needs vary across communities. E. Next, the analyst must decide how the various components of need should be weighted in the creation of a summary index of need. One can look to previous research, relevant legislative guidance, “common-sense rules of thumb,” or other methods to combine the output from Step D into a single index. F. Finally, the analyst must figure out how to use the components of need developed in Step D or the single index developed in Step E to measure needs at a different point in time. This report discusses how each of these steps was accomplished and what was learned in resolving the issues involved in each step. The report provides a list of needs indicators Page 5
that are generally available for all cities with populations of 65,000 or more and that will be available in the future for smaller cities and for counties. From the needs indicators, a set of three factors that summarizes the types of needs associated with the needs indicators was found, and the report used alternative ways to combine the factors into a single index. The report applies the factors and a single-valued index based on the factors to explore relative need among 370 cities in 2005 and to monitor changes in need from 2000 to 2005. This report contains the following six chapters: 1. Introduction. 2. Indicators of Community Needs – Chapter 2 identifies cities as the entities being studied and defines the range of conditions to be considered as “needs” at the city level. It deals with all of the issues involved in Steps A and B and the issues in Step C associated with conditions that make it more difficult for cities to deal with problems. After examining data on various measu res of need, we selected 26 needs indicators to be used in the factor analysis. 3. Factor Analysis, Dimensions of Need, and a Community Needs Index – Chapter 3 applies factor analysis to the data on needs indicators and identifies three common themes (factors) that encompass the conditions measured by the needs indicators. The chapter combines these three factors into a single index using four alternative sets of weights for the factors and compares the alternatives indices. Finally, the chapter develops the methodology to apply the factor analysis developed using 2005 data to measure city needs in 2000. The chapter deals with the issues involved in Steps D, E, and F. As such, it lays out the methodology used in this study and proposed to be used with future rounds of ACS data. 4. Community Needs in 2000 and 2005 – Chapter 4 applies the results from Chapter 3 to examine how cities differ in needs in 2005 and how city needs changed from 2000 to 2005. 5. Measuring Fiscal Capacity – Chapter 5 develops a methodology for measuring fiscal capacity, implements the methodology for 292 cities in 2005, and explores how one could combine a measure of fiscal capacity with a community needs index to obtain a complete picture of the relative dependence of cities on federal aid. This chapter deals with the issues involved in Step C associated with fiscal capacity. 6. Implications – Chapter 6 summarizes the lessons learned in Chapters 2 through 5 and applies them to the three main objectives of this study: developing techniques for measuring community needs that can be used with future rounds of ACS data, developing techniques for tracking changes in needs for individual communities, and exploring ways to measure progress in resolving needs at the neighborhood level. Page 6
Chapter 4 contains the most important empirical results—a comparison of needs in 370 cities in 2000 and 2005. Chapter 5 presents the empirical findings related to fiscal capacity and the joint consideration of community needs and fiscal capacity. This project involved a substantial amount of methodological work, both in conceptualizing and implementing the analysis. Chapter 2 contains the conceptual work related to the selection of needs indicators. Chapter 3 presents the methodology involved in the factor analysis and in applying factor analysis in multiple time periods. Chapter 5 describes the rationale and processes involved in constructing an index of real fiscal capacity and in combining the index of community needs and the index of real fiscal capacity. Appendix B describes the methodology behind the hedonic-type analysis and contains the results of that work. Appendix A contains supplemental tables. Appendix C compares the 2000 factor analysis performed by Richardson in his 2005 study with a 2000 factor analysis using the needs indicators developed in this study. Page 7
This page left blank intentionally. Page 8
2. Indicators of Community Needs HUD designed this study to test the use of ACS data to measure a variety of community needs and track changes in needs over time. This focus shaped the choice of data, time period, and type of communities used in the analysis. The first section of this chapter discusses these choices. The second section describes the range of problems and conditions considered as community needs, proposes a set of needs indicators, and explains why certain variables were not included as needs indicators. The third section examines data on the proposed indicators to test whether they are reliable measures of need prior to the factor analysis in Chapter 3. The fourth section contains our assessment of the accuracy and comprehensiveness of the needs indictors. 2.1. Data Issues in Building an Index of Community Needs This section discusses the issues involved in gathering data for an index of community needs. 2.1.1. Type and Size of Community In this report, community will mean a city with a population in 2005 of 65,000 or more. The recent availability of ACS data motivated this research and, as of now, the Census Bureau has released ACS data only for states and for cities and counties with populations of 65,000 or more. Data on places with populations of 20,000 or more will become available in 2008. Counties are not included in the analysis for two reasons: non-urban counties probably have a different mix of problems, and including them with cities and urban counties could produce misleading results. Second, in the community development area, HUD usually deals with units of governments, which means that it may work with both a county and with cities within that county. This creates data difficulties because, depending upon the issue, HUD may want data at the county level that relate to community needs for both the overall county and individual cities within the county. In awarding CDBG funds to an urban county, HUD considers only the portion of the county outside of cities that receives CDBG funds directly from HUD. Conceptually, it would be possible to construct ACS estimates for these pieces of counties, but the current 65,000-minimum-population rule and confidentiality constraints that limit reporting in individual tables would eliminate numerous urban counties from the analysis. For these reasons, HUD excluded counties from the analysis. The Census Bureau lists 499 places for which it has released ACS data, including 7 in Puerto Rico. We dropped the seven places in Puerto Rico because of problems with particular variables. For five cities (Indianapolis, Louisville, Nashville, Augusta, and Page 9
Page 10 Athens), the information is reported for “balance” of the jurisdiction. Based on correspondence with the Census Bureau, these data cover the consolidated city and county, but omit incorporated places within the consolidated city/county. It is possible that the city/county government is responsible for community needs in the omitted incorporated areas, but there is nothing in the ACS data to confirm or deny responsibility. These five places are included in the analysis. Twenty-seven of the 499 places are “census designated places” (CDPs). Of these, one (Honolulu) is a CDBG central city, and six are CDBG noncentral city entitlement cities. We kept these seven places. 4 New Orleans and other Gulf Coast cities present special problems. Two-thirds of the ACS data for these areas were collected prior to Hurricane Katrina and one-third afterwards, although response rates were probably low. If the primary purpose of this study were to rank cities by need, we might have eliminated many of the Gulf Coast cities because their needs today are probably much different from their needs measured by the 2005 data. 5 But, as noted previously, this study concentrates mainly on developing techniques for measuring need. The ACS data provide a reasonable good measure of conditions in these places prior to Katrina and, therefore, we included them as part of the universe of places used to test the techniques. After these adjustments, the analysis begins with 472 cities; the actual number included in any analysis depends upon how many cities have data for all the measures used in that analysis. Of the 472 cities, 235 are principal cities, and 137 are suburban cities. 2.1.2. Sources of Data Used to Measure City Needs The types of analyses that are used to construct a community needs index require that the data be defined and collected consistently across cities. This means that we must rely on national databases and ignore valuable local data sources. 6 The databases used are: • The 2005 American Community Survey. • The 2000 decennial census. • The 1997 and 2002 Economic Censuses. • United States Postal Service (USPS) vacancy d ata for 2006. • 2005 Home Mortgage Disclosure Act data. • FBI Uniform Crime Report data for 2000 and 2004. 4 Initially we also retained an eighth CDP—Arlington, VA—which is a CDBG urban county. However, Arlington dropped out of the analysis at an early stage because of missing information for some variables. 5 The Census Bureau published a report using ACS data to compared conditions in Gulf Coast states prior to and after Katrina. The report provides data at the state level, distinguishing between the set of counties designated as disaster areas and the balance of the state. See http://www.census.gov/acs/www/Products/Profiles/gulf_coast/index.htm. 6 Examples of relevant local records are city records that bear on neighborhood conditions, such as building code violations or abandoned cars, and county records that bear on real estate conditions, such as property transactions, property valuations, housing court, and evictions. Unfortunately, the methods for collecting and storing these sources of information are not standardized.
• Bureau of Labor Statistics Local Area Unemployment Statistics for 2000 and 2005. Information from other sources—principally, the 2002 Census of Local Governments and the 2001 Residential Finance Survey—was used to adjust data from the ACS and the Economic Censuses to create measures of fiscal capacity. 2.1.3. Time Consistence of Needs Indicators To encompass as wide a range of city problems as feasible, we combined ACS data with data from other nationally available sources. In almost every instance, we used the version of those data collected in 2005 or as close to 2005 as possible. The most recently released FBI uniform crime data were compiled in 2004. The most recent Economic Census covered calendar year 2002. HUD used USPS data (2006) to construct an estimate of abandoned structures. Because of the extensive work involved in calculating this estimate, HUD used the most recent data (2006) and decided not to construct a separate 2005 estimate for this study. In some cases, the Census Bureau has not yet produced tables for the ACS that it published for the 2000 decennial census; in other cases, comparable ACS tables are published, but the tables were empty for a number of our 472 cities because of small sample sizes. Tables reporting tenure, age of structure, and poverty status jointly are examples of the first situation, and the table on overcrowded housing is an example of the second situation. In these situations, we used data from the 2000 census. We also used 2000 census data for those indicators of need that require information at the census-tract level. The ACS will be adding tables in upcoming years, and census-tract data will be available in 2010. 2.2. Selection of Needs Indicators 2.2.1. How the Needs Indicators Were Selected Econometrica team members met with HUD to discuss the research at an Orientation Meeting on October 12, 2006. One of the issues on the agenda was the range of city problems to be considered in gathering data for the needs index. As noted in Chapter 1, the previous HUD studies had focused on needs that are eligible for assistance under the Community Development Block Grant program. The question posed to HUD was whether to focus strictly on problems that can be treated using CDBG funds or to take a wider perspective on community needs. The goals of the CDBG program are broad, and very few things are excluded de facto as eligible activities. So, using CDBG eligibility as the criterion would not significantly limit the types of needs included in the analysis. Page 11
Page 12 Nevertheless, the participants in the Orientation Meeting agreed that the study should adopt a broad definition of community needs—that is, a definition that included needs beyond those typically eligible for unrestrict ed funding under the CDBG program. The participants reasoned that HUD’s mission extends to most problem areas that affect cities and other communities. Given this direction, we investigated a wide range of data sources and developed a list of potential variables for discussion at an expert panel meeting on November 21, 2006. The list drew upon the variables used in the previous HUD studies and ideas developed by GAO for its ongoing study. The participants in the November expert panel meeting stressed certain principles in selecting variables, including: • Variables should clearly relate to city-level needs. • Proxies should be avoided in deference to direct measures of need. • Failure of a city to respond to a problem should not be considered a need. • Variables should be defined to avoid spurious needs, such as the low income of college students who receive support from their families. Using these principles, the panel rejected a number of variables on the list and suggested some additions to the list. In some cases, the panel suggested we investigate alternative measures of particular needs and make the final selection after reviewing the data. A revised list of variables was submitted on December 4, 2006, and work was begun on collecting data to implement the variables. Discussion continued via e-mail on how to construct useful measures from the data collected under the Home Mortgage Disclosure Act (HMDA). 2.2.2. Needs Indicators Table 1 identifies the 27 needs indicators that HUD and Econometrica jointly selected. The list includes variables related to population subgroups with special needs, housing needs, social needs, neighborhood needs, economic needs, conditions that make it more difficult for cities to respond to various needs, and indicators of unfavorable long-run trends. These needs indicators deal with the broad range of problems covered by the five objectives relating to supporting community development in HUD’s strategic plan. Table 1 classifies each indicator with respect to the category of problem that it measures. However, many of the indicators relate to more than one type of problem.
Table 1. Needs Indicators for Developing an Index of Community Needs Page 13 Variable (Short-name Used in Tables) Comments Definition Populations with Needs: The first eight indicators identify subgroups in the populations that may have specialized needs that require the attention and resources of city governments. POOR PERSONS (POORPERS) Poverty in cities has always been a central concern to HUD. The CDBG program requires that cities use 70 percent of program funds to benefit low and moderate income persons. In line with previous research, we eliminate poor college students on the grounds that most receive support from their parents that is not included in income. Ratio of persons age 3 and over not enrolled in college who live in households with below poverty incomes to all persons age 3 and over who live in households. POOR CHILDREN (POORCHILD) Children living in poor households require different and perhaps more city-supplied services than poor working age adults. Percent of persons under 18 (children) in the household population living in households with below poverty incomes POOR ELDERLY (POOROVER74) The elderly require different services and perhaps more city- supplied services than poor working age adults. We chose “over 74” rather than “over 64” for two reasons: (1) with long life spans, “over 74” seems to be a better identifier of “the elderly” who are likely to have special needs, and (2) it is less correlated with overall poor population (POORPERS) and therefore more likely to identify different types of needs. Percent of persons over 74 living in households with below poverty incomes LOW INCOME HOUSEHOLDS (LWINCHHDS) This variabl e was added to pick up low-income households whose incomes exceed the poverty level. The CDBG emphasis on low and moderate income persons argues for including more than just the poverty level population. Percent of persons living in households with incomes greater than the poverty level and less than 50 percent of area median income. Note the decennial census does not contain the table needed to calculate this variable, thus the 2005 ACS data are use for both 2000 and 2005. SINGLE-PARENT FAMILIES WITH CHILDREN (SGLPRNTFAM) Single-parent households frequently require city-supplied services and unsupervised children in some of these households may create neighborhood problems. We chose “single-parent” over “female-headed” because the needs associated with these families are not limited to “female-headed” families. Previous studies had used female-headed families. Percent of families that are single parent-headed with own children under 18. UNEDUCATED POPULATION (UNEDUCADULTS) Adults without a high school diploma generally have lower skills than other workers and may require some support and training during periods of unemployment and may not have adequate preparation for post-employment living support. Percent of household population over 18 without a high school diploma. UNDEREDUCATED WORKING AGE POPULATION (UNDEREDWORKAGE) These workers are more vulnerable to being unemployed and have greater difficulty finding new jobs. Percent of household population over 24 and less than 65 without a college degree. RECENT IMMIGRANT POPULATION (RCNTIMMIG) Language problems and cultural differences create adjustment problems for many members of this group. Percent of household population that is foreign born and entered the United States within the last 15 years.
Page 14 Table 1. Needs Indicators for Developing an Index of Community Needs (continued) Variable (Short-name Used in Tables) Comments Definition Housing Needs: The next five indicators identify problems with housing, housing markets, or housing finance that require city attention or resources or reduce the attractiveness of a city. LACK OF AFFORDABLE RENTAL HOUSING (LACKAFFDRENTALS) There is no good measure of affordability problems in rental housing. HUD has successfully used a close variant of this measure to identify cities where housing costs relative to income justify additional assistance under the Low Income Housing Tax Credit. Ratio of median gross rent (city) to median family income (city). OVERCROWDED HOUSING (OVERCROWD2000) A comprehensive study by the British Government has found potential links between overcrowded housing and health and development problems. Percent of households living in units where the number of person per room is 1.01 or greater. Note this variable is available only for 2000 and the 2000 values are used for both 2000 and 2005. POOR QUALITY HOUSING (PR70RENTPOV) Previous studies have used the percent of housing built prior to 1940 as both an indicator of deteriorated housing and older infrastructure. This variable has been criticized for being an inaccurate indicator of either housing or infrastructure deterioration. Richardson (2005) found that the percent of the housing stock that (1) was built prior to 1970, (2) was rental, and (3) was occupied by a household with below poverty income was a better indicator of poor quality housing. We use the Richardson (2005) indicator. We have a separate indicator of infrastructure problems. Percent of occupied housing units built prior to 1970 and occupied by a poor renter household. Note this variable is available only for 2000 and the 2000 values are used for both 2000 and 2005. DENIAL RATE FOR MORTGAGE APPLICATIONS (DENIAL) This variable identifies cities where lenders are restricting credit because of poor appreciation prospects or some combination of inadequat e income or credit problems on the part of potential buyers. Percent of loan applications denied. POOR HOUSING APPRECIATION IN HIGH POVERTY NEIGHBORHOODS (POORAPPRECHIGH- POVNGHS) HUD and panel members wanted a variable that could discriminate between poor neighborhoods and poor neighborhoods with poor appreciation potential. Percent change between 2000 and 2005 in average mortgage amount on loans in high poverty neighborhoods. (As explained in the text, this variable was not calculated.) Neighborhood Needs: The next three indicators identify the extent to which cities may have seriously troubled neighborhoods. HIGH POVERTY NEIGHBORHOODS (PCTPOPHIGH- POVNGHS Previous research has found that social problems are markedly greater in neighborhoods with a high percentage of poor persons. The research typically uses 40 percent as the crucial percentage. Percent of city population living in census tracts with poverty rates of 40 percent or higher. Note this variable is available only for 2000 and the 2000 values are used for both 2000 and 2005. MODERATE POVERTY NEIGHBORHOODS (PCTPOPMOD- POVNGHS) Since there is a relationship between concentrated poverty and neighborhood problems, we included this variable to identify neighborhoods – other than the highest poverty neighborhoods – where the poverty concentration may be a problem. Percent of city population living in census tracts with poverty rates greater than or equal to 20 percent but less than 40 percent. Note this variable is available only for 2000 and the 2000 values are used for both 2000 and 2005.
Table 1. Needs Indicators for Developing an Index of Community Needs (continued) Page 15 Variable (Short-name Used in Tables) Comments Definition ABANDONMENT (PCTVACMOD- POVCITY) Abandoned buildings are a blighting influence and could affect community health. HUD has always wanted a reliable measure of abandoned building but data on abandonment are neither universally nor consistently collected at the city level. To solve this problem, HUD analysts have collected data on vacant housing units from the USPS and have counted the number of such structures in moderate to high poverty neighborhood under the presumption that vacant units in such neighborhood have a high probability of being or becoming abandoned. Ratio of vacant housing units (from 2006 USPS surveys) in tracts with 20 percent or more poor (identified from 2000 census) to total housing units in city (from 2005 ACS). Note only one version of this variable is available and is used for both 2000 and 2005. City-Wide Social or Economic Problems: The next four indicators identify social or economic problems at the city level. PART 1 CRIME (PT1CRIME) Part 1 crimes include violent crimes and serious nonviolent crimes. Crimes of this nature are a social problem in themselves and have a blighting influence on neighborhoods. Number of part 1 crimes per 100,000 population. Based on 2004 FBI data on arrests for murder, rape, burglary, motor vehicle theft, arson, and other part 1 crimes - see http://www.fbi.gov/ucr/cius_04/appendices/appendix_02.html for definitions of Part 1 and Part 2 crimes. PART 2 CRIME (PT2CRIME) Part 2 crimes include offenses that are less serious but that nevertheless reduce the quality of life of city residents. Number of part 2 crimes per 100,000 population. Based on 2004 FBI data on arrests for forgery, fraud, simple assault, prostitution, drug offenses, drunkenness, disorderly conduct, and other part 2 crimes - see http://www.fbi.gov/ucr/cius_04/appendices/appendix_02.html for definitions of Part 1 and Part 2 crimes. POOR SCHOOL AGED POPULATION (SCHPOPPOOR) This variable was included as a measure of the problems faced by a city in carrying out its responsibility to provide quality education to its youth. Percent of the school aged population (between 5 and 17) living in households wi th below poverty income. UNEMPLOYMENT RATE (UNEMPCEN) We chose the unemployment rate measured by the ACS and the decennial census over a Bureau of Labor Statistics variable for two reasons: (1) its is calculated from sample data not estimated by a model and (2) its less precise definition of labor force may be successful at capturing disguised unemployment, that is, the unemployment of persons who have left the labor force because of discouragement. Percent of household population over 16 that is unemployed and looking for work (in labor force). This variable is calculated from the 2000 decennial census or the ACS. Conditions that Complicate Dealing with Other Problems: The next four variables were added to identify conditions that might make it more difficult for cities to deal with the needs of people, housing, neighborhoods, or general social or economic problems. LINGUISTICALLY ISOLATION (LINGISOL) This variable identifies language difficulties that may complicate a city’s efforts to provide services and may generate the need for additional services. Percent of households in which all adults (high school age and older) have some limitation in communicating in English. (A household is classified as “linguistically isolated” if no household members age 14 years and over spoke only English, and no household members age 14 years and over who spoke a language other than English spoke English “Very well.”)
Page 16 Table 1. Needs Indicators for Developing an Index of Community Needs (continued) Variable (Short-name Used in Tables) Comments Definition MEASURE OF RACIAL DISSIMILARITY (MINCON) Cities in highly segregated metropolitan areas may experience additional difficulties in providing ordinary services and will have to deal with segregation and its consequences. The CDBG program has as an objective promoting an increase in the diversity of neighborhoods. Ratio of minority population rate in city to minority population rate in metropolitan area INDEX OF ECONOMIC DISSIMILARITY (POVCON) Cities in areas where the poverty population is concentrated may experience additional difficulties in providing ordinary services and will have to deal with poverty concentration and its consequences. The CDBG program has as an objective reducing the isolation of income groups. Poverty rate in city divided by poverty rate in metropolitan area LOCAL FISCAL DISPARITY (MEDINCCBS2CITY) Disparity in incomes between central cities and suburbs make it difficult for central cities to meet their needs and can create disparity problems that affect the entire local economy. Median family income of metro area relative to median family income of jurisdiction Long-run Decline: The final three indicators identify cities that are suffering from long-run decline. EXCESS INFRASTRUCTURE (EXCSINFRA) The panel initially focused on this variable as a good indicator of the extent to which a city may be faced with maintaining more infrastructure than it needs. The indicator also identifies declining cities. Ratio of maximum population measured in households (without reference to boundary changes) at 1970, 1980, 1990, and 2000 to current population CHANGE IN EMPLOYMENT BASE (CHNGEMPLOYBASE) This variable focuses on the recent performance of the city economy. It compares growth in the labor force to growth in actual jobs within the city. The ratio of two ratios: the first ratio is labor force in 2005 to labor force in 2000 from BLS; the second ratio is jobs in the city from the 2002 economic census to jobs in the city from the 1997 economic census. CHANGE IN THE CONCENTRATION OF LOW INCOME FAMILIES (CHGLOWINCCON) This variable measures how well incomes in the city are keeping pace with incomes throughout the country with special attention to a city’s relative share of low income families. Calculate the proportion of families in a city th at have incomes in the bottom quintile for all families in the country and then take the ratio of this proportion in 2005 (or 2000) and divides by the proportion in 1970 (based on 1969 income).
Page 17 We defined each indicator in percentage or per capita terms or as a ratio, so that the value of the indicator would depend only on conditions in a city and not on city size. We also defined each indicator in such a way that an increase in the value of the indicator means that conditions measured by that indicator have worsened. Consistent definition of the indicators will make it easier to interpret the factor analysis in Chapter 3. The poverty variable (POORPERS) is based on an estimate from the ACS, using a national poverty-level income, of the number of poor persons. 7 It is reasonable to expect that the consequences of poverty are greater in high-cost areas than in low-cost areas. Because the count is based on data available only to the Census Bureau, there is no easy way to adjust these data for cost-of-living differences. UNEDUCADULTS and UNDEREDWORKAGE were developed based on similar but slightly different concerns, and therefore are defined using different age qualifications. UNEDUCATDULTS refers to all persons over age 18, whereas UNDEREDWORKAGE refers to the 18-65 years-old population. In both cases, lack of education was considered to place persons at greater risk of unemployment, and therefore the focus on working age is appropriate for both variables. In addition, persons without a high school education may not have had the earning capacity during their working years to adequately prepare themselves for retirement. Therefore, UNEDUCADULTS also focuses on persons over age 65. This different focus creates some problems in Appendix B, but the rationale seems reasonable. The recent immigrant population variable (RCNTIMMIG) counts all persons who were foreign born and immigrated to the United States during the previous 15 years, including both citizens and non-citizens. According to the Census Bureau: 8 The American Community Survey questionnaires do not ask about immigration status. The population surveyed includes all people who indicated that the United States was their usual place of residence on the survey date. The foreign-born population includes naturalized U.S. citizens, Lawful Permanent Residents (immigrants), temporary migrants (e.g., foreign students), humanitarian migrants (e.g., refugees), and unauthorized migrants (people illegally present in the United States). Legal and illegal immigrant households can present similar problems for local governments, most notably, language difficulties in the workplace, language difficulties in schools, and the need for medical services. To the extent that the Census Bureau is successful in including illegal immigrants in the ACS, they should be counted in this indicator. HMDA data were used to construct the DENIAL RATE FOR MORTGAGE APPLICATIONS indicator. 7 There are separate poverty levels for Alaska and Hawaii. 8 American Community Survey, Puerto Rico Community Survey, 2005 Subject Definitions, page 31, http://www.census.gov/acs/www/UseData/Def.htm.
Page 18 We also planned to use HMDA data to identify low-income neighborhoods with stagnant or declining housing markets. This is the POOR HOUSING APPRECIATION IN HIGH POVERTY NEIGHBORHOODS indicator listed in Table 1. The construction of this indicator proved to be too complicated for the limited scope of this project, and therefore this indicator is not used in the subsequent analysis. 9 2.2.3. Other Indicators Considered But Not Used We considered a large number of potential indicators and excluded many for various reasons. The following two exclusions deserve additional discussion: • Persons with a Disability Limiting Employment: We had originally planned to u se this variable. We dropped it because changes in the skip pattern used to ask this question appear to have produced a substantial downward shift in the percentage between 2000 and 2005. • Decline of the Middle Class: We constructed an indicator that focused on the proportion of families in a city that are middle-income families. We defined middle income as having an income higher than the incomes of the poorest 20 percent of American families but lower than the richest 20 percent of American families. We took the ratio of this proportion in 2005 (or 2000) to the ratio in 1970 to determine whether the city was gaining or losing middle-class families. We decided not to use this variable because a city can have a lower proportion of middle-class families as a result of growing poorer or growing richer. Among the 100 cities that had the largest decline in middle-income families between 1970 and 2005, 34 also had a decline in the proportion of poor families. In these 34 cities, only the proportion of rich families was growing. We used CHANGE IN THE CONCENTRATION OF LOW-INCOME FAMILIES as an indicator of long-term trends instead of the decline in the middle-class indicator. 2.3. Review of Indicators Prior to Index Building 2.3.1. Correlations Among the Needs Indicators After gathering data on the needs indicators for both 2000 and 2005, we examined the distribution of each variable and its correlation with the other variables to determine 9 Construction of the variable would require HMDA data from two different years. The main problem is that the 2000 HMDA data do not contain a variable to identify mobile homes. In constructing the DENIAL variable, we eliminated investor loans and mobile-home loans because of concern that their inclusion would affect the results. We had the same concern about the PRICE APPRECIATION indicator. To eliminate mobile homes would require matching the HMDA data to a list prepared by HUD analysts of lenders who specialize in mobile-home lending.
whether the indicator is performing as anticipated and to uncover any problems with the indicator. It was this analysis that led to the elimination of the employment disability and decline of the middle-class indicators discussed above. The first test in the correlation analysis was to determine if any of the needs indicators is highly correlated with population. As noted, we defined the indicators so that their values should be independent of city size and, therefore, we expected to find low correlations between population and the various needs indicators. None of the 26 variables had a correlation with population greater in absolute value than 0.20. Next, we examined the correlations among the needs indicators. This analysis provides some prior indication of how the factor analysis will sort the indicators and can identify problem with the indicators as implemented. Table A.1 in Appendix A reports these correlations. The most interesting findings from the correlation analysis were: • POORPERS has correlations of 0.60 or higher with 15 other indicators. These include all but two of the variables that use income or poverty in their definitions, but also include SGLPRNTFAM, DENIAL, UNEMPCEN, MINCON, UNEDUCADULTS, and EMPLOYDISB. • UNEMPCEN seems to correlate with the same variables with which POORPERS correlates, but at lower rates. • OVERCROWD_2000, the crime variables, and RCNTIMMIG have low correlations with POORPERS. • RCNTIMMIG has correlations of 0.60 or higher only with LINGISOL and OVERCROWD_2000. • POOROVER74 is weakly related to all the other variables. • The crime indicators (PT1CRIME and PT2CRIME) correlate most highly with each other, but the correlation is only 0.53. These crime indicators correlate weakly with all the other variables. The low correlation of the crime variables with each other is puzzling and, as we shall see in Chapter 3, these variables do not perform well in the factor analysis. We discuss these indicators more in Section 2.3.3. MEDINCCBS2CITY, POVCON, and MINCON are indicators that measure conditions recognized in previous studies as either problems in themselves or as factors that complicate the solution of other problems. While the concepts behind POVCON and MINCON are not new, the definitions of these indicators are new. Richardson (2005) used a dissimilarity index to measure the extent of racial segregation. We considered using dissimilarity indices to measure both racial and income segregation, but were persuaded in the November 21 meeting that the definitions in Table 1 were simpler to Page 19
implement and provided much the same information. The correlation analysis reveals no problems with these variables as defined. • MEDINCCBS2CITY correlates highly with the other variables we used to characterize city/suburb differences: POVCON (0.86) and MINCON (0.73). It also correlates highly with the poverty variables. Six of the needs indicators in Table 1 were defined for the first time in this study; these are: LACKAFFDRENTALS, DENIAL, PCTVACMODPOVCITY, EXCSINFRA, CHNGEMPLOYBASE, and CHGLOWINCCON. • LACKAFFDRENTALS correlates most highly with UNEDUCADULTS (0.66), a somewhat surprising result. It has only modest to low correlations with the other housing variables: OVERCROWD_2000 (0.58), PR70RENTPOV (0.49), DENIAL (0.28), and PCTVACMODPOVTOCITY (0.15). LACKAFFDRENTALS appears to pick up different types of housing problems than the other housing indicators. • DENIAL correlates most highly with PCTVACMODPOVTOCITY (0.76) and has correlations above 0.60 with some of the indicators of population groups with special needs and with UNEMPCEN, MINCON, and MEDINCCBS2CITY. • PCTVACMODPOVTOCITY correlates most highly with DENIAL (0.76) and has correlations of greater that 0.60 with seven other indicators, including poor persons, poor children, poor school-aged children, excess infrastructure, poor quality rental housing, and minority concentration. • EXCSINFRA correlates most highly with PCTVACMODPOVTOCITY (0.64) and has correlations over 0.50 only with DENIAL and PR70RENTPOV. DENIAL appears to be picking up some “older city” problems. • CHNGEMPLOYBASE does not correlate highly with any of the other variables; its highest correlations are with the education variables, UNDEREDWORKAGE (0.18) and UNEDUCADULTS (0.13). The correlation results for CHNGEMPLOYBASE are not disturbing. It is intended to identify a type of need different from that of the other indicators and some association with education limitations should be expected. • CHGLOWINCCON has modest correlations with three variables and low correlations with the remaining variables; its highest correlations are with MEDINCCBS2CITY (0.47), DENIAL (0.44), and LWINCHHDS (0.42). Page 20
2.3.2. Analysis of Means After the correlation analysis, we compared the means of the indicators in 2000 and 2005 for two reasons: to identify data errors and to obtain a sense of how conditions changed between 2000 and 2005. Table 2 reports these comparisons. Table 2. Comparison of Means for the Needs Indicators Variable 2005 Mean 2000 Mean Absolute Difference Percent Difference POORPERS 0.140 0.125 0.0148 11.8% POORCHILD 0.207 0.181 0.0265 14.7% POOROVER74 0.109 0.108 0.0002 0.2% SGLPRNTFAM 0.181 0.160 0.0201 12.5% UNEDUCADULTS 0.169 0.204 -0.0350 -17.2% UNDEREDWORKAGE 0.691 0.715 -0.0239 -3.3% RCNTIMMIG 0.099 0.091 0.0081 8.9% LACKAFFDRENTALS 0.180 0.159 0.0208 13.1% SCHPOPPOOR 0.198 0.175 0.0233 13.3% UNEMPCEN 0.076 0.066 0.0096 14.5% LINGISOL 0.071 0.060 0.0109 18.0% MINCON 1.281 1.255 0.0267 2.1% POVCON 1.196 1.139 0.0573 5.0% MEDINCCBS2CITY 1.125 1.081 0.0436 4.0% EXCSINFRA 1.025 1.015 0.0105 1.0% CHG LOWINCCON 1.245 1.310 0.0650 5.2% PT1CRIME 957.036 1004.970 -47.9337 -4.8% PT2CRIME 4133.420 4311.42 -178.0000 -4.1% LWINCHHDS 0.299 0.299 Same data Same data OVERCROWD_2000 0.083 0.083 Same data Same data PR70RENTPOV 0.052 0.052 Same data Same data PCTVACMODPOVCITY 0.010 0.010 Same data Same data DENIAL 0.220 0.220 Same data Same data PCTPOPHIGHPOVNGHS 0.026 0.026 Same data Same data PCTPOPMODPOVNGHS 0.175 0.175 Same data Same data CHNGEMPLBASE 0.836 0.836 Same data Same data For the last eight indicators in Table 2, we were unable to calculate values for both 2000 and 2005, and therefore used the same values of these variables for each city in both years. We had data only for 2000 for OVERCROWD_2000, PR70RENTPOV, PCTPOPHIGHPOVNGHS, and PCTPOPMODPOVNGHS. We had data only for 2005 for LWINCHHDS, PCTVACMODPOVCITY, DENIAL, and CHNGEMPLBASE. Therefore, Table 2 does not calculate the absolute or percent differences for the means of these eight indicators. Only two of the changes in means are surprising. The 17-percent decline in uneducated adults seems remarkably large for social statistics that frequently change at glacial rates. However, the national data show a 25-percent decline in this ratio. The Census Bureau Page 21
Page 22 report comparing the 2000 decennial census with the ACS-type C2SS collected in 2000 reveals no substantial differences between the two surveys on educational attainment. The ACS Subject Definition document mentions changes in the questions prior to 1999, but no changes that would have affected the 2000 comparisons. The 18-percent increase in linguistic isolation also seems large. The base is small, and the national change was a 16-percent increase. Except for the crime and education indicators, all the needs indicators have higher mean values in 2005. This suggests that conditions in cities, on average, worsened over the period from 2000 to 2005. 2.3.3. Analysis of the Crime Variables The results involving the crime variables were puzzling. The 2004 versions of PT1CRIME and PT2CRIME had only a 0.53 correlation with each other and had low correlations with all the other variables. Both indicators had low loading on all the factors and, as noted in Chapter 3, the test of sampling adequacy indicated that PT1CRIME was not a good candidate for factor analysis. All the previous studies used some indicator of crimes, and our panel of experts concurred in the inclusion of the two crime measures among the needs indicators. So, at HUD’s request, we investigated why the crime indicators did not perform better. First, we examined the correlations between our crime variables in 2000 and 2004. It turns out that PT1CRIME as measured in 2000 had a correlation of only 0.20 with PT1CRIME measured in 2004, and PT2CRIME as measured in 2000 had a correlation of only 0.27 with PT2CRIME measured in 2004. These low correlations heighten our concern about the crime data. The FBI distinguishes between Part 1 crimes and Part 2 crimes in its Uniform Crime Report data. In the Part 1 crime data, the FBI reports separately on: criminal homicide, forcible rape, robbery, aggravated assault, burglary, larceny-theft, motor vehicle theft, and arson. Normally, the FBI data on Part 1 crimes include both known offenses and those cleared by arrests, but the data we received contained only arrest records. In retrospect, we are not sure why the Department of Justice gave us only arrest data for Part 1 crimes. The Uniform Crime Report data on Part 2 crimes are based solely on arrest data; perhaps the DOJ data staff provided the Part 1 and Part 2 data on a matching basis—that is, using information from arrests only. From the FBI Web site, we downloaded data on Part 1 crimes for 2005; these data include both arrests and known occurrences. 10 The correlation between these data, which the FBI Web site label “violent crime,” and our PTICRIME variable (b ased on 2004 data) was 0.20. It appears that either there are substantial differences between the arrest data 10 http://www.fbi.gov/ucr/05cius/data/table_08.html.
and the combination of arrests and known offenses, or there are problems with our arrest data. (We checked the programs used to read the data carefully.) The correlations between VIOLCRIME and the other needs indicators are also reported in Table A.1 in Appendix A. VIOLCRIME has correlations between 0.5 and 0.6 with DENIAL, PCTVACMODPOVCITY, POORCHILD, POORPERS, SCHPOPPOOR, and SGLPRNTFAM; all the other correlations are less than 0.5. While VIOLENT crime has stronger associations with other needs indicators than PT1CRIME, its correlations with the other indicators are modest to low. 2.4. Assessment of Needs Indicators Table 1 includes eight indicators that are designed to identify population groups that may have needs for city services beyond those of the typical citizen. The variables measure separately the poor and low-income persons, poor children, the elderly poor, single- parent households, persons with limited education, and immigrants. The ACS provides detailed information on important subgroups, and it should be possible to implement all eight indicators in future years for cities of all sizes and counties, including HUD-defined urban counties. Table 1 lists five indicators of problems with housing, but we were only able to implement four of the indicators. Of these four, two used 2000 census data. ACS data were used only for LACKAFFDRENTALS. In the future, however, HUD should be able to use ACS data to estimate the overcrowded housing variable and the poor quality rental housing variable as well as this variable for cities of all sizes and for counties. Also in the future, HUD will be able to estimate the DENIAL indicator using HMDA data and perhaps will be able to use these data to create a useful indicator that identifies high- poverty neighborhoods with poor appreciation of owner-occupied housing. At this time, the housing indicators are limited in scope, but some of these limitations appear to be temporary. Table 1 contains three indicators of the extent of neighborhood problems. All three variables are based on tract-level data from the 2000 decennial census. But beginning in 2010, HUD should be able to employ ACS tract-level data in these indicators. The primary indicator is proportion of the population living in high-poverty tracts, defined as tracts with a poverty rate of 40 percent or more. The urban literature has highlighted these neighborhoods as having problems related to the concentration of poverty. The abandonment variable was developed by HUD in an attempt to identify problems associated with abandoned buildings. The distribution and correlation analyses suggest that the abandoned building indicator is working reasonably well. Table 1 contains only four indicators of city-wide social or economic problems. As noted above and discussed further in Chapter 3, the crime variables seem to be surprisingly unrelated to other indicators of need. We were unable to find any city-level data on health needs, and we were able to find only one measure to associate with problems in Page 23
education. For our education indicator, we chose percentage of the school-age population that is poor, because it appears to represent the problem facing a city in providing quality education. There were three reasons for the lack of more measures related to education: inconsistent definition of measures across schools and school districts, difficulty converting the education data provided by the federal government for schools and school districts into city-level measures, and concern that some of the measures represented failures on the part of cities to meet their education responsibilities rather than difficulties of the education challenges pres ented to cities. With respect to the third reason, test scores are an example of a measure that could indicate either poor performance or difficult-to-educate populations. Table 1 contains measures of four conditions—LINGISOL, MEDINCCBS2CITY, POVCON, and MINCON—that we believe complicate the ability of cities to deal with their problems. All four indicators can be easily calculated using ACS data and should be available in the future for all cities and for all counties. Table 1 contains three indicators of unfavorable long-run trends that affect cities. EXCSINFRA both measures a condition that is a problem in its own right—having to maintain more infrastructure per household than the typical city—and identifies cities in long-run decline—that is, cities that are losing households. CHGLOWINCCON identifies cities that are losing their middle- and upper-income households. Both indicators represent important dimensions of long-run change at the city level. A third dimension is economic vitality. The indicator chosen to represent this dimension is limited in three ways: the indicator changes only every 5 years; its calculation requires the integration of data from two different sources; and the confidentiality rules applied to the Economic Census data introduce noise into the calculation. At this stage, we believe that we have very good indicators of the needs associated with certain demographic groups, fairly good indicators of housing problems, good indicators of neighborhood-level problems, only marginally adequate measures of city-wide social and economic problems, good measures of complicating conditions, and adequate measures of two out of three of the key long-term trends. As more ACS data become available, the problems with the housing indicators should be eliminated, and all of the indicators should be available for all cities, for counties, and for HUD-defined urban counties. Future work to improve indicators of community needs should focus on: • Getting good measures of education and health needs. • Testing other formulations of the national crime data to find a more robust crime indicator. • Getting a better measure of the impact of long-term economic forces on cities. Page 24
Page 25 3. Factor Analysis, Dimensions of Need, and a Community Needs Index In developing an index of community needs, analysts face two problems. These can be succinctly, and fairly accurately, described as “too much data” and “not enough data.” 11 The ACS and other national data sets provide a wealth of information on persons and housing, the local economy, and some social problems such as crime. Using these sources, we constructed 26 indicators of need. We could have included many more, such as per capita income, persons with disabilities, and retail and wholesale jobs. 12 The previous chapter described how we worked with HUD and other experts to narrow down the list of potential indicators. Now having selected indicators, we need to figure out how to make sense of so much information. How does a one-percentage-point higher unemployment rate compare in neediness to a three-percentage-point higher poverty rate or 15 more serious crimes per 100,000 population? Does the overcrowded housing indicator simply indicate low income or high housing prices and therefore duplicate the poverty or lack of affordable housing indicators, or does it represent a different type of problem? Questions like these need to be answered in order to create a mathematical formula that translates the needs indicators into a single number. On the other hand, despite all the data available from the Census Bureau and other sources, our information on community needs is incomplete. While we have multiple indicators of needs related to specific demographic groups and to housing, we have limited information in other areas. As Sectio n 2.4 points out, we have only one indicator of education problems and no indicators of health problems. We also have only one indicator of long-term economic trends, and it has limitations. The available economic data are limited by the timeliness of the every 5-year economic census and, even more so, by the confidentiality requirements that result in suppressed data for many useful variables—for example, the total number of jobs in a city. It would be comforting to know that the some of other 24 indictors were providing reasonable proxies for the type of problems measured by our lone education and economic indicators. Previous research has turned to factor analysis to solve the problems discussed in the two preceding paragraphs—that is, to aggregate multiple measures into simpler indicators of need and to identify distinct dimensions of need that hopefully span the full range of problems that affect cities. This chapter carries out factor analysis using the indicators 11 The advent of the ACS eliminated a third problem that seriously limited previous efforts to measure community needs: the absence of timely data. 12 We did not use per capita income because we consider it a measure of capacity, and this analysis considers need and capacity separately. We did not use persons with disabilities because the consensus at the November 21 meeting was that employment disabilities, and not disabilities in general, is the more relevant needs indicator. As mentioned in Chapter 2, the persons with employment disabilities indicator was dropped because of suspected data problems. We did not use retail and wholesale jobs because that data series provides an incomplete picture of economic activity, and the data needed to properly balance the picture do not exist.
developed in Chapter 2; it explains factor analysis and explains how we use the results of factor analysis to create a single-valued community needs index and to track conditions in cities over time. Chapter 4 uses the results of our factor analysis to compare the needs of different cities and to see how the needs of individual cities changed from 2000 to 2005. The discussion in this chapter is organized as follows: • Section 3.1 provides a simple explanation of factor analysis. • Section 3.2 uses factor analysis to reduce the needs indicators developed in Chapter 2 to three independent dimensions of need—based, for the most part, on 2005 data. The long discussion in Section 3.2 is divided into six subsections: o Section 3.2.1 lists the expectations we carried into the factor analysis; these expectations guided the choices we made during factor analysis. o Section 3.2.2 reports the results using all 26 needs indicators. The footnotes in this Section give technical details on how we did the factor analysis. o Section 3.2.3 discusses tests that we ran to determine whether the factor analysis would support the uses to which we plan to put it. o Section 3.2.4 considers the possibility of dropping some needs indicators to increase the number of cities for which we can report results. o Section 3.2.5 carries out a second factor analysis using fewer needs indicators. The factors developed in this section are used in Section 3.4 to construct a single-valued community needs index and in Chapter 4 to compare conditions in different cities in 2000 and 2005. Table 3 presents the factor loading for these factors. o Section 3.2.6 explains how we calculated factor scores for each city. • Section 3.3 explains the problems involved in using factor analysis to compare needs across time and describe the methodology we used to make comparisons between needs in 2000 and 2005. • Section 3.4 explains the problems involved in combining the separate dimensions of need that are the product of factor analysis into a single-valued index of communi ty needs. We consider six alternative ways to combine the dimensions and compare results from the alternative indices. • Section 3.5 provides a brief summary of the chapter. Page 26
Page 27 3.1. A Brief Introduction to Factor Analysis Factor analysis is a statistical technique that examines multifaceted data to find simpler, underlying patterns. Researchers apply factor analysis to data sets that have three characteristics: (1) there are a sizable number of units being observed (472 cities in our case); (2) there are a variety of pieces of information (variables) on each unit (26 needs indicators in our case); and (3) there are reasons to believe that there are certain natural groupings among the variables that reflect common contributory sources. 13 The third characteristic is the defining characteristic of factor analysis. The first two characteristics provide the data structure needed for the analysis, while the third characteristic motivates the analysis. The goal of factor analysis is to uncover patterns in the data that can be characterized in a useful fashion. Factor analysis achieves “data reduction”—that is, it replaces a large number of variables with a smaller number that approximates the range of joint variation found in the data set with the larger number of variables. There are other statistical techniques that result in data reduction, most notably, principal components analysis. We could have applied principal components analysis to our database of 26 indicators for 472 cities. The results would have looked similar to those from factor analysis. The output from both factor analysis and principal components analysis would be three or four new variables that represent the range of variation found in the data set with 26 indicators. But the techniques used to obtain the smaller set of variables are different; the statistical properties of the new variables are different; and the interpretation of the new variables is different. 14 The fundamental difference between the two techniques is the difference in interpretation. Factor analysis posits the existence of unobservable “causes” that produce the correlations among the original variables; principal components analysis does not look for underlying causes. Because of this different orientation, the two techniques use different statistical algorithms. The principal components algorithm attempts to explain as much of the variance in the original data as possible, while the factor analysis algorithm attempts to explain the correlation among the variables in the original data. The output of factor analysis is in the format of the table that follows this paragraph. The factor loadings are numbers between 1 and -1 that relate each unobserved factor to the observed variables. For reading the rows, the larger the factor loading, the more important a factor is in determining the value of that variable. For reading the columns, the larger the factor loading, the stronger the association is between that factor and those variables. Understanding what variables are associated with a factor helps the analyst 13 In factor analysis, the assumption is that every variable V i is a result of factors that act on two or more of the variables and other unique circumstances that act on that variable only. Statistically, this assumption translates into the following equation: V i = Σα j F j + U i , where U i are influences unique to that variable and F j are the common factors. Factor analysis assumes that there are common underlying forces that are providing joint causation, but it cannot prove their existence or identify what they are. 14 See Jae-On Kim and Charles W. Mueller, FACTOR ANALYSIS Statistical Methods and Practical Issue, Series: Quantitative Applications in the Social Sciences, Number 14, Sage Publica tions, 1978, pp. 14-23, for a discussion of the similarities and differences between factor analysis and principal components analysis.
understand the nature of the unobserved factor. In the work described in this chapter, we will pay particular attention to the factor loadings. Factor 1 Factor 2 Factor 3 Factor 4 Variable 1 Factor loading Factor loading Factor loading Factor loading Variable 2 Factor loading Factor loading Factor loading Factor loading Variable 3 Factor loading Factor loading Factor loading Factor loading Variable 4 Factor loading Factor loading Factor loading Factor loading Variable 5 Factor loading Factor loading Factor loading Factor loading Variable 6 Factor loading Factor loading Factor loading Factor loading Variable 7 Factor loading Factor loading Factor loading Factor loading Variable 8 Factor loading Factor loading Factor loading Factor loading Unlike most statistical techniques, factor analysis does not produce definitive results in the following three senses: • The mathematical formulas involved in factor analysis will always produce a table like the one above, whether or not there are unobserved causes at work in determining the value of the variables. Although there is a test to determine whether data are suitable for factor analysis, there is no test to prove the existence of a factor. • The mathematic formulas involved in factor analysis will identify many possible “factors” that may or may not be real. Although there are techniques for deciding on how many of the possible factors to use, the choice always involves some judgment. • The factor loadings are not unique. After determining how many factors to select, one can apply certain statistical techniques and produce different tables of the type above from the same data. Each of the tables identifies the same number of factors, and those factors explain the same amount of variation in the data. The tables differ in factor loading, and therefore offer different perspectives on the possible underlying factors. Choosing which of the possible tables to use also requires some judgment. Using factor analysis, we will find three factors that appear to underlie the 26 needs indicators. We interpret these factors as measuring more fundamental “dimensions of need” than the 26 individual needs indicators—that is, we view the “factors” or “dimensions” as logical aggregations of needs that have similar origins or that occur together. Using the needs indicators, factor analysis provides a technique to generate a “score” for each city on each factor. We will build a needs index around these three factor scores. This last step also requires judgment since nothing in the factor analysis process tells us how much importance to give to each factor. Page 28
3.2. Application of Factor Analysis to Needs Indicators 3.2.1 Anticipated Results from Factor Analysis The need to exercise judgment at various stages in performing factor analysis makes it essential to have a clear sense before beginning the analysis of what to expect to find in the way of factors. For this reason, HUD and Econometrica team members discussed their expectations regarding the factors that would be revealed. Based on previous experiences with factor analysis and an understanding of the forces affecting cities, HUD and Econometrica anticipated finding up to four factors and expected those factors to be related to poverty, immigration, economic decline, and city/suburb disparities. Every previous study that has used factor analysis to identify dimensions of community needs has found a “poverty” factor, and “poverty” has always been the first factor—that is, the factor that accounts for most of the variation among the chosen needs indicators. One reason for the primacy of the poverty factor is that the natio nal data sources, particularly the decennial census (and now the ACS), contain an abundance of data that relate to various forms or manifestations of poverty. Among our 26 needs indicators, we have an indicator of overall poverty, poor elderly persons, poor children, and poor school-age children. Our measures also include single-parent families—a group that generally has lower income—and households with incomes higher than the poverty level but lower than 50 percent of metropolitan-area median income. In addition, we measure the percent of the population living in neighborhoods with poverty levels of 40 percent or more, and the percent living in neighborhoods with poverty levels between 20 and 40 percent. In choosing these needs indicators, we were careful to avoid needless duplication. For example, we measured poverty among the elderly, poverty among children, and poverty among school-age children in addition to overall poverty, because we thought each of these groups required somewhat different responses from local government. Also, previous research has found that social problems are greater where poverty is concentrated, particularly in neighborhoods where the poverty rate is 40 percent or more. We did not include female-headed households along with single-parent households. Immigration, economic decline, and city/suburb disparities can be thought of as conditions that generate problems. “Poverty” can be thought of both as a condition that generates problems and as a problem in itself that results from other underlying causes. Poverty could result from a variety of causes: lack of human capital (that is, poor education); market imperfections, such as discrimination or a spatial mismatch between jobs and housing; or general economic decline. In evaluating the findings from factor analysis, we will look at “poverty” in both ways. Page 29
Page 30 3.2.2. Initial Results from Factor Analysis Using 2005 data on the 26 needs indicators for the 472 cities, we performed a standard factor analysis using the factor analysis feature of the statistical program known as SAS™. We were not able to calculate values for all 26 needs indicators for every city. Missing crime data was the primary reason for dropping cities. In other cases, the missing data resulted from suppression of data by the Census Bureau—either to protect the confidentiality of respondents or because the sample sizes were too small to justify reporting the results. Because of missing data, only 292 cities were used in the initial factor analysis. The procedures in SAS™ first examine the data to determine whether there are sufficient relationships among the indicators to justify factor analysis. The 2005 data seem to be well-suited to factor analysis. The SAS™-provided measure of sampling adequacy was strong for all the variables except CHNGEMPLBASE and PT1CRIME. 15 Only the PT1CRIME measure was considered unacceptable and then only marginally so; the implication is that PT1CRIME is not determined by any of the factors that appear to be related to the other indicators. The overall measure of sample adequacy was considered strong. Any factor analysis program, such as SAS™, reports factors in the order in which they help explain the variation in the data. 16 In deciding how many of the reported factors to use, we considered two rules. The first rule selects factors until the group selected account for all of the variation. This approach led to the selection of seven factors. Examination of the factor loading indicated that after the first three or four factors, the remaining factors had no useful interpretation. Therefore, we used a second rule that indicated that only three factors should be considered. 17 The three factors selected appeared to be readily interpretable. The first factor appeared to be associated with poverty, central city/su burb disparities, and long-run decline; the second factor appeared to be associated with immigration and housing affordability; and the third factor appeared to be associated with more immediate limited economic prospects. The factor loading for these factors is reported in Table A.2 in Appendix A. 18 Next, we used a feature available in all factor analysis programs and “rotated” the factors. Rotation is a process that finds alternative factor loadings that are equivalent to the initial 15 This is known as the Kaiser test. The Kaiser measure was strong (0.78 or better) for all the variables except CHNGEMPLBASE (0.60) and PT1CRIME (0.49). The overall Kaiser measure was 0.91, which is considered strong. 16 We used the “principal factor” approach to extract factors. 17 The second rule is known as the Eigenvalue rule. An Eigenvalue is computed for each factor, and only factors with an Eigenvalue greater than one are chosen. In simplest terms, an Eigenvalue greater than 1 means that a factor explains more than the average amount of variation explained by all the factors. 18 These initial loadings are called the “unrotated” loadings because the next step in the factor analysis process is to “rotate” the factors to produce alternative loadings that may be more easily interpreted.
Page 31 loading but may be more easily interpreted. 19 We chose to obtain “orthogonal” factors, that is, factors that are uncorrelated with each other. 20 Having factors that are statistically uncorrelated with each other is useful in developing an index, because it enables us to consider the components of the index as independent contributors to overall need. 21 Table A.3 in Appendix A reports the factor loading for these orthogonal factors. Despite the small differences between the rotated and unrotated factors, the rotated factors provide a somewhat clearer interpretation, such as: • Factor 1 could be interpreted as a “poverty” factor, but it can also be interpreted as a “city/suburb disparities combined with long-term decline” factor. • Factor 2 is a combined “immigration and housing affordability” factor. • Factor 3 is a “weak economy” factor because of the importance of the two education indicators, viewed in a human capital context, and the change in employment base indicator. Statistically, there is no reason to favor rotated factors over unrotated factors or the interpretation attributed to one set of factors over another. Both sets of factors explain the same amount of variation in the data; both satisfy the test for sample adequacy and the same criterion for selecting factors. Fortunately, the interpretations of the unrotated and rotated factors are sufficiently close that use of the rotated factors raises no concerns. 3.2.3. Testing the Robustness of the Factor Analysis Before proceeding further with the analysis, we carried out three important tests. First, we repeated the factor analysis using 2000 data. A prime objective of the research, as specified by HUD, is to develop an index that can be used to track conditions in cities over time. To do this, we need to be confident that the factors that appear to explain the 2005 data are also capable of explaining the 2000 data. To test this, we compared the factor loadings from factor analysis on the 2000 data with the factor loading from factor analysis on 2005 data. Table A.4 in Appendix A reports the results of this analysis. 22 19 The factor loading are equivalent in the sense that they explain the same amount of variation in the data and represent the same hypothetical factors. The mathematical techniques used in factor analysis involve the solution to a matrix algebra problem that has no unique solution; the unrotated and rotated factor loading are alternative solutions. 20 We employed the Varimax rotation, which maximizes the variance of the squared factor loading for each factor. 21 While the orthogonal factors are derived in such a way that the unobserved factors are uncorrelated, the factor scores will not be uncorrelated. 22 Because of missing information or other difficulties, some of the needs indicators have the same values in the 2000 data as in the 2005 data. These are: CHNGEMPLBASE, DENIAL, OVERCROWD, LWINCHHDS, PCTPOPHIGHPOVNGHS, PCTPOPMODPOVNGHS, PCTVACMODPOVCITY, and PR70RENTPOV.
Page 32 The factor patterns in Table A.4 are close, but not identical, to those in Tables A.2 and A.3. In 2000 data, PR1CRIME (arrests related to more serious crimes) loads most heavily on Factor 1, and UNEDUCADULTS (adults without a high school diploma) loads most heavily on rotated Factor 2 instead of Factor 3. When we look at the entire pattern of factor loadings, only the PT1CRIME loadings are noticeably different. In our opinion, the patterns of factor loadings are close enough to justify applying the 2005 factor scoring coefficients to 2000 data. Section 3.3 explains how one should perform a similar test if, for example, one were to use the 2005 factor analysis to compare conditions in 2005 and 2010. The second test examined whether the same factors explain conditions in large cities and small cities. Splitting the sample of cities roughly in half, we ran the factor analysis separately for cities with 200,000 or more residents and cities with less than 200,000 residents. Table A.5 in Appendix A reports the results for the rotated factors. The main difference between factor analysis applied to large and small cities is that the rule used to select factors for the large cities calls for using four factors. The new fourth factor is most strongly associated with declining household population as measured by EXCSINFRA, but EXCSINFRA still loads strongly on Factor 1. There are three other noteworthy changes from the analysis involving all cities, including: • The change in employment base indicator (CHNGEMPLBASE) is more strongly associated with Factor 2 for large cities than Factor 3. • The lack of affordable rental indicator (LACKAFFDRENTALS) is more strongly associated with Factor 1 than Factor 2 for the large cities. • The uneducated adults indicator (UNEDUCADULTS) is more strongly associated with Factor 2 than Factor 3 for large cities. Despite these differences, we believe that it is appropriate to apply factor analysis to the combined database that includes both large and small cities. The fourth factor adds little to the analysis and the other differences are minor. 23 Therefore, with only minor reservations, we proceed with the factor analysis. As noted in Section 2.3.3, we have some concerns about the quality of the crime data. For this reason, we replaced PT1CRIME with VIOLCRIME and carried out the factor analysis again. Table A.5 in Appendix A reports the results. Previously, PT1CRIME had very low loading on all three factors; now VIOLCRIME has a loading of 0.56 on Factor 1. Despite the higher loading, VIOLCRIME is only the 16 th most important of the 26 needs indicators for this factor. 24 A comparison of Table A.6 with Table A.5 shows that 23 The test of sampling adequacy suggests that CHNGEMPLBASE is not an appropriate variable for inclusion in either the large city or small city analysis. 24 In Richardson’s 2000 analysis, the comparable crime variable had higher factor loadings but was not among the most important variables in defining either the unrotated or rotated factors. See Todd Richardson (2007), “Analyzing a Community Development Needs Index,” in Cityscape.
Page 33 replacing PT1CRIME with VIOLCRIME had virtually no effect on th e interpretation of the three factors. 3.2.4. Culling the Needs Indicators After the initial results, we explored alternative ways to structure the factor analysis. In particular, we considered dropping variables that had a large number of missing values in order to increase the number of cities in the analysis. The crime indicators (PT1CRIME and PT2CRIME) were missing for 107 cities. 25 The next four indicators in terms of missing values were: (1) change in the low-income concentration (CHGLOWINCCON), with 42 missing values; (2) the relative concentration of minorities in the central city (MINCON), with 36 missing values; (3) the proportion of immigrants who entered the United States in the last 15 years (RCNTIMMIG), with 25 missing values; and (4) change in the employment base over a recent 5-year period (CHNGEMPLBASE), with 17 missing values. We decided to eliminate the crime indicators because they had low loadings on all the factors; the test of sampling adequacy indicated that, at least, PT1CRIME was not a good candidate for factor analysis; and the analysis in Section 2.3.3 suggests that there may be some problems with PT1CRIME. We decided to keep the remaining variables because all four were important to the interpretation of the factors on which they loaded. In addition, MINCON and RCNTIMMG had high factor loadings on their respective factors. Eliminating the crime indicators increased the number of cities in the factor analysis from 292 to 370, with a total population of 83,246,832 in 2005. 3.2.5. Derivation of the Final Factors Used in this Report We reran the factor analysis for those 370 cities using 24 needs indicators. Again, the rule we used in the initial analysis and our judgment led us to select three factors. Table 3 contains the factor loadings for those three factors based on an orthogonal rotation. To make it easier to interpret each factor, we ranked the indicators by their loadings on each factor. The final factors (below) are easy to interpret: • Factor 1: Three of the four indicators that identify various types of poverty loaded heavily on Factor 1. These are the overall proportion of poor persons (POORPERS), the proportion of children living in households with poverty incomes (POORCHILD), and the proportion of school-age children living in households with poverty incomes (SCHPOPPOOR). The proportion of persons over 74 living in poverty (POOROVER74) had a modest loading of Factor 1, but this indicator loaded more heavily on this factor than either of the other factors. The indicators that related neighborhood poverty also load heavily on Factor 1. 25 Early in the project, HUD had suggested experimenting with excluding some indicators to determine whether their exclusion made any difference to the analysis.
Page 34 The three measures of city/suburb disparity load heavily on Factor 1; these are POVCON, MEDINCCBS2CITY, and MINCON. Finally, although the two indicators of long-term trends, EXCSINFRA and CHGLOWINCCON, have modest loadings on Factor 1, it is the factor on which they load most heavily. • Factor 2: The proportion of households that are linguistically isolated (LINGISOL) and the proportion of the population who are recent immigrants to the United States (RCNTIMMIG) load heavily on Factor 2. Overcrowded housing (OVERCROWD2000) and the lack of affordable rental housing also load heavily on Factor 2. The proportion of adults without a high school diploma (UNEDUCADULTS) loads heavily on Factor 2. • Factor 3: Only the proportion of adults between 25 and 65 years of age without a college degree (UNDEREDWORKAGE) and the proportion of adults without a high school diploma (UNEDUCADULTS) load heavily on Factor 3, although UNEDUCADULTS loads slightly more heavily on Factor 2. The change in the employment base over a recent 5-year period (CHNGEMPLBASE) has its highes t loading on this factor. The proportion of mortgage applications that are denied (DENIAL) also has a modest loading on this factor.
Table 3. Rotated Factor Loadings for Final Factor Analysis, Each Factor Sorted by Loadings Factor 1: Poverty and Structural Problems Factor 2: Immigration and Housing Affordability Factor 3: Limited Economic Prospects POORPERS 0.92728 LINGISOL 0.91800 UNDEREDWORKAGE 0.78216 POORCHILD 0.91442 RCNTIMMIG 0.85752 UNEDUCADULTS 0.54446 POVCON 0.89969 OVERCROWD2000 0.82505 DENIAL 0.43440 SCHPOPPOOR 0.89838 LACKAFFDRENTALS 0.64700 UNEMPCEN 0.28849 PR70RENTPOV 0.87989 UNEDUCADULTS 0.59240 OVERCROWD2000 0.25630 SGLPRNTFAM 0.85591 LWINCHHDS 0.32419 PCTPOPMODPOVNGHS 0.24034 MEDINCCBS2CITY 0.85455 MEDINCCBS2CITY 0.29163 CHNGEMPLBASE 0.23059 LWINCHHDS 0.85063 PCTPOPMODPOVNGHS 0.27201 CHGLOWINCCON 0.20700 MINCON 0.78498 UNDEREDWORKAGE 0.20307 POORPERS 0.18606 PCTVACMODPOVCITY 0.77777 POOROVER74 0.20099 SGLPRNTFAM 0.17529 PCTPOPMODPOVNGHS 0.77370 CHGLOWINCCON 0.18643 LACKAFFDRENTALS 0.15992 DENIAL 0.73175 PR70RENTPOV 0.14105 PCTVACMODPOVCITY 0.15615 PCTPOPHIGHPOVNGHS 0.68063 POORPERS 0.12609 POORCHILD 0.13772 UNEMPCEN 0.63363 POVCON 0.10734 SCHPOPPOOR 0.11026 EXCSINFRA 0.58278 SCHPOPPOOR 0.09173 LINGISOL 0.09723 UNEDUCADULTS 0.47999 POORCHILD 0.06533 MEDINCCBS2CITY 0.09539 POOROVER74 0.45757 UNEMPCEN 0.06183 PCTPOPHIGHPOVNGHS 0.05213 LACKAFFDRENTALS 0.44787 CHNGEMPLBASE 0.02851 POOROVER74 0.03401 CHGLOWINCCON 0.36726 PCTPOPHIGHPOVNGHS 0.02778 EXCSINFRA 0.01141 UNDEREDWORKAGE 0.35296 SGLPRNTFAM 0.01532 MINCON -0.00458 LINGISOL 0.04718 MINCON 0.00792 LWINCHHDS -0.00897 OVERCROWD2000 -0.01856 EXCSINFRA -0.10178 PR70RENTPOV -0.14311 RCNTIMMIG -0.11242 DENIAL -0.15728 RCNTIMMIG -0.15913 CHNGEMPLBASE -0.16812 PCTVACMODPOVCITY -0.28525 POVCON -0.16557 Page 35
Page 36 Based on these loadings, we ascribe these interpretations to the three factors: • Factor 1 is the poverty-structural problems factor. • Factor 2 is the immigration-housing factor. • Factor 3 is the limited economic prospects factor. The low education of the work force, combined with recent declines in jobs relative to the labor force, result in this label. We think the modest loading of DENIAL and the unemployment rate (UNEMPCEN) are consistent with this interpretation. Factor 3 is the least well-defined factor. Lack of clear definition probably results from the paucity of good information on economic trends in cities that was discussed in the conclusion to Chapter 2. Chapter 4 compares cities based on these three factors and on an index derived from these factors in Section 3.5. 3.2.6. Calculating Factor Scores Having identified three common dimensions of need among cities with populations of 65,000 or more, the next step is to calculate a score for each city on each factor so that we can compare the need level in different cities on each dimension. A factor-loading table, such as Table 3, provides information on the relationship between the unobserved factors and the observed needs indicators. This information is useful in characterizing the unobserved factors, but it cannot be used to estimate the factors. In general, factors are not linear combinations of the variables used to identify them. 26 Techniques have been developed to use the observed variables to create linear approximations of the unobserved factors. These techniques first transform the observed variables into standardized form and then use one of several methods to create a set of “standardized scoring coefficients.” 27 The standardized variables are multiplied by the standardized scoring coefficient to provide a linear approximation of each factor and, using this approximation, to create a score for each city on each factor. To derive the standardized scoring coefficients, we used a technique that employs a regression method to minimize the squared deviation between the “estimated” factors and the unobserved factors. Table A.7 in Appendix A presents the standardized scoring coefficients. Because the scores are linear combinations of standardized indicators, the expected value of the score for each scored factor is zero. Because we have defined each indicator such that higher values indicate worse conditions on that indicator, score values greater than zero indicate higher-than-average problems. The unobserved factors are uncorrelated 26 This is another example of how factor analysis differs from principal components analysis. By definition, principal components are linear combinations of the variables that they represent. 27 A standardized value is calculated by the formula: (value – mean value)/standard deviation.
Page 37 with each other, but because the standardized scoring coefficients only create linear approximations of the unobserved factors, the set of factor scores have non-zero correlations. Table 4 shows that the three sets of factor scores are almost uncorrelated— an indication that the scores measure distinctly different conditions. Table 4. Correlations among Factor Scores 28 Correlations Factor 1 Factor 2 Factor 3 Factor 1 1.000 0.012 0.003 Factor 2 1.000 0.016 Factor 3 1.000 3.3. Comparing Needs at Different Times Typically analysts use factor analysis to compare conditions in different cities at a given point in time—for example, to compare the community needs of Denver and Wichita in 2005. But, often analysts want to know whether conditions in a given city have improved or worsened between two points in time. For example, does Denver have more community needs in 2005 than it had in 2000? Factor analysis can also be used for this purpose, but analysts need to take some conceptual issues into account. A factor score is calculated as the weighted sum of the number of standard deviations above (+) or below (-) the mean for each need indicator. The weights are the factor scoring coefficients calculated as part of the factor analysis. With each new wave of ACS data, there will be new means and standard deviations for the needs indicators and, if the factor analysis is repeated, new factor scoring coefficients. Potentially a new factor analysis could even reveal new factors or major changes in factor definitions. With so many possible changes in the inputs used to compare conditions between the two time periods, it is important to define a process that yields a result that has a clear interpretation. The approach we proposed and used has the following steps: 1. Choose a base year. We used 2005 as the base year because the project focused on using the 2005 ACS data. The comparison year is 2000. 2. Derive factors in the base year and save the standardized scoring coefficients to use as weights in both the base year and the comparison year. Table A.7 contains the standardized scoring coefficients. 3. Do a new factor analysis with each new comparison year. Use this to determine whether conditions have changed so much as to make the use of the base-year factor analysis no longer legitimate. Section 3.2.4 reports the comparison we made between the 2000 and 2005 factor analyses. If the base-year factors appear 28 Because we have adhered strictly to the protocols involved in factor analysis, we treat the factor scores as cardinal measurements and therefore use Pearson correlation coefficients.
Page 38 to be the same as the factors in the comparison year, proceed to the following steps. 4. Translate the needs indicators into standardized form in both the base year and the comparison year using the means and standard deviations calculated in the base year. This step is crucial because the standardized scoring coefficients derived from the base year are desi gned to produce factor scores using standardized needs indicators, defined by the means and standard deviations of the base year. In this way, conditions on each need indicator are measured by the distance from the mean of that indicator in the base year using the base-year standard deviation as the unit of measure. Therefore, conditions on each need indicator are measured consistently in both years. 5. Compute the weighted sum of the standardized needs indicators in both the base year and the comparison year. Subtract the base-year score from the comparison- year score. A positive difference indicates that community needs on that factor increased between the base year and the comparison year; a negative difference indicates that community needs on that factor decreased between the base year and the comparison year. 6. For each factor, compute the mean factor score for all cities in the base year and in the comparison year. Subtract the mean factor score in the base year from the mean factor score in the comparison year. This difference indicates whether community needs as measured by that factor have improved or worsened on average. This comparison is not weighted by the size of the cities. There are two important points that need to be made about Steps 5 and 6: • While Steps 5 and 6 are described in terms of a single factor, the same procedure could be applied to a single-value needs index that is calculated as a linear combination of the factors. Instead of using the standardized scoring coefficients for a single factor as the weights in Step 5, one would use a linear combination of the standardized scoring coefficients as weights. The same linear combination would be applied to the standardized scoring coefficients as the one applied to the factor scores in computing the single-valued index. • In both Steps 5 and 6, we use the differences between the scores rather the ratio of the scores. Because the scores can be both positive and negative, the ratio of the scores will not produce a consistent ranking. 29 Chapter 4 presents the results of Steps 1 through 6. Steps 1 through 6 should produce reasonable results for comparisons between points in time that are close together. Over longer time periods, it is possible that factor analysis 29 Section 5.4.2 discusses this issue in more detail with respect to combining a measure of fiscal capacity with a single-valued needs index.
Page 39 will not produce a similar set of factors using the base-year and comparison-year data on the needs indicators. Using the experience with price indices as a guide, we suggest the following approach to handling this problem: • If the comparison-year factor structure is no longer consistent with the base-year structure, then use the comparison-year factor structure as a new baseline. This is analogous to using a new basket of goods and services for a price index. One could continue to report the old index along with the new index. This approach used the method applied to price indices until the introduction of chain-linked indicators. We considered whether it would be possible to create the equivalent of a chain-linked indicator to handle this situation. Chain-linked indices have two key characteristics: they allow the weights to evolve over time, and applying the technique year-by-year over a period of years produces the same result as applying it to the beginning and end years of the period. To achieve these two characteristics, chain-link indicators use geometric means instead of arithmetic means. Unfortunately, factor scoring is based on arithmetic averaging instead of geometric averaging. Therefore, we cannot construct a chain-link index to compare needs over time. 3.4. Creating a Single-Valued Index of Community Needs 3.4.1. Alternative Indices The explicit goal of this project is to produce an index of community needs—a formula that will assign one number to each city to indicate its relative need. The needs index will be a function of the three factor scores; but because the factor scores are linear combinations of the needs indicators, it will also be a function of the 24 needs indicators. Section 3.1 noted that factor analysis provides no information that can be used to choose weights to combine the factors into a single-valued index. In this section, we construct six alternative indices; some are based on external rationales while others are created to test how sensitive the index results are to the choice of weights. Table 5 defines the alternative indices and the reasons we constructed them.
Page 40 Table 5. Alternative Single-Valued Community Needs Indices Index Index Name Factor 1 (Poverty and Structural Problems) Weight Factor 2 (Immigration and Housing Affordability) Weight Factor 3 (Limited Economic Prospects) Weight Rationale for Index 1 Equal weight 1/3 1/3 1/3 This index treats all three factors the same. It is the standard to which we compared the other indices. 2 Triple weight to poverty and structural problems 0.60 0.20 0.20 Legislation provides virtually no guidance in choosing weights. However, the CDBG statute does give precedence to “the development of viable urban communities, by providing decent housing and suitable living environment and expanding economic opportunities, principally for persons of low and moderate income [emphasis added].” Therefore, we provide a triple weight to the factor that relates to poverty. We decided on triple weights so that Indices 2 and 3 would parallel Index 4. 3 Triple weight to immigration and housing affordability factor 0.20 0.60 0.20 Indices 2 and 4 provide extra weights to Factors 1 and 3 respectively. We added this index to see what happens when we add extra weight to Factor 2 alone. 4 Triple weight to limited economic prospects factor or hedonic weights 0.20 0.20 0.60 The hedonic analysis in Section B.3 in Appendix B indicates that Factor 3 should receive three times the weight of Factor 1. It provides no information on how to weight Factor 2; we gave Factor 2 the same weight as Factor 1. 5 Richardson weights 0.80 0.15 0.05 Richardson chose these weights for his unrotated factors. Our rotated factors do not match well with either Richardson’s unrotated or rotated factors but they are more similar to his unrotated factors. 6 Partial hedonic weights 0.60 0.28 0.12 Using the hedonic analysis in Appendix B, we chose these weights to obtain the closest match between the weighted sum of the scoring coefficients for eight needs indicators to beta coefficients for those variables in the hedonic-type equation. See Section B.4 for an explanation of how we derived these weights.
Page 41 The equal weight index (Number 1) is the standard to which we compare the other indices; it treats all the factors equally. We have some rationale for Indices 2, 4, 5, and 6. Index 2 puts a triple weight on the poverty and structural problems factor because the CDBG legislation emphasizes assistance to low- and moderate-income person indices. Index 5 uses the same weights that Richardson employed when he constructed an index based on 2000 census data. Richardson used four unrotated factors, which he identified with poverty, immigration, high poverty concentration, and income growth. He gave these factors weights of 0.80, 0.15, 0.05, and 0.00 respectively. Our Factors 1 and 2 correspond roughly to Richardson’s first two factors; there appears to be little overlap between our Factor 3 and Richardson’s third factor. We based Index 5 on the Richardson weights. The hedonic analysis reported in Appendix B provides some guidance on weighting the factors. A regression involving the factor scores sug gests that Factor 3 should be receive three times the weight of Factor 1 but provides no guidance on what weight should be given to Factor 2. We incorporated this information into the weights for Index 4. A separate regression involving the 24 needs indicators provided useful information on how eight of the indicators affect property values. We incorporated this information into the weights for Index 6. Index 3 was added to test the sensitivity of the results to added weight to Factor 2. 3.4.2. Comparisons of Scores on Alternative Indices Table A.8 in Appendix A contains the scores on all six indices for the 370 cities for which we computed factor scores. Table 6 presents some key statistics on the indices. Indices 2, 5, and 6 have larger ranges than the other three indices because these indices give a heavy weight to the poverty and structural problems factor and because the scores for that factor have a larger range than the scores for the other two factors. 30 The equal weight index has the smallest standard deviation while the Richardson index has the largest. Table 6. Basic Statistics on the Alternative Indices Index 1: Equal Weight Index 2: Triple Weight to Poverty and Structural Problems Index 3: Triple Weight to Immigration and Housing Affordability Factor Index 4: Hedonic Weights Index 5: Richardson Weights Index 6: Partial Hedonic Weights Mean -0.01 -0.01 -0.01 -0.02 -0.01 -0.01 Variance 0.28 0.38 0.36 0.36 0.57 0.39 Std Dev 0.53 0.61 0.60 0.60 0.76 0.62 Max 2.03 3.02 2.45 1.62 3.82 3.11 Min -1.21 -1.26 -0.96 -1.91 -1.34 -1.21 Range 3.25 4.29 3.41 3.53 5.17 4.32 30 The ranges for the three factor scores are: 6.0, 5.0, and 5.2.
Page 42 Table 7 presents the correlations between population and the six indices and the correlations among the indices. This table shows that all the indices except the hedonic index (Index 4) have some correlation with population. Larger cities appear to have more need; this effect is small in all cases. The needs indicators were defined in per capita or percentage terms so that they would be independent of city size, and Table A.1 in Appendix A shows that correlations between population and the needs indicators were very low, ranging from -0.06 to 0.17. So the small positive correlations between five of the index scores and population suggest that larger cities have somewhat greater community needs. 31 Table 7. Correlations Among the Alternative Indices Population Index 1: Equal Weight Index 2: Triple Weight to Poverty and Structural Problems Index 3: Triple Weight to Immigration and Housing Affordability Factor Index 4: Hedonic Weights Index 5: Richardson Weights Index 6: Partial Hedonic Weights Population 1.000 0.122 0.152 0.140 0.026 0.163 0.172 Index 1 0.122 1.000 0.874 0.875 0.872 0.719 0.863 Index 2 0.152 0.874 1.000 0.644 0.641 0.964 0.987 Index 3 0.140 0.875 0.644 1.000 0.649 0.492 0.704 Index 4 0.026 0.872 0.641 0.649 1.000 0.420 0.565 Index 5 0.163 0.719 0.964 0.492 0.420 1.000 0.964 Index 6 0.172 0.863 0.987 0.704 0.565 0.964 1.000 The correlations between the equal weight index (Index 1) and all the other indices are strong. 32 The Richardson index (Index 5) has the lowest correlation with the equal weight index (0.719); the other four indices have correlations of approximately 0.87. This suggests that an equal weight index is a reasonable approximation to a wide range of weighted indices. Weighting does affect the scoring. While the equal weight index correlates well with the other indices, the correlations among the weighted indices vary more. We focus on Indices 2, 3, and 4 because they represent, respectively, emphasizing Factors 1, 2, or 3 heavily. Correlations among these indices are in the range of 0.60 to 0.65. The Richardson index weighs Facto r 1 very highly and gives small weight to the other two factors. It correlates highly with Indices 2 and 6, which also weigh Factor 1 highly but has correlations in the 0.40 to 0.50 range with Indices 3 and 4. Table 8 compares the scores on the equal weight index (Index 1) to the score from Indices 2, 3, and 4, which successively give triple weight to Factors 1, 2, and 3. Scores on Index 1 varied from a high of 2.03 to a low of -1.21, a range of 3.25 points. The 31 The correlations between population and the factor scores were: 0.148 for Factor 1, 0.127 for Factor 2, and -0.065 for Factor 3. 32 The Spearman rank-order correlations among the six indices are very close to the Pearson correlations reported in Table 7 and display the same pattern.
scores for all the cities on Indices 2, 3, and 4 are within 1.00 points of their scores on Index 1; the scores for over 90 percent of the cities are within 0.50 points of their scores on Index 1; and the scores for over 60 percent of the cities are within 0.25 points of their scores on Index 1. Camden had the largest difference in scores between Indices 1 and 2; it scored 0.99 points higher on Index 2. Despite this large difference in scores, Camden was the city with the highest score on both Index 1 and Index 2. Miami had the largest difference in scores between Indices 1 and 3; it scored 0.87 points higher on Index 3. Miami was ranked as the 56 th most needy city on Index 1 and was ranked as the 16 th most needy city on Index 3. Cambridge, MA had the largest difference in scores between Indices 1 and 4; it scored 0.96 points lower on Index 4. Cambridge was ranked as the 357 th most needy city on Index 1 and was ranked as the 368 th most needy city on Index 4. Table 8. Comparison of Scores between Index 1 and Indices 2, 3, and 4 for 370 Cities Absolute Difference between Score on Index 1 and Score on -- Index 2: Triple Weight to Poverty and Structural Problems Index 3: Triple Weight to Immigration and Housing Affordability Factor Index 4: Triple Weigh to Limited Economic Prospects Mean 0.23 0.23 0.23 Std Dev 0.18 0.18 0.18 Max 0.99 0.87 0.96 Number of Cities whose Score on Index 1 is Index 2 Index 3 Index 4 Within 1.00 370 370 370 Within 0.50 339 338 340 Within 0.25 234 223 230 The data in Tables 6, 7, and 8 indicate that, from a statistical perspective, an equal weight index provides scores and rankings similar to those provided by indices that weigh the factor scores unevenly. The numeric and rank-order correlations between Index 1 and Indices 2, 3, and 4 are all above 0.85. Using Indices 2, 3, or 4, rather than Index 1, would affect the scores of 60 percent of the cities by less than 0.25 points. For this reason, we will use the equal weight index as our single-valued index in comparing conditions across cities and between 2000 and 2005 in Chapter 4. Statistical closeness does not mean that the ranking of some cities are not substantially different depending upon the index used. Washington, DC had the biggest change in ranking between Index 1 and Index 2; it is ranked 243 rd on Index 1 and 80 th on Index 2. Sunnyvale, CA had the biggest change in ranking between Index 1 and Index 3; it is ranked 284 th on Index 1 and 98 th on Index 3. Providence had the biggest change in ranking between Index 1 and Index 4; it is ranked 48 th on Index 1 and 256 th on Index 4. If HUD were to use one of these indices to allocate funds to cities, the choice of index would be of great concern to individual cities. But, if HUD is interested primarily in analyzing the variation in needs across cities and over time, then the results from the equal weight index will be similar to those from any index that applies reasonable weights to the factor scores. Page 43
Page 44 3.4.3. Transformation of the Factor Score Functions into F unctions of Needs Indicators Because the factor scores are linear combinations of the needs indicators, the choice of index determines which needs indicators will have the greatest impact on the index score. 33 Interpreting the indices in terms of the needs indicators helps identify the cities that might do best or worst on a particular Index. Table 9 uses the standardized scoring coefficients in Table A.7 to transform Indices 1 through 4 from weighted sums of factor scores into weighted sums of the 24 needs indicators. The entries in Table 9 tell how much increase in the relevant index score would result from a one standard deviation increase in need on a given need indicator. SCHPOPPOOR, LWINCHHDS, MEDINCCBS2CITY, MINCON, and EXCSINFRA are needs indicators that contribute to high scores on Index 2. As expected, RCNTIMMG, LINGSOL, OVERCROWD_2000, and LACKAFFRDRENTALS contribute to high scores on Index 3. DENIAL, CHNGEMPLBASE, UNDEREDWORKAGE, UNEDUCADULTS, and UNEMPCEN contribute to high scores on Index 4. 33 Increases in some needs indicators—for example, POOROVER74—would decrease the index score for that city. The factor loading and the standardized scoring coefficients take into account correlations among the needs indicators, and thus some of the loading and some of the scoring coefficients are negative.
Table 9. Transformation of Factor-Scoring Coefficients into Scoring Coefficients for Needs Indices Need Indicator Index 1: Equal Weight Index 2: Triple Weight to Poverty and Structural Problems Index 3: Triple Weight to Immigration and Housing Affordability Factor Index 4: Triple Weight to Limited Economic Prospects POORPERS 0.1413 0.1596 0.0881 0.1761 POORCHILD 0.025 0.0505 -0.0378 0.0633 SCHPOPPOOR 0.0070 0.0525 0.0392 -0.0707 POOROVER74 -0.0061 -0.0008 -0.0040 -0.0135 LWINCHHDS 0.0078 0.0160 0.0442 -0.0367 SGLPRNTFAM 0.0170 0.0342 0.0016 0.0152 PCTPOPHIGHPOVNGHS -0.0126 -0.0019 -0.0116 -0.0244 PCTPOPMODPOVNGHS 0.0249 0.0185 0.0331 0.0232 PCTVACMODPOVCITY -0.0006 0.0365 -0.0530 0.0148 MEDINCCBS2CITY 0.0970 0.1397 0.0988 0.0525 MINCON 0.0038 0.0193 -0.0077 -0.0003 POVCON -0.1085 -0.0261 -0.0628 -0.2367 EXCSINFRA 0.0082 0.0247 -0.0040 0.0039 CHGLOWINCCON 0.0134 -0.0010 0.0056 0.0357 RCNTIMMIG 0.0178 0.0076 0.1093 -0.0637 LINGISOL 0.086 0.039 0.1966 0.0241 OVERCROWD_2000 0.0659 0.0325 0.1052 0.0601 LACKAFFDRENTALS 0.0290 0.0107 0.0630 0.0132 PR70RENTPOV -0.0288 0.0432 -0.0136 -0.1160 DENIAL 0.0516 0.0530 -0.0116 0.1134 CHNGEMPLBASE 0.0159 0.0066 0.0114 0.0297 UNDEREDWORKAGE 0.1037 0.0395 0.0424 0.2292 UNEDUCADULTS 0.1688 0.0849 0.1533 0.2682 UNEMPCEN 0.0129 0.0156 -0.0038 0.0270 3.5. Summary of Factor Analysis This chapter applied standard factor techniques to a set of 26 needs indicators developed in Chapter 2. The majority of these needs indicators use data from the 2005 American Community Survey. The factor analysis identified three dimensions that represent community needs in 2005 for the 292 cities for which we have data. We tested the factor analysis results in three ways. First, we compared the factor analysis using 2005 data for most needs indicators to factor analysis using 2000 data for most needs indicators. The two analyses identified factors that were nearly identical. This process gives us confidence that we could apply factors developed using 2005 data to 2000 data on needs indicators. Second, we split the sample of cities into those with populations of 200,000 or more and those with populations of less than 200,000. Factor analysis applied separately to the two samples produced results that were very similar. These results gave us some confidence that community needs are similar in larger and Page 45
smaller cities. Finally, we substituted a different measure for violent crimes than the measure used in the initial analysis (PT1CRIME). The resu lts of the factor analysis did not seem to vary significantly when the alternative measure of violent crimes was used. This relieved some concerns we had about the original measure of violent crimes. Next, we examined the needs indicators to see where problems with missing data caused a large number of cities to drop out of the analysis. Based on this examination, we eliminated PT1CRIME and PT2CRIME from the set of indicators and reran the factor analysis. When applied to the smaller set of needs indicators, factor analysis identified the same factors found with the full set of indicators. Eliminating these two variables increased the number of cities included from 292 to 370. This factor analysis is the one that we use for the remainder of the analysis in the report. We interpret the factors to represent the needs associated with: • Poverty and structural problems, • Immigration and lack of affordability housing, and • Limited economic prospects. The first two factors are well-defined; the third factor is weakly defined. We ascribed the weak definition of the limited economic prospects factor to the lack of multifaceted data on economic conditions and trends in cities. In Section 3.3, we discussed technical issues in applying factors developed at one point to data on the same needs indicators at a different point in time. This provided the conceptual background for the comparisons in Chapter 4. Finally, we examined six alternative single-valued needs indicators based on linear combinations of scores from the three factors. Examining the correlations among the indices and other statistics, we concluded that an equal weight index would provide adequate information on the variation in community needs across cities and across time. We use the equal weight index in Chapters 4 and 5. Page 46
4. Community Needs in 2000 and 2005 While this research project has multiple objectives, the two principal goals are to test the feasibility of using ACS data to measure community needs and to test the feasibility of measuring changes in community needs over time. Chapter 2 identified 26 indicators of problems at the city level, most of which either use ACS data or will be capable of using ACS data once the ACS begins to release 5-year moving average data for census tracts. Chapter 3 performed factor analysis using 24 of the 26 indicators and identified three factors that track different dimensions of community needs. Chapter 3 also examined several single-valued indices based on the three factors and explained how to apply factor analysis in different years. This chapter compares conditions in 370 cities in 2000 and 2005 using each of the three factors and also using the equal weight index developed in Chapter 3. The chapter examines changes in each factor between 2000 and 2005 to obtain a fuller picture of how conditions in individual cities are changing. The equal weight index provides a convenient summary of these changes. As noted in Chapter 3, the results from the equal weight index are similar in scope and general details to that from other indices that weigh the factors unequally. While unequal weighting can markedly change the scoring of individual cities, the overall patterns are more stable. The reader should keep in mind the following facts about how the analysis in this chapter was carried out: (1) 2005 data were used to identify the factors and to develop standardized scoring coefficients; (2) the standardized scoring coefficients were applied to standardized data on 24 indicators in 2000 and 2005; and (3) standardization of the indicators was achieved in both 2000 and 2005 by taking the value of the indicator in the relevant year and subtracting the mean value of the indicator in 2005 and dividing the difference by the standard deviation of the indicator in 2005. Table A.9 in Appendix A presents the results of these calculations for 2000 and 2005 for the three factors and for the equal weight index for all 370 cities. Section 4.1 looks at the how conditions changed on average for the 370 cities between 2000 and 2005. Sections 4.2, 4.3, and 4.4 examine changes across individual cities for each of the three factors. Section 4.5 uses the equal weight index to compare changes in overall community needs. Section 4.6 contains a summary of findings. 4.1. Changes in Community Needs for Cities with Populations of 65,000 or More Table 10 computes the average score on each factor in 2000 and 2005 and the average score on the equal weight index in both years. On each factor and on the index, an increase in the scores (a positive change) indicates an increase in community needs while a decrease in scores (a negative change) indicates a decrease in community needs. Page 47
Before looking at the numbers in Table 10, it is important to call attention to two previous results. Table 9 indicated that, of the 24 needs indicators, UNEDUCADULTS has the largest impact on the equal weight index. (Table A.7 in Appendix A indicates that UNEDUCADULTS also has a strong impact on the scoring for Factor 3.) Table 2 noted that the mean of the unstandardized data for UNEDUCADULTS declined by 17 percent between 2000 and 2005. The discussion of Table 2 expressed surprise at the size of this decline, but examination of Census Bureau reports comparing the decennial census with the ACS failed to find any indication of problems with this variable. In addition, the observed decline for the cities studied closely paralleled the decline in the data for the entire United States. Because we found no evidence of problems with this variable, we included it among the needs indicators. Table 10. Average Factor Scores and Average Equal Weight Index Scores in 2000 and 2005 Factor 1 (Poverty and Structural Problems) Factor 2 (Immigration and Housing Affordability) Factor 3 (Limited Economic Prospects) Equal Weight Index Mean - 2005 -0.006 0.004 -0.034 -0.012 Mean - 2000 -0.154 -0.067 0.192 -0.010 Change 0.149 0.070 -0.226 -0.002 Ratio of change to standard deviation in 2005 16.1% 7.8% -25.4% -0.5% Number of cities worse off 283 231 44 168 Number of cities no worse off 87 139 326 202 Table 10 shows that, on average, community needs—as measured by the equal weight index—decreased slightly between 2000 and 2005. This decline was due solely to improvement in the needs represented by Factor 3, the limited economic prospects factor. As discussed above, the improvement in Factor 3 and the equal weight index can be attributed to the substantial reduction in the percentage of adults without a high school diploma (UNEDUCADULTS) between 2000 and 2005. Table 10 also shows that community needs related to poverty and structural problems (Factor 1) and immigration and housing affordability (Factor 2) worsened between 2000 and 2005. Conditions worsened most with respect to poverty and structural problems. The average city experienced a move of 1/6 of a standard deviation up in the score on this factor whereas the average city experienced a move of only 1/12 of a standard deviation up in the score on the immigration and housing affordability factor. Consistent with the relative size of the average changes, the number of cities that were worse off (had higher scores) in 2005 was larger for Factor 1 than Factor 2. Page 48
Page 49 Correlation analysis found little evidence of a relationship between changes in the score on one factor and changes in the scores on either of the other two factors between 2000 and 2005. Changes in the score of Factor 1 have a correlation of 0.18 with changes in the scores of Factor 2. The other two pairings have negative correlations of -0.01 and -0.06. 34 In addition, there was no relationship between population and changes in the scores on any of the factors or on th e score for the equal weight index. 4.2. Comparison of Scores in 2000 and 2005 on Factor 1 The Factor 1 scores rank cities on community needs related to poverty and structural problems. Between 2000 and 2005, 283 of the 370 cities became worse off on this dimension of need. Table 11 shows how the scores on this factor varied by region and by size class of cities. Table 11. Changes in Factor 1 Scores between 2000 and 2005, by Region and Population Region Number of Cities 2000 Factor 1 2005 Factor 1 Difference South 110 -0.11 0.03 0.14 West 150 -0.48 -0.40 0.08 Midwest 69 -0.06 0.18 0.24 Northeast 41 0.77 1.05 0.28 Population 1,000,000+ 9 0.46 0.59 0.13 500,000-999,999 21 0.33 0.53 0.20 300,000-499,999 23 0.30 0.44 0.14 200,000-299,999 37 0.07 0.23 0.16 100,000-199,999 124 -0.22 -0.08 0.15 under 100,000 156 -0.32 -0.18 0.14 All cities 370 -0.15 -0.01 0.15 The Northeast has the highest average scores on the poverty and structural problems factor in both 2000 and 2005 and the largest increase in average scores between the two years. The West region has the lowest average scores on this factor in both years and the smallest increase between the two years. Using 2005 as the standard, only the average scores in the Northeast were above the average for 2005 in 2000; by 2005, all regions except the West had above-average scores. There appears to be a systematic relationship between the scores on the poverty and structural problems factor and city size. The average score declined by size class in both 2000 and 2005. The change in scores is approximately the same for all the size classes— except for cities with populations between 500,000 and a million, which have a slight higher increase in average scores. 34 As expected, scores of a factor in 2000 are highly correlated (approximately 0.97) with scores on the same factor in 2005.
Table 12 lists the 40 cities that had the largest increase in need on this factor between 2000 and 2005. The list contains cities that already had serious problems related to this factor in 2000—such as Camden, Detroit, Cleveland, Rochester, Reading, and Syracuse—and cities that were relatively well off on Factor 1 in 2000—such as Redwood, CA; West Covina, CA; Hillsboro, OR; Garland, TX; Upland, CA; and Cedar Rapids. The latter cities moved up sharply in the ranking on Factor 1. The Northeast region is heavily represented among those cities with the largest increases in Factor 1 scores. Of the 41 Northeast cities, 14 are on the list of 40 cities with the largest increase in Factor 1 needs between 2000 and 2005. Page 50
Table 12. Forty Cities with the Largest Increases in Factor 1 Scores, 2000-2005 City State 2005 Population 2000 Factor 1 2005 Factor 1 Difference 2000 Rank 2005 Rank 1 Lawrence city Massachusetts 82,191 1.88 2.89 1.01 12 5 2 Hillsboro city Oregon 82,732 -0.74 0.19 0.94 276 120 3 Camden city New Jersey 73,305 3.62 4.51 0.89 1 1 4 Reading city Pennsylvania 81,302 2.04 2.89 0.85 8 6 5 Passaic city New Jersey 68,422 0.93 1.76 0.83 32 20 6 Scranton city Pennsylvania 67,314 0.28 1.07 0.80 91 42 7 Redwood City city California 81,195 -1.00 -0.25 0.74 312 204 8 Gainesville city Florida 100,879 0.24 0.93 0.69 97 52 9 West Covina city California 116,371 -0.97 -0.33 0.64 309 223 10 Baton Rouge city Louisiana 205,442 0.79 1.44 0.64 45 27 11 Dayton city Ohio 132,679 1.92 2.56 0.64 11 9 12 Springfield city Massachusetts 146,948 1.32 1.96 0.64 22 18 13 Birmingham city Alabama 222,154 1.82 2.46 0.64 14 11 14 Hammond city Indiana 72,507 0.35 0.97 0.62 84 50 15 Cleveland city Ohio 414,534 2.52 3.14 0.62 4 2 16 Somerville city Massachusetts 74,869 -0.12 0.48 0.60 152 85 17 Cedar Rapids city Iowa 119,670 -0.62 -0.04 0.58 255 171 18 Nampa city Idaho 67,112 -0.27 0.31 0.58 180 111 19 Allentown city Pennsylvania 105,231 0.89 1.4 6 0.57 38 26 20 Albany city New York 78,402 1.53 2.10 0.57 18 13 21 Detroit city Michigan 836,056 2.59 3.13 0.54 3 3 22 Gresham city Oregon 95,334 -0.43 0.11 0.54 222 138 23 Syracuse city New York 132,495 2.02 2.55 0.53 10 10 24 Pueblo city Colorado 101,302 0.35 0.86 0.51 82 57 25 Avondale city Arizona 61,666 -0.42 0.08 0.50 219 144 26 Bryan city Texas 56,277 0.09 0.59 0.50 119 75 27 Lansing city Michigan 119,675 0.57 1.06 0.49 67 43 28 Rochester city New York 189,312 2.42 2.90 0.48 6 4 29 Vancouver city Washington 155,488 -0.31 0.17 0.47 190 126 30 Garland city Texas 235,750 -0.73 -0.27 0.46 273 210 31 Lowell city Massachusetts 96,876 0.65 1.10 0.44 56 38 32 South Bend city Indiana 97,070 0.65 1.08 0.44 57 41 33 Milwaukee city Wisconsin 556,948 1.29 1.72 0.43 24 22 34 Lynn city Massachusetts 83,419 0.66 1.09 0.43 54 39 35 Pawtucket city Rhode Island 72,896 0.50 0.92 0.42 72 53 36 Toledo city Ohio 285,937 0.72 1.14 0.42 47 34 37 High Point city N Carolina 101,852 -0.15 0.26 0.42 163 112 38 Upland city California 74,420 -0.70 -0.29 0.41 265 214 39 Rockford city Illinois 139,173 0.26 0.67 0.41 92 71 40 Tyler city Texas 87,687 0.09 0.50 0.40 117 82 Page 51
Table 13 lists the 40 cities that experienced the greatest improvement on the poverty and structural problems factor between 2000 and 2005. Table 13. Forty Cities with the Largest Decreases in Factor 1 Scores, 2000-2005 City State 2005 Population 2000 Factor 1 2005 Factor 1 Difference 2000 Rank 2005 Rank 1 Miami city Florida 361,701 0.84 0.10 -0.73 42 140 2 Glendale city California 194,620 -0.38 -0.88 -0.50 209 306 3 Turlock city California 74,883 -0.38 -0.84 -0.46 210 298 4 Rialto city California 93,284 -0.26 -0.67 -0.41 178 277 5 Pomona city California 161,257 0.16 -0.24 -0.40 111 203 6 Oceanside city California 162,259 -0.58 -0.98 -0.40 247 320 7 Wilmington city N Carolina 91,207 0.30 -0.05 -0.35 88 173 8 Alexandria city Virginia 133,479 -0.51 -0.85 -0.34 232 303 9 Redding city California 89,362 -0.19 -0.51 -0.32 170 255 10 Westminster city California 97,946 -0.77 -1.08 -0.31 281 338 11 Santa Monica city California 82,777 -0.74 -1.00 -0.27 274 324 12 Hemet city California 77,076 0.01 -0.26 -0.27 131 206 13 Pompano Beach Florida 94,892 -0.02 -0.28 -0.26 135 213 14 Bethlehem city Pennsylvania 68,144 0.20 -0.04 -0.24 100 167 15 Richmond city Virginia 180,757 1.22 0.98 -0.24 26 48 16 Long Beach city California 463,956 0.62 0.38 -0.24 62 104 17 San Marcos city California 77,445 -0.71 -0.94 -0.23 270 314 18 Escondido city California 133,017 -0.46 -0.69 -0.22 226 280 19 Columbia city S Carolina 88,450 0.93 0.72 -0.21 34 67 20 Deltona city Florida 85,979 -0.86 -1.04 -0.18 298 330 21 Newark city New Jersey 254,217 2.21 2.02 -0.18 7 16 22 Peoria city Illinois 102,136 0.86 0.68 -0.18 41 70 23 Fullerton city California 142,064 -0.84 -1.00 -0.16 292 323 24 Quincy city Massachusetts 84,080 -0.69 -0.85 -0.16 262 302 25 Stockton city California 278,515 0.52 0.38 -0.14 69 102 26 Orange city California 137,994 -1.00 -1.14 -0.14 315 350 27 Inglewood city California 120,204 0.63 0.48 -0.14 60 84 28 Suffolk city Virginia 77,922 -0.26 -0.39 -0.13 179 235 29 Chino city California 69,732 -1.16 -1.29 -0.13 338 364 30 McKinney city Texas 92,337 -0.95 -1.07 -0.13 307 334 31 Merced city California 65,391 0.68 0.55 -0.12 52 77 32 El Cajon city California 92,507 0.06 -0.06 -0.12 123 177 33 North Las Vegas Nevada 165,061 -0.35 -0.47 -0.12 201 247 34 Melbourne city Florida 76,373 -0.45 -0.58 -0.12 225 267 35 Riverside city California 294,059 -0.37 -0.49 -0.12 204 253 36 Buena Park city California 76,062 -0.82 -0.94 -0.12 289 315 37 Simi Valley city California 116,722 -1.18 -1.30 -0.12 340 365 38 Berkeley city California 90,432 0.19 0.08 -0.11 104 145 39 Clearwater city Florida 108,382 -0.35 -0.46 -0.10 202 245 40 Modesto city California 202,971 -0.29 -0.39 -0.10 188 234 Page 52
Twenty-four of the 40 cities are in California; 11 others are in the South region. Only Long Beach and Miami have populations over 300,000. Interesting cases include Newark, which moved from 7 th highest score in 2000 to the 16 th highest score in 2005, and Richmond, which moved from the 26 th highest score to the 48 th highest. 4.3 Comparison of Scores in 2000 and 2005 on Factor 2 The Factor 2 scores rank cities on community needs related to immigration and the housing affordability. Between 2000 and 2005, 231 of the 370 cities became worse off on this dimension of need. Table 14 shows how the scores on this factor varied by region and by size class of cities. Table 14. Changes in Factor 2 Scores between 2000 and 2005, by Region and Population Region Number of Cities 2000 Factor 2 2005 Factor 2 Difference South 110 -0.40 -0.33 0.07 West 150 0.34 0.41 0.06 Midwest 69 -0.63 -0.59 0.04 Northeast 41 0.28 0.41 0.14 Population 1,000,000+ 9 0.70 0.76 0.06 500,000-999,999 21 -0.14 -0.10 0.04 300,000-499,999 23 0.21 0.23 0.02 200,000-299,999 37 -0.27 -0.16 0.11 100,000-199,999 124 -0.12 -0.05 0.07 under 100,000 156 -0.05 0.02 0.08 All cities 370 -0.07 0.00 0.07 Cities in the Northeast experienced the greatest worsening of conditions on this factor, an increase of 0.14 standard deviations, which was twice the national average. With the exception of cities with over a million residents, there appears to be little relationship between population size and the prevalence of problems related to immigration and housing affordability. The largest cities had an average score of 0.70 or more in both 2000 and 2005. Table 15 lists the 40 cities that had the greatest increase in the score on Factor 2. California, Texas, and Florida account for 26 of the 40 cities. None of the cities on this list had populations above 300,000. Only Salinas, CA and Lawrence, MA had been ranked in the top 20 in 2000, and only five of these cities had been ranked in the top 50 in 2000. Mesquite, TX and Cape Coral, FL had the biggest increase in rank on this factor. Mesquite moved from 248 th to 153 rd while Cape Coral moved from 244 th to 179 th . Page 53
Table 15. Forty Cities with the Largest Increases in Factor 2 Scores, 2000-2005 City State 2005 Population 2000 Factor 2 2005 Factor 2 Difference 2000 Rank 2005 Rank 1 Deerfield Beach Florida 71,599 0.27 1.11 0.84 102 47 2 Mesquite city Texas 126,895 -0.58 0.02 0.61 248 153 3 Camden city New Jersey 73,305 0.77 1.36 0.59 61 29 4 Union City California 65,239 1.16 1.74 0.58 32 20 5 San Bernardino California 204,552 1.00 1.56 0.55 41 24 6 Trenton city New Jersey 77,471 0.15 0.69 0.54 115 72 7 Redwood City California 81,195 0.82 1.34 0.52 56 30 8 Aurora city Illinois 170,490 0.46 0.96 0.49 83 58 9 Reading city Pennsylvania 81,302 0.12 0.62 0.49 119 83 10 Hemet city California 77,076 0.11 0.60 0.49 122 85 11 Lowell city Massachusetts 96,876 0.64 1.10 0.46 67 49 12 Rialto city California 93,284 0.65 1.11 0.46 66 48 13 Garland city Texas 235,750 0.26 0.72 0.46 104 68 14 Palmdale city California 145,800 0.22 0.66 0.44 108 77 15 Gresham city Oregon 95,334 -0.09 0.35 0.44 151 105 16 Salinas city California 156,950 2.14 2.56 0.43 10 4 17 Cape Coral city Florida 134,388 -0.58 -0.16 0.42 244 179 18 Fremont city California 210,387 0.86 1.26 0.40 54 34 19 Hollywood city Florida 138,412 0.29 0.68 0.39 99 73 20 Turlock city California 74,883 0.32 0.71 0.39 95 69 21 Newark city New Jersey 254,217 1.27 1.65 0.38 29 22 22 Tracy city California 82,218 -0.20 0.18 0.38 172 126 23 Pompano Beach Florida 94,892 0.18 0.54 0.36 110 88 24 Lawrence city Massachusetts 82,191 2.05 2.41 0.35 13 7 25 Worcester city Massachusetts 154,398 0.12 0.47 0.35 120 95 26 Antioch city California 103,339 -0.28 0.07 0.35 183 145 27 Pasadena city Texas 150,180 0.70 1.05 0.34 63 52 28 Fairfield city California 102,642 -0.13 0.21 0.34 154 122 29 Hesperia city California 79,714 -0.19 0.15 0.34 170 131 30 Kent city Washington 84,979 0.09 0.43 0.34 124 98 31 Victorville city California 93,042 -0.04 0.29 0.32 142 115 32 Salem city Oregon 142,006 -0.32 0.00 0.32 187 157 33 Palm Bay city Florida 90,102 -0.56 -0.25 0.31 239 191 34 Bloomington city Indiana 55,406 -0.47 -0.16 0.31 213 178 35 Wyoming city Michigan 68,960 -0.52 -0.21 0.31 226 187 36 Irving city Texas 212,262 0.85 1.16 0.31 55 40 37 Costa Mesa city California 105,333 0.88 1.19 0.31 51 38 38 Scranton city Pennsylvania 67,314 -0.95 -0.65 0.30 337 274 39 Lewisville city Texas 81,484 -0.37 -0.08 0.29 199 166 40 Bryan city Texas 56,277 -0.02 0.27 0.29 138 117 Page 54
Table 16 lists the 40 cities that had the largest decrease in scores for the immigration and housing affordability factor. Table 16. Forty Cities with the Largest Decreases in Factor 2 Scores, 2000-2005 City State 2005 Population 2000 Factor 2 2005 Factor 2 Difference 2000 Rank 2005 Rank 1 Pasadena city California 129,400 0.92 0.34 -0.58 47 107 2 Santa Barbara city California 90,708 0.63 0.14 -0.49 69 134 3 San Marcos city California 77,445 0.79 0.33 -0.47 57 110 4 Baldwin Park city California 84,812 2.82 2.36 -0.46 4 8 5 Santa Ana city California 302,302 3.81 3.39 -0.42 1 3 6 Fort Lauderdale Florida 141,307 0.14 -0.26 -0.40 116 196 7 Elizabeth city New Jersey 121,137 2.32 1.93 -0.39 6 16 8 Alhambra city California 76,309 2.50 2.15 -0.35 5 12 9 McKinney city Texas 92,337 -0.27 -0.60 -0.33 182 260 10 Hayward city California 135,474 1.31 1.00 -0.30 28 57 11 Killeen city Texas 98,434 -0.65 -0.91 -0.27 271 332 12 Mountain View California 69,427 1.11 0.87 -0.24 39 61 13 Alameda city California 77,058 0.31 0.11 -0.20 96 139 14 San Jose city California 887,330 1.31 1.12 -0.19 27 45 15 Southfield city Michigan 75,053 -0.67 -0.87 -0.19 276 321 16 New Bedford city Massachusetts 84,898 0.30 0.10 -0.19 98 140 17 Chico city California 71,298 -0.36 -0.55 -0.19 195 249 18 Waukesha city Wisconsin 62,690 -0.72 -0.91 -0.19 291 330 19 Glendale city California 194,620 2.29 2.11 -0.18 8 13 20 Evanston city Illinois 62,258 -0.47 -0.65 -0.18 215 275 21 Westminster city Colorado 99,305 -0.54 -0.72 -0.18 234 290 22 Upland city California 74,420 -0.11 -0.29 -0.17 153 199 23 Chattanooga city Tennessee 139,158 -1.07 -1.24 -0.17 353 365 24 San Francisco city California 719,077 1.12 0.95 -0.17 37 59 25 Carson city California 92,156 0.66 0.49 -0.17 64 93 26 Suffolk city Virginia 77,922 -0.84 -1.00 -0.17 320 344 27 North Las Vegas Nevada 165,061 0.99 0.82 -0.16 42 63 28 Simi Valley city California 116,722 -0.38 -0.54 -0.16 201 245 29 Sioux City city Iowa 78,395 -0.58 -0.74 -0.16 245 298 30 Lorain city Ohio 65,476 -0.76 -0.91 -0.15 305 331 31 Roanoke city Virginia 90,074 -0.91 -1.06 -0.15 331 347 32 Pawtucket city Rhode Island 72,896 0.49 0.34 -0.14 79 108 33 Westland city Michigan 80,284 -0.68 -0.82 -0.14 277 315 34 Peoria city Illinois 102,136 -1.21 -1.35 -0.14 364 369 35 Honolulu CDP Hawaii 362,252 0.71 0.57 -0.14 62 87 36 Round Rock city Texas 81,639 -0.51 -0.65 -0.14 225 273 37 Newton city Massachusetts 82,383 -0.43 -0.57 -0.14 208 254 38 Los Angeles city California 3,731,437 2.01 1.88 -0.14 14 18 39 Fargo city North Dakota 88,809 -0.79 -0.92 -0.13 309 333 40 Buena Park city California 76,062 1.16 1.03 -0.13 33 55 Page 55
The five cities with the largest decreases, and 18 of the top 40, are in California; the California cities include three very large cities—Los Angeles, San Jose, and San Francisco. Five of the 10 cities with the highest scores on this factor in 2000 were among the 40 cities with the largest decreases. Fort Lauderdale had the greatest change in rank, moving from 116 th in 2000 to 196 th in 2005. 4.4. Comparison of Scores in 2000 and 2005 on Factor 3 The Factor 3 scores rank cities on community needs related to limited economic prospects. Between 2000 and 2005, the average score on Factor 3 declined, indicating that conditions improved on average for cities on th is dimension of need. Table 17 shows how the changes in scores for this factor varied by region and size class of cities. Table 17. Changes in Factor 3 Scores between 2000 and 2005, by Region and Population Region Number of Cities 2000 Factor 3 2005 Factor 3 Difference South 110 0.29 0.08 -0.21 West 150 0.26 0.02 -0.24 Midwest 69 0.05 -0.10 -0.14 Northeast 41 -0.07 -0.44 -0.36 Population 1,000,000+ 9 0.21 -0.08 -0.29 500,000-999,999 21 -0.16 -0.40 -0.24 300,000-499,999 23 0.09 -0.12 -0.21 200,000-299,999 37 0.22 0.03 -0.19 100,000-199,999 124 0.28 0.07 -0.21 under 100,000 156 0.17 -0.07 -0.24 All cities 370 0.19 -0.03 -0.23 On average, cities in every region and in every size class improved on this factor between 2000 and 2005. Cities in the Northeast had the lowest scores on this factor in 2000 and showed the greatest improvement between 2000 and 2005. There does not appear to be any consistent relationship between city size and either the Factor 3 scores or the changes in the Factor 3 scores. Table 18 lists the 40 cities that had the largest increase in community needs on the limited economic prospects factor between 2000 and 2005. Only 44 of the 370 cities became worse off on this factor. For 24 of these cities, the increases in Factor 3 scores were negligible, 0.10 standard deviations or less. Carrollton, TX; Gastonia, NC; and Cedar Rapids, IA had the largest increases. These increases combined with the general pattern of decreases created some large changes in the rankings on Factor 3. Carrollton moved from 214 th to 125 th ; Gastonia moved from 70 th to 27 th ; and Cedar Rapids moved from 268 th to 194 th . Page 56
Table 18. Forty Cities with the Largest Increases in Factor 3 Scores, 2000-2005 City State 2005 Population 2000 Factor 3 2005 Factor 3 Difference 2000 Rank 2005 Rank 1 Carrollton city Texas 122,699 0.12 0.42 0.30 214 125 2 Gastonia city N Carolina 72,183 0.91 1.14 0.23 70 27 3 Cedar Rapids city Iowa 119,670 -0.23 -0.01 0.22 268 194 4 Lawton city Oklahoma 79,486 0.96 1.15 0.19 63 26 5 Southfield city Michigan 75,053 0.43 0.62 0.19 162 91 6 Westminster city California 97,946 0.71 0.88 0.17 114 48 7 Irving city Texas 212,262 0.35 0.51 0.16 175 107 8 Arlington city Texas 348,965 0.48 0.63 0.16 152 85 9 Mesquite city Texas 126,895 1.47 1.62 0.15 19 5 10 Topeka city Kansas 117,326 0.31 0.45 0.14 184 122 11 Sioux City Iowa 78,395 0.77 0.90 0.13 101 44 12 Glendale city California 194,620 -0.62 -0.49 0.13 312 266 13 Bloomington city Indiana 55,406 -2.34 -2.21 0.13 366 364 14 Champaign city Illinois 65,600 -1.62 -1.51 0.11 355 348 15 High Point city N Carolina 101,852 0.65 0.76 0.11 122 66 16 Lorain city Ohio 65,476 1.42 1.53 0.11 21 8 17 Cary town N Carolina 107,446 -1.26 -1.16 0.10 344 331 18 Fayetteville city N Carolina 128,777 0.72 0.82 0.10 111 56 19 Lakewood city Colorado 142,434 0.05 0.14 0.09 234 173 20 Turlock city California 74,883 1.04 1.14 0.09 50 28 21 Midland city Texas 100,799 1.02 1.11 0.09 54 31 22 Dayton city Ohio 132,679 0.55 0.64 0.08 141 84 23 Orlando city Florida 221,299 -0.21 -0.13 0.07 264 220 24 Killeen city Texas 98,434 0.62 0.69 0.07 130 76 25 Chico city California 71,298 -0.93 -0.87 0.06 334 312 26 Gresham city Oregon 95,334 0.45 0.50 0.05 157 110 27 Plano city Texas 251,648 -0.72 -0.68 0.05 324 291 28 Berkeley city California 90,432 -3.26 -3.22 0.04 370 369 29 Tempe city Arizona 166,171 -0.64 -0.61 0.04 316 278 30 Wichita city Kansas 354,582 0.36 0.39 0.03 174 134 31 Garland city Texas 235,750 1.14 1.17 0.03 41 23 32 Toledo city Ohio 285,937 1.00 1.03 0.03 57 34 33 Hemet city California 77,076 1.34 1.37 0.03 24 13 34 Madison city Wisconsin 203,704 -2.04 -2.02 0.02 365 362 35 Pontiac city Michigan 59,472 1.13 1.16 0.02 42 24 36 Lubbock city Texas 199,789 0.68 0.70 0.02 115 74 37 Aurora city Colorado 291,317 0.37 0.40 0.02 170 133 38 Spokane city Washington 192,777 -0.11 -0.09 0.02 254 209 39 Clearwa ter city Florida 108,382 0.14 0.16 0.02 208 170 40 Cleveland city Ohio 414,534 0.72 0.73 0.01 110 70 Page 57
Table 19 lists the 40 cities that showed the greatest improvement on the limited economic prospects factor. Table 19. Forty Cities with the Largest Decreases in Factor 3 Scores, 2000-2005 City State 2005 Population 2000 Factor 3 2005 Factor 3 Difference 2000 Rank 2005 Rank 1 Deerfield Beach Florida 71,599 0.67 -0.27 -0.94 116 245 2 Davie town Florida 88,683 0.56 -0.20 -0.76 137 228 3 Indio city California 65,091 1.65 0.89 -0.76 12 46 4 Tustin city California 79,811 0.31 -0.40 -0.71 185 260 5 Newark city New Jersey 254,217 0.44 -0.24 -0.67 160 236 6 Hawthorne city California 100,754 1.41 0.74 -0.67 22 68 7 Upland city California 74,420 0.40 -0.27 -0.67 164 241 8 Allentown city Pennsylvania 105,231 0.21 -0.44 -0.65 197 262 9 Worcester city Massachusetts 154,398 -0.50 -1.14 -0.64 300 330 10 Lawrence city Massachusetts 82,191 -0.24 -0.88 -0.64 272 314 11 Miami Beach city Florida 84,086 -0.07 -0.70 -0.63 248 296 12 Jersey City New Jersey 246,335 -0.49 -1.11 -0.62 298 327 13 Baltimore city Maryland 608,481 0.20 -0.41 -0.62 200 261 14 Alhambra city California 76,309 -0.03 -0.64 -0.61 243 284 15 Cambridge city Massachusetts 81,260 -2.67 -3.28 -0.60 367 370 16 Cranston city Rhode Island 77,025 0.75 0.17 -0.57 105 169 17 Miramar city Florida 115,444 0.66 0.09 -0.57 118 182 18 Birmingham city Alabama 222,154 0.82 0.26 -0.56 90 155 19 Fayetteville city Arkansas 58,839 -0.89 -1.44 -0.55 329 342 20 Napa city California 73,085 0.21 -0.33 -0.55 196 254 21 Pittsburgh city Pennsylvania 284,366 -0.22 -0.76 -0.54 265 305 22 Carson city California 92,156 1.56 1.02 -0.54 15 35 23 Washington city District of Columbia 515,118 -1.74 -2.28 -0.54 359 366 24 Santa Maria city California 88,817 1.19 0.65 -0.54 35 83 25 Camden city New Jersey 73,305 0.77 0.23 -0.54 100 159 26 Hollywood city Florida 138,412 0.46 -0.07 -0.53 153 202 27 Santa Fe city New Mexico 66,453 -0.22 -0.75 -0.53 267 301 28 Fall River city Massachusetts 97,612 1.10 0.58 -0.52 44 97 29 Suffolk city Virginia 77,922 0.96 0.45 -0.51 62 121 30 Pleasanton city California 67,018 -0.68 -1.19 -0.50 321 332 31 Lowell city Massachusetts 96,876 -0.33 -0.83 -0.50 280 310 32 Portsmouth city Virginia 95,183 0.80 0.30 -0.50 92 149 33 Bellingham city Washington 69,057 -0.94 -1.44 -0.49 335 341 34 Livermore city California 87,054 -0.09 -0.58 -0.49 252 275 35 Paterson city New Jersey 148,353 0.61 0.13 -0.48 131 177 36 Baldwin Park city California 84,812 1.99 1.50 -0.48 3 9 37 New York City New York 7,956,113 -0.60 -1.08 -0.48 310 326 38 Savannah city Georgia 117,478 0.49 0.01 -0.48 149 190 39 Lafayette city Louisiana 108,175 0.39 -0.08 -0.47 167 207 40 Clovis city California 80,529 0.85 0.38 -0.47 82 135 Page 58
Deerfield Beach, FL—the city that experienced the greatest worsening of Factor 2— showed the greatest improvement on Factor 3. New York, Washington, Pittsburgh, Newark, Jersey City, and Birmingham were among the cities with the largest decreases in Factor 3 scores. There were 13 Northeastern cities on the list, nine more than would have been expected by chance. 4.5. Comparison of Scores in 2000 and 2005 on the Equal Weight Index Table 20 shows how the changes in scores for the equal weight index varied by region and size class of cities. According to the equal weight index, conditions improved in the West, were stable in the South, and worsened in the Midwest and Northeast. Overall conditions were generally stable; if the differences in the last column were carried to three decimal places as was done in Table 10, the difference for all cities would be -.002, a very small improvement at the national level. There was no consistent pattern in the changes by city size. According to the equal weight index, conditions were stable or got better in 202 cities. Table 20. Changes in Equal Weight Index Scores between 2000 and 2005, by Region and Population Region Number of Cities 2000 Equal Weight Index 2005 Equal Weight Index Difference South 110 -0.08 -0.07 0.00 West 150 0.04 0.01 -0.03 Midwest 69 -0.21 -0.17 0.05 Northeast 41 0.33 0.34 0.02 Population 1,000,000+ 9 0.46 0.42 -0.03 500,000-999,999 21 0.01 0.01 0.00 300,000-499,999 23 0.20 0.18 -0.02 200,000-299,999 37 0.01 0.03 0.02 100,000-199,999 124 -0.02 -0.02 0.00 under 100,000 156 -0.07 -0.07 -0.01 All cities 370 -0.01 -0.01 0.00 Table 21 lists the 40 cities that experienced the biggest worsening of conditions between 2000 and 2005. Nineteen of the 40 are cities in the Northeast and Midwest; the proportionate share of the 40 from these two regions would be 11. Dallas is the only city with 500,000 or more population. Camden, Passaic, and Lawrence, MA had high scores in 2000 and experienced big increases between 2000 and 2005. We were able to score 11 cities in North Carolina; four of them made the list of worst change in overall condition. Page 59
Table 21. Forty Cities with the Largest Increases in Equal Weight Index Scores, 2000-2005 City State 2005 Population 2000 EW Index 2005 EW Index Difference 2000 Rank 2005 Rank 1 Reading city Pennsylvania 81,302 0.93 1.34 0.40 17 6 2 Gresham city Oregon 95,334 -0.02 0.32 0.34 169 89 3 Garland city Texas 235,750 0.22 0.54 0.31 107 51 4 Camden city New Jersey 73,305 1.72 2.03 0.31 2 1 5 Redwood City city California 81,195 -0.18 0.12 0.30 228 141 6 Mesquite city Texas 126,895 -0.02 0.28 0.29 167 101 7 Springfield city Massachusetts 146,948 0.52 0.81 0.29 50 27 8 Dayton city Ohio 132,679 0.35 0.64 0.28 83 38 9 Union City city California 65,239 0.03 0.29 0.26 153 96 10 Hillsboro city Oregon 82,732 -0.17 0.09 0.26 226 147 11 Lawrence city Massachusetts 82,191 1.23 1.47 0.24 10 5 12 Bloomington city Indiana 55,406 -0.90 -0.66 0.24 361 334 13 Passaic city New Jersey 68,422 1.49 1.71 0.23 5 2 14 Cedar Rapids city Iowa 119,670 -0.60 -0.37 0.23 332 276 15 Scranton city Pennsylvania 67,314 -0.02 0.21 0.23 168 119 16 Hammond city Indiana 72,507 0.36 0.57 0.21 79 47 17 Gainesville city Florida 100,879 -0.41 -0.20 0.21 288 231 18 Palmdale city California 145,800 0.44 0.64 0.21 61 36 19 Irving city Texas 212,262 0.21 0.42 0.21 109 65 20 Trenton city New Jersey 77,471 0.86 1.07 0.21 22 14 21 Bryan city Texas 56,277 0.34 0.54 0.20 85 50 22 Cleveland city Ohio 414,534 0.75 0.96 0.20 30 18 23 Baton Rouge city Louisiana 205,442 0.10 0.30 0.20 135 94 24 Dallas city Texas 1,144,946 0.60 0.80 0.20 46 28 25 Carrollton city Texas 122,699 -0.33 -0.14 0.20 273 213 26 Aurora city Colorado 291,317 -0.04 0.16 0.19 175 131 27 Wyoming city Michigan 68,960 -0.09 0.10 0.19 195 142 28 West Covina city California 116,371 0.16 0.35 0.18 118 82 29 Greensboro city N Carolina 208,552 -0.25 -0.08 0.18 257 192 30 Gastonia city N Carolina 72,183 0.23 0.40 0.17 106 72 31 Aurora city Illinois 170,490 0.07 0.24 0.17 141 110 32 Arlington city Texas 348,965 -0.11 0.06 0.17 203 153 33 Rochester city New York 189,312 0.60 0.77 0.17 45 29 34 Champaign city Illinois 65,600 -0.76 -0.59 0.17 347 324 35 High Point city N Carolina 101,852 -0.01 0.16 0.16 164 130 36 Kansas City city Kansas 142,341 0.41 0.57 0.16 68 45 37 Lawton city Oklahoma 79,486 0.02 0.18 0.16 156 122 38 Winston-Salem N Carolina 183,467 -0.09 0.07 0.16 199 152 39 Somerville city Massachusetts 74,869 -0.28 -0.13 0.16 264 208 40 Brooklyn Park Minnesota 66,408 -0.38 -0.22 0.15 281 241 Page 60
Table 22 lists the 40 cities that experienced the greatest improvement in community needs between 2000 and 2005 as measured by the equal weight index. Table 22. Forty Cities with the Largest Decreases in Equal Weight Index Scores, 2000-2005 City State 2005 Population 2000 EW Index 2005 EW Index Difference 2000 Rank 2005 Rank 1 Pasadena city California 129,400 0.07 -0.28 -0.34 143 254 2 San Marcos city California 77,445 0.24 -0 .09 -0.33 103 197 3 Alhambra city California 76,309 0.68 0.37 -0.31 38 78 4 McKinney city Texas 92,337 -0.38 -0.67 -0.29 282 336 5 Suffolk city Virginia 77,922 -0.05 -0.32 -0.27 183 265 6 North Las Vegas Nevada 165,061 0.66 0.41 -0.25 40 68 7 Santa Ana city California 302,302 1.83 1.59 -0.24 1 3 8 Oceanside city California 162,259 0.06 -0.18 -0.24 149 225 9 Peoria city Illinois 102,136 -0.20 -0.43 -0.23 236 290 10 New York city New York 7,956,113 0.55 0.33 -0.22 48 87 11 Fort Lauderdale Florida 141,307 0.13 -0.09 -0.22 125 196 12 Miami city Florida 361,701 0.61 0.39 -0.22 44 74 13 Redding city California 89,362 -0.09 -0.31 -0.22 197 264 14 Santa Barbara city California 90,708 -0.17 -0.37 -0.20 225 274 15 Baldwin Park city California 84,812 1.52 1.32 -0.20 4 8 16 Miramar city Florida 115,444 0.04 -0.16 -0.20 151 215 17 Wilmington city N Carolina 91,207 -0.27 -0.47 -0.20 263 302 18 Buena Park city California 76,062 0.40 0.21 -0.19 70 118 19 Hawthorne city California 100,754 1.16 0.97 -0.19 12 16 20 Columbia city S Carolina 88,450 -0.12 -0.31 -0.19 209 263 21 Los Angeles city California 3,731,437 0.87 0.69 -0.19 20 33 22 Glendale city California 194,620 0.43 0.25 -0.18 63 109 23 Tustin city California 79,811 0.06 -0.13 -0.18 150 209 24 Indio city California 65,091 1.26 1.09 -0.18 7 13 25 Miami Beach city Florida 84,086 1.65 1.47 -0.18 3 4 26 Santa Monica city California 82,777 -0.74 -0.91 -0.18 344 362 27 Long Beach city California 463,956 0.62 0.45 -0.17 43 62 28 Pomona city California 161,257 1.26 1.09 -0.17 8 12 29 Alexandria city Virginia 133,479 -0.54 -0.72 -0.17 321 342 30 Burbank city California 100,053 -0.08 -0.25 -0.17 192 245 31 Asheville city N Carolina 74,889 -0.32 -0.49 -0.17 270 305 32 Stockton city California 278,515 0.78 0.61 -0.17 28 42 33 Fayetteville city Arkansas 58,839 -0.54 -0.71 -0.17 320 341 34 Roanoke city Virginia 90,074 0.12 -0.05 -0.16 129 183 35 Carson city California 92,156 0.42 0.26 -0.16 67 106 36 Des Moines city Iowa 196,917 -0.14 -0.30 -0.16 220 260 37 Cambridge city Massachusetts 81,260 -0.72 -0.88 -0.16 341 357 38 Newark city New Jersey 254,217 1.30 1.15 -0.16 6 10 39 Fullerton city California 142,064 0.00 -0.16 -0.16 162 218 40 Evanston city Illinois 62,258 -0.94 -1.10 -0.16 364 368 Page 61
The two largest cities, New York and Los Angeles, were among the biggest improvers. Consistent with the improvement in the West region, 21 of the 40 are California cities— 11 more than would have been expected by chance. Among the biggest improvers were cities that, in 2000, had been ranked number 1 in community needs (Santa Ana); number 3 (Miami Beach); number 4 (Baldwin Park, CA); number 6 (Newark); number 7 (Indio, CA); and number 8 (Pomona, CA). 4.6. Summary Factor analysis has allowed us to represent 24 needs indicators by three dimension of need: poverty and structural problems, immigration and housing affordability, and limited economic prospects. This chapter successfully applied factor analysis to compare conditions in 370 cities in 2000 and 2005 on each of the dimensions of needs and an equal weight index of community needs. While the equal weight index offers a reliable summary statistic on community needs, the analysis in this chapter shows that considering the individual factors separately paints a fuller picture of what is happening in American cities. The factor-by-factor analysis revealed the following: • Between 2000 and 2005, cities on average became worse off with respect to poverty and structural problems and with respect to immigration and housing affordability problems but became better off with respect to the limited economics prospects factor. • Regional differences appeared on the individual factors. o The Northeast has the highest average scores on the poverty and structural problems factor in both 2000 and 2005 and the largest increase in average scores between the two years. The West region has the lowest average scores on this fa ctor in both years and the smallest increase between the two years. o For the immigration and housing affordability factor, the average scores of cities in the Northeast and West were higher than the national average in both 2000 and 2005. Cities in the Northeast had the highest average change between 2000 and 2005. o Cities in the Northeast had the lowest scores on the limited economic prospects factor in 2000 and showed the greatest improvement between 2000 and 2005. Page 62
• Differences by class size of cities were less common. o There appears to be a systematic relationship between the scores on the poverty and structural problems factor and city size. The average score declines by size class in both 2000 and 2005. The change in scores is approximately the same for all the size classes, except for cities with populations between 500,000 and one million, which have a slight higher increase in average scores. o With the exception of cities with over a million residents, there appears to be little relationship between population size and the prevalence of problems related to immigration and housing affordability. The largest cities had an average score of 0.70 or more in both 2000 and 2005; the national average was 0.00 in 2005. • There were also some interesting patterns in the lists of cities with the biggest increase in scores (becoming worse off) and the lists of cities with the biggest decreases in scores (becoming better off). o Some of the worse off cities on the poverty and structural problems factor experienced big increases on this factor between 2000 and 2005; the cities were Camden, Detroit, Cleveland, Rochester, Reading, and Syracuse. o Compared with the other states, California has the most cities—95— among the 370 scored. Still, California cities appeared in higher than expected proportions on the lists of the 40 biggest losers and gainers. One would expect, proportionally, 10 cities from California on each list, yet: − Twenty-four of the 40 cities with the biggest improvements on the poverty and structural problems factor were California cities. − Fifteen of the 40 cities with the worse changes on the immigration and housing affordability factor were California cities. − The five cities with the largest improvements on the immigration and housing affordability factor, and 18 of the top 40 were California cities. The equal weight index showed that, on average, community needs decreased slightly between 2000 and 2005. According to the index, conditions were stable or got better in 202 cities. However, the chapter notes that the observed improvement appears to be related strongly to the substantial increase in the proportion of adults with a high school diploma between 2000 and 2005, a fact that was questioned in Section 2.3.2. Page 63
This page left blank intentionally. Page 64
Page 65 5. Measuring Fiscal Capacity The federal government, in general, and HUD in particular, are interested in developing an index of community needs because they want to know the extent to which communities require federal assistance. But a needs index answers only one-half of this question; the federal government also needs to know the extent to which communities are capable of dealing with their problems without federal assistance. At the Orientation Meeting for this project on October 12, 2006, the question was raised as to whether the project should attempt to construct a measure of fiscal capacity so that needs and capacity could be looked at jointly. Opinions differed. Those opposed to looking at capacity argued that needs are independent of the capacity of the local government to address those needs. In general, those present seemed to favor developing a parallel measure of capacity, because federal aid must take both needs and capacity into account. For this reason, we attempted to create a measure of capacity and to combine the needs and capacity measures to achieve an integrated view of local conditions. 35 This chapter explores the issues involved in estimating capacity, develops a methodology to measure capacity, implements the methodology, and then explores how to combine a measure of need with a measure of capacity. The capacity measure developed is a significant advancement in assessing need at the city level. It allows one to look at cities from two different perspectives—needs and capacity. While the work on combining the needs and capacity measures is only exploratory, the results are reasonable and provide the first comprehensive assessment of the relative need for federal assistance. Section 5.1 describes the methodology we used to construct a measure of fiscal capacity. Section 5.2 lists the variables we used to compute fiscal capacity and how we calculated the weights applied to those variables. Section 5.3 reports the results of implementing the measure of fiscal capacity. Section 5.4 explores how to combine a measure of fiscal capacity with an index of community needs. Section 5.5 reports the results from combining the equal weight index developed in Chapter 3 with the fiscal capacity measure developed in this chapter. Section 5.6 provides a brief summary. 5.1. General Approach We interpret capacity as “access to resources.” This approach equates capacity with the ability to raise money through taxation—that is, we are concerned with the fiscal capacity of cities. We measure fiscal capacity, independent of competence. If a city has an ineffective government but the same access to resources as another city with an effective government, then, in our opinion, the two cities should be considered to have equal capacity. We take this perspective based on the premise that achieving effective 35 The GAO has indicated that it intends to look at capacity in its study of community needs and the CDBG formula.
Page 66 government at the local level is the responsibility of citizens at the local level and not the responsibility of the federal government. A similar logic guides us to focus on the ability to raise money for community needs rather than actual performance in raising money or using funds for community needs. How much to tax and how to spend tax revenue are matters of choice. A measure of fiscal capacity should be independent of choice. This orientation affects how we measure fiscal capacity and the data sources we use. The Census of Governments collects extensive information on expenditures by class and on tax revenue by source, for units of government ranging from states to special districts. Unfortunately, we cannot use this information directly, because those data record the choices made by cities rather than the capacity of cities. Cities may differ in terms of what they spend on parks and recreation or how much they collect from property taxes, either because they differ in the capacity to raise revenue and address local needs or because they choose to tax and spend differently. For this reason, we chose to estimate what cities could raise through taxes, rather than what they do raise. This approach requires us to estimate various tax bases separately, such as income, sales, and property values, and to find a way to aggregate these bases. Income is a flow, while property values are a stock, and for this reason they are taxed differently. Adding personal income to the value of all property would not be a useful measure of fiscal capacity. The methodological problem is to find a set of weights to combine different tax bases; the weights should be chosen so as to represent the potential to raise taxes from each of the bases. We selected average tax rates as weights because they represent actual experience in translating taxable poten tial into tax revenue. 36 The choice of average tax rates raises two questions. First, the use of average rates may bias the measure toward current practice instead of what could be achieved if cities taxed to maximum capacity. The goal is not to create a measure of maximum potential tax revenue; instead the goal is create a measure that treats each city fairly in portraying its fiscal capacity. Actual practice seems to be the best way to achieve this fairness. Second, since state laws prohibit some cities from taxing certain bases, is it fair to use average rates when the actual tax rate applicable in a particular city may be zero? We do not consider this a major limitation. First, states could eliminate any restrictions on what can be taxed. The rationale behind creating a capacity measure is that the federal government should not respond to a problem if the local government has the capacity to solve the problem. Failure of a state to provide jurisdictions with the tools needed to deal with their problems should not be a reason for federal support. Second, if a jurisdiction cannot tax one source, it can increase the rate at which it taxes other sources. The real 36 Table 23 shows how we computed the average tax rates and applied them to estimates of income, value, or sales as appropriate. In general, we used information from the 2002 Census of Government on revenue raised by various taxes as the numerator for the average tax rates, and estimates of the taxable source in 2002 from other data sets. For property taxes, we used information in the 2001 Residential Finance Survey to estimate the average tax rate.
Page 67 limit on the ability to raise taxes is the willingness of voting taxpayers to be taxed. The willingness or ability of taxpayers to bear taxes should depend more on how much is raised than on how the taxes are collected. A final concern is the difference in the cost of providing services across cities. Cities with the same taxing capacity may not be able to provide the same level of services because it costs more in one city to provide services than it does in the other city. We solve this problem by dividing the dollar measure of capacity by the average wage of all government employees in the core-based statistical area in which the city is located. 5.2. Variables Used to Measure Fiscal Capacity Table 23 explains how we constructed the dollar measure of capacity. • Variables 2 through 6 are our estimates of the various tax bases potentially available to cities, each weighted by an estimate of the applicable average tax rate. • Variable 1 sums variables 2 through 6 and puts the combined taxing capacity on a per capita basis; it is our estimate of the capacity that cities have available from their own resources. • Variable 7 is our estimate of what cities can expect in funding from states. • Variable 8 is our estimate of fiscal capacity in dollar terms. It sums the capacity available to cities from their own resources (variable 1) and the capacity available to them from their respective states (variable 7). • Variable 9 is per capita income. We compare our estimate with an estimate based strictly on per capita income. • Variable 10 is our government wage variable. 37 • Variable 11 is our estimate of real capacity; it is variable 8 divided by variable 10. A HUD reviewer pointed out that the use of local government wages as a cost-of-services deflator may create a bias, because cities differ in how they contract out services. If one city contracts out services that employ low-wage workers, its wage rate may be higher than an identical city that provides those services directly. We acknowledge that this is a potential bias, but we believe it is minimized by our use of local wage data at the level of 37 A HUD reviewer suggested using the federal locality pay tables instead of the BLS data on government wages. We think that it would be complicated trying to adjust the federal tables for the local mix of white collar and blue collar workers. The BLS data cover all local employees.
Page 68 Core-Based Statistical Areas (CBSA). 38 Contracting patterns will vary across the CBSA, and there should be some averaging-out of practices. Another HUD reviewer noted that we use 2005 population to transform our estimates of capacity into per capita estimates while some of our data are from 2002. Measures 2, 3, 4, and 7 use 2005 data; measures 5 and 6 use 2002 data. 39 We think 2005 population is probably the best choice for this mix. 38 CBSAs are defined by OMB using criteria that make CBSAs coterminous with local labor markets, and therefore all governments in a CBSA should face the same wage scale. 39 The weights we apply to measures 2, 3, 4, and 7 are based on data from 2001 and 2002, but the numbers that vary across cities are from 2005.
Table 23. Variables Used to Measure Fiscal Capacity Variable Short-Name Definition 1 LOCAL FISCAL CAPACITY LOCFISCAP Sum variables 2 through 6 and divide the sum by city population 2 INCOME TAX CAPACITY INCTAXCAP Aggregate household income multiplied by the ratio of local income taxes from 2002 Census of Governments to national household income from 2002 ($17,185,681,000/$6,142,192,043,242 = 0.003) 3 RESIDENTIAL PROPERTY TAX CAPACITY (owner-occupied) OWNPROPTAXCAP Aggregate property value multiplied by ratio of real estates paid to owner- occupied housing value from 2001 Residential Finance Survey (RFS) (0.01). 4 RESIDENTIAL PROPERTY TAX CAPACITY (rental) RENTPROPTAXCAP Aggregate contract rent multiplied by ratio of net operating income to rent income from 2001 RFS divided by national cap rate (from Goodman 40 ) multiplied by ratio of real estates paid to rental housing value from 2001 Residential Finance Survey. We do this estimate separate for central cities and non-central cities because found that cap rates vary between central cities and suburbs. For central cities: (Estimate aggregate contract rent)*95*0.01 For non-central cities: (Estimate aggregate contract rent)*125*0.01 5 SALES TAX CAPACITY SALESTAXCAP Retail sales in 2002 from the Economic Census multiplied by ratio of local revenue from the sum of the general sale taxes and selective sales taxes in all localities from the 2002 Census of Governments to total retail sales in 2002 from Census of Economic Census ($61,761,893,000/$3,056,421,997,000 = 0.02) 6 BUSINESS TAX CAPACITY (includes payroll taxes, business property taxes, and corporate income taxes) BUSTAXCAP Aggregate payrolls for the 12 sectors for which the 2002 Economic Census provides place data times 0.05. 0.05 was derived by estimating commercial real estate taxes, local corporate taxes, and proportion of local income taxes attributed to non-residents working in city; summing these items; and taking the ratio of this sum to the US total of payrolls from the 12 sectors. 7 STATE CAPACITY TO ASSIST JURISDICTION STATECAP Aggregate state income times (the ratio of state aid to local governments to total state revenue from all sources) times (the ratio of total state revenue from all sources to aggregate state income) times (the ratio of city population to the sum of the populations of all cities in our list in that state) divided by (city population). (Aggregate state income times 0.007 divided by the sum of the populations of all cities in our list in that state.) Note that this provides a per capita amount conceptually available to all cities. 8 TOTAL FISCAL CAPACITY TOTFISCAP Sum of variables 1 and 7 9 PER CAPITA INCOME PERCAPINC City per capita income 10 AVERAGE GOVERNMENT WAGES CBSAGOVWAGE The average wage rate of local government employees measured at the CBSA level using 2004 Bureau of Labor Statistics data. 11 REAL FISCAL CAPACITY REALFISCAP Variable 8 divided by variable 10. Page 69 40 Issues in Housing Finance: An Analysis of Data from the 2001 Residential Finance Survey, Chapter 11: Estimating Capitalization Rates for Multifamily Rental Properties with the 2001 Residential Finance Survey, a report submitted by Econometrica, Inc. to HUD, October 30, 2006.
Page 70 We can estimate state funding for all our cities and have data on wages for all but five cities. We were able to estimate variables 2 through 6 for 266 of the 473 cities, and therefore could estimate real fiscal capacity only for these 266 cities. The next section contains our findings. First, however, we note that the correlation of per capita income with the revenue that a city itself can raise (variable 1) is 0.78. If one wanted to create estimates of real fiscal capacity for other cities, one could do so by using per capita income and other relevant variables to model variable 1. Second, we also note that adjusting capacity for the cost of providing services eliminates the need to adjust needs for differences in costs of living across jurisdictions. This does not apply to the counts of poor persons, but does apply to dealing with the needs of citizens, including poor persons. 5.3. Variations in Fiscal Capacity Deflating total fiscal capacity by average local government wages translates a dollar measure of capacity into a measure of the fraction of a year that a city could apply to a local government worker’s time on a per capita basis to the needs of its citizens. Across the 266 cities for which we have data, this measure ranges between 0.026 in Laredo, TX to 0.136 in Charleston, SC. The mean is 0.059 with a standard deviation of 0.018, and the median is 0.057. This measure implies that, on average, cities should be able to devote the equivalent of 6 percent of a year’s work from the typical city employee to solving the needs of a specific citizen. We say “equivalent,” because a city will devote some of its revenue to purchases of goods and services other than labor, for example, interest on bonds, rent on buildings, and supplies. Since our estimate sums several different sources of tax revenue, it probably overestimates the amount available on average since most cities do not use all the sources. This is not a problem because the index does not try to measure absolute capacity—only relative capacity. To provide some impression of how the fiscal capacity measure performs, we produced three tables. Table 24 lists the 25 cities with the highest fiscal capacity scores based on the sum of real own resources and state resources per capita; Table 25 lists the 25 cities with the lowest fiscal capacity scores, and Table 26 shows how this measure varies across the 25 largest cities for which we have sufficient data to estimate capacity. 41 Tables 24 and 25 show how the presence or absence of resources (income and wealth) and the costs of providing services interact to determine rank on this measure. Table 24 contains both cities with high per capita income and cities with low government wages. Eight of the 25 cities in Table 24 (Santa Monica, Scottsdale, Cambridge, Bellevue, Santa Barbara, Santa Fe, Atlanta, and Boulder) are among the 25 with the highest per capita income among the 266 cities for which we have data. At the same time, four of the 25 (Fort Smith, Boise, Salt Lake City, and North Charleston) are among the 25 with the 41 Since we have data on both needs and real capacity for only 266 cities, many of the cities discussed in Chapter 4 do not appear in this analysis. This includes some cities with very high need scores such as Camden and Trenton.
lo west annual wages for government employees based on data at the metropolitan-area level. Table 25 contains both cities with low per capita income and cities with high government wages. Nine of the 25 cities in Table 25 (Brownsville, Laredo, Hialeah, Gary, San Bernardino, Pomona, Inglewood, and Flint) are among the 25 with the lowest per capita income among the 266 cities for which we have data. At the same time, two of the 25 (Mount Vernon and Yonkers) are among the 25 with the highest annual wages for government employees based on data at the metropolitan-area level. Brownsville is an interesting case; it has the lowest per capita income of any of the 25 cities and the eighteenth lowest annual government wages. Despite low government wages, Brownsville ranks as the city with the third lowest real fiscal capacity. Table 24. Twenty-five Cities with Greatest Real Fiscal Capacity City State Population Real Own + State Resources Per Cap 1 Charleston city South Carolina 109,151 0.136 2 North Charleston city South Carolina 70,001 0.121 3 Cambridge city Massachusetts 81,260 0.116 4 Bloomington city Minnesota 80,055 0.106 5 Boulder city Colorado 83,432 0.105 6 Santa Monica city California 82,777 0.104 7 Atlanta city Georgi 394,929 0.101 8 Bellevue city Washington 114,748 0.100 9 Nashua city New Hampshire 84,632 0.099 10 Fort Smith city Arkansas 81,054 0.096 11 Little Rock city Arkansas 176,924 0.095 12 Asheville city North Carolina 74,889 0.092 13 Billings city Montana 92,844 0.092 14 Fargo city North Dakota 88,809 0.090 15 Salt Lake City Utah 182,670 0.089 16 West Palm Beach city Florida 86,804 0.089 17 Fort Lauderdale city Florida 141,307 0.088 18 Scottsdale city Arizona 215,933 0.087 19 St. Cloud city Minnesota 59,624 0.086 20 Santa Barbara city California 90,708 0.085 21 Santa Fe city New Mexico 66,453 0.085 22 Norwalk city Connecticut 86,354 0.085 23 Boston city Massachusetts 520,702 0.085 24 Boise City Idaho 191,667 0.085 25 Manchester city New Hampshire 109,308 0.085 Page 71
Table 25. Twenty-five Cities with the Least Real Fiscal Capacity City State Population Real Own + State Resources Per Cap 1 Laredo city Texas 207,787 0.026 2 Inglewood city California 120,204 0.028 3 Brownsville city Texas 171,528 0.028 4 Pomona city California 161,257 0.030 5 Pasadena city Texas 150,180 0.032 6 Mount Vernon city New York 65,354 0.032 7 Garland city Texas 235,750 0.032 8 San Bernardino city California 204,552 0.033 9 Gary city Indiana 97,057 0.033 10 Hemet city California 77,076 0.033 11 El Paso city Texas 583,419 0.034 12 Detroit city Michigan 836,056 0.034 13 Garden Grove city California 192,345 0.034 14 Glendale city Arizona 229,913 0.035 15 Mesquite city Texas 126,895 0.035 16 Kansas City Kansas 142,341 0.035 17 Hialeah city Florida 213,791 0.035 18 Stockton city California 278,515 0.036 19 Flint city Michigan 111,948 0.036 20 Peoria city Arizona 141,941 0.036 21 Fresno city California 477,251 0.037 22 Long Beach city California 463,956 0.037 23 Mesa city Arizona 442,445 0.037 24 Yonkers city New York 193,327 0.037 25 Arlington city Texas 348,965 0.037 Page 72
Table 26. Real Fiscal Capacity for 25 Largest Cities with Data City State Population Real Own + State Resources Per Cap 1 New York city New York 7,956,113 0.041 2 Los Angeles city California 3,731,437 0.043 3 Chicago city Illinois 2,701,926 0.049 4 Houston city Texas 1,941,430 0.053 5 Philadelphia city Pennsylvania 1,406,415 0.047 6 Phoenix city Arizona 1,377,980 0.043 7 San Diego city California 1,208,331 0.065 8 San Antonio city Texas 1,202,223 0.038 9 Dallas city Texas 1,144,946 0.056 10 San Jose city California 887,330 0.051 11 Detroit city Michigan 836,056 0.034 12 Jacksonville city Florida 768,537 0.050 13 Indianapolis (balance) Indiana 765,310 0.064 14 San Francisco city California 719,077 0.064 15 Columbus city Ohio 693,983 0.057 16 Austin city Texas 678,457 0.059 17 Memphis city Tennessee 642,251 0.049 18 Baltimore city Maryland 608,481 0.066 19 Fort Worth city Texas 604,538 0.041 20 Charlotte city North Carolina 601,598 0.066 21 El Paso city Texas 583,419 0.034 22 Milwaukee city Wisconsin 556,948 0.048 23 Denver city Colorado 545,198 0.066 24 Las Vegas city Nevada 538,653 0.040 25 Seattle city Washington 536,946 0.082 In the 25 largest cities, local governments can devote between 0.03 to 0.08 percent of a city government employee’s time to the needs of each resident. The largest cities tend to have lower fiscal capacity, 17 of the 25 largest have a fiscal capacity score below the average of 0.059. 5.4. Combining Need and Fiscal Capacity 5.4.1. Background The primary reason for developing a measure of real fiscal capacity was to complete the picture of city need for federal community development assistance. It was felt that the need for federal help depends both on community needs and the resources available at the community level to deal with those needs. However, an important question remains: How does one combine information on community needs and real fiscal capacity to obtain an accurate assessment of need? Our aims in this section are modest. We discuss Page 73
some ideas we have about combining the needs and real fiscal capacity measures and explore one simple way to combine them. We experiment with combining our estimate of city needs from Chapter 3 with the estimate of the capacity of cities to deal with their needs from this chapter. There are 234 cities for which we have both an equal weight index score and an estimate of real fiscal capacity. 5.4.2. Simple Options for Combining the Measures In Chapter 4, we looked at city needs in two different ways: we looked at conditions measure by each of the three factors from the factor analysis in Chapter 3, and then we combined the factor scores into a single-valued index of need using several different weighing options. At this stage, we have to choose a single-valued needs index because it would be meaningless to combine the real fiscal capacity measure with the individual factors (“meaningless” in the sense that we would still be missing parts of the puzzle— namely, needs measured by the omitted factors). For the purposes of this work, we will use the equal weight index because it is the simplest. Any of the other options could have been combined with the real fiscal capacity measure in the manner described below. Our first thoughts involved taking a ratio of needs to capacity as measured by the two indices. The attraction of this approach was the possibility of making statements—such as a city’s needs exceed its capacity by 10 percent. However, the ratio approach, and all approaches involving a multiplicative joining of the indices, falters because the two measures have both positive and negative values. There is no way to combine series with positive and negative values in a multiplicative way and obtain consistent results. We define a positive needs score as meaning a city has above-average community needs and a positive capacity score as meaning a city has above-average real fiscal capacity. Negative scores mean being below average on both of the component indices. Using these conventions, the following table shows how the various possibilities combine to produce either positive or negative ratios. Sign of the ratio of indices Above average needs Below average needs Above average fiscal capacity Positive Negative Below average fiscal capacity Negative Positive In this table, the best possible situation (having below-average needs and above-average capacity) has the same sign as the worst possible situation (having above-average needs and below-average capacity). This result is not desirable. One could try to avoid this problem by redefining the real fiscal capacity index so that having above-average capacity results in a negative score and havi ng below-average capacity results in a positive score. If one does this, then the previous table becomes: Page 74
Page 75 Sign of the ratio of indices Above average needs Below average needs Above average fiscal capacity Negative Positive Below average fiscal capacity Positive Negative Now the best possible situation and the worst possible situation both produce positive ratios. Again, this is not a desirable result. The next simplest approach was an additive approach. We standardized both the scores from the equal weight index and our estimates of real fiscal capacity. After standardization, a need score of +1.00 indicates that the needs of a city are one standard deviation above the needs of the average city, while a real fiscal capacity score of +1.00 means that a city’s capacity to meet the needs of its citizens is one standard deviation above the average city. We subtract the standardized capacity score from the standard needs score, so that a city with needs of +1.00 and capacity of +1.00 would have a combined score of 0.00. Both the needs score and the capacity score refer to 2005. The following table characterizes the results of combining the indices using this approach. Sign from the subtraction of indices Above average needs Below average needs Above average fiscal capacity Uncertain Negative Below average fiscal capacity Positive Uncertain Now the best possible situation (below-average needs and above-average capacity) has a negative sign, while the worst possible situation (above-average needs and below-average capacity) has a positive sign. The signs of the other two alternatives depend on the magnitude of the indices involved in the subtraction. If needs are above average by more than capacity is above average, the score will be positive. If needs are below average by less than fiscal capacity is below average, the score will also be positive. While the technique of subtracting the real fiscal capacity index from the community needs index is simple, it does produce a consistent ranking of cities. 42 The scores produced by subtracting the index of real fiscal capacity from the index of community needs only rank cities by relative need; they do not indicate whether an individual city needs federal assistance and, if so, how much. Consider the situation of a city with a community needs score of 1.00 and a real fiscal capacity score of 1.00; the combined index produces a score of 0.00. This score tells us nothing about the absolute need for federal assistance; it does tell us only that the city needs assistance more than cities with negative scores and less than cities with positive scores. 42 This approach assumes an implicit equivalence between the standard deviations of needs and real capacity; in other words, it assumes that being one standard deviation away from the average has the same implications for the two indices. A city with average needs but real capacity one standard deviation below average will have a score of 1.00; a city with average needs one standard deviation above average but with average real capacity will also have a score of 1.00. This approach presumes that these two scores indicate the same level of need from the federal government.
By standardizing the two components before combining them, we produce an index where 0.00 is the average value. Does a city where needs and capacity balance need assistance? On average, are cities able to deal with their problems or on average do they need assistance? None of the analysis in this chapter or anywhere in this report is capable of answering these questions. This is perhaps the fundamental issue of fiscal federalism. 5.5. An Index of Needs Adjusted for Capacity Combining the two indices produces an index of community needs adjusted for real fiscal capacity. As explained in Section 5.4.2, we calculate it, first, by standardizing both the equal weight index of community needs and the index of real fiscal capacity across the 234 cities for which we have data and then by subtracting the index of real fiscal capacity from the equal weight index of community needs. For simplicity, we call the resulting index, the adjusted needs index. Table A.10 in Appendix A reports the score on the adjusted needs index for all 234 cities. The adjusted needs index has a mean of 0.00 and a standard deviation of 1.76. Positive values indicate adjusted needs higher than average, and negative values indicate adjusted needs lower than average. Adjusted needs ranges from 4.29 (Santa Ana, TX) to -5.28 (Charleston, SC). The range is large (9.57), and the standard deviation is large relative to the standard deviations of the component indices because there is moderately strong negative correlation between community needs and fiscal capacity. The correlation between the two component indices is -0.55. In general, the greater the needs of a city, the less real fiscal capacity the city has. This is an important finding. From the 234 cities for which we have data on both needs and capacity with the highest index of community needs, Table 27 takes the 50 cities with the highest scores on the equal weight index of community needs and shows how taking fiscal capacity into account would affect their ranking. Because of the negative correlation between the two component indices, adjusting needs for capacity does not produce large changes in the rankings of these cities. Only nine of the cities ranked among the top 50 on community needs have a rank lower than 50 on the combined index; these are Dayton (25 to 52), Birmingham (34 to 98), Lynn (36 to 53), Irving (43 to 115), Waco (44 to 69), Jersey City (45 to 57), Miami Beach (46 to 81), Memphis (48 to 62), and Escondido (50 to 65). Page 76
Table 27. Impact of Combining Needs and Fiscal Capacity for 50 Cities with the Highest Community Needs Equal Weight Index Rank City State Population Equal Weight Index Real Fiscal Capacity Adjusted Needs Index Adjusted Needs Index Rank 1 Passaic city New Jersey 68,422 3.28 -0.959 4.24 2 2 Santa Ana city California 302,302 3.04 -1.245 4.29 1 3 Miami city Florida 361,701 2.82 -0.116 2.93 13 4 Lawrence city Massachusetts 82,191 2.81 -1.011 3.82 3 5 Reading city Pennsylvania 81,302 2.55 -0.277 2.83 15 6 Salinas city California 156,950 2.52 -1.239 3.76 4 7 Paterson city New Jersey 148,353 2.29 -0.984 3.28 10 8 Newark city New Jersey 254,217 2.18 -0.704 2.89 14 9 San Bernardino California 204,552 2.12 -1.531 3.65 6 10 Pomona city California 161,257 2.07 -1.692 3.76 5 11 Detroit city Michigan 836,056 1.89 -1.460 3.34 9 12 Elizabeth city New Jersey 121,137 1.84 -0.726 2.56 16 13 Cleveland city Ohio 414,534 1.81 -0.394 2.21 24 14 Inglewood city California 120,204 1.81 -1.809 3.62 7 15 Oxnard city California 178,871 1.81 -1.127 2.93 12 16 Pasadena city Texas 150,180 1.79 -1.564 3.36 8 17 Garden Grove California 192,345 1.58 -1.437 3.01 11 18 Ontario city California 156,679 1.55 -0.729 2.28 21 19 Houston city Texas 1,941,430 1.54 -0.365 1.91 29 20 Springfield city Massachusetts 146,948 1.52 -0.407 1.93 28 21 Dallas city Texas 1,144,946 1.51 -0.237 1.74 35 22 Rochester city New York 189,312 1.46 -0.278 1.74 37 23 Yakima city Washington 79,517 1.37 -0.317 1.69 40 24 Los Angeles city California 3,731,437 1.29 -0.961 2.25 22 25 Dayton city Ohio 132,679 1.20 -0.240 1.44 52 26 Anaheim city California 329,483 1.17 -0.930 2.10 26 27 Stockton city California 278,515 1.13 -1.342 2.48 19 28 New Bedford Massachusetts 84,898 1.12 -0.782 1.90 30 29 Kansas City Kansas 142,341 1.07 -1.383 2.45 20 30 Hemet city California 77,076 1.06 -1.488 2.55 17 31 Providence city Rhode Island 160,264 1.04 -0.831 1.87 32 32 Buffalo city New York 256,492 1.03 -1.178 2.21 23 33 Garland city Texas 235,750 1.00 -1.548 2.55 18 34 Birmingham city Alabama 222,154 1. 00 0.706 0.29 98 35 Fall River city Massachusetts 97,612 0.95 -0.783 1.74 36 36 Lynn city Massachusetts 83,419 0.90 -0.512 1.41 53 37 Pawtucket city Rhode Island 72,896 0.86 -1.119 1.98 27 38 Milwaukee city Wisconsin 556,948 0.86 -0.637 1.49 50 39 Lowell city Massachusetts 96,876 0.85 -0.699 1.54 47 40 Philadelphia city Pennsylvania 1,406,415 0.83 -0.744 1.58 44 41 Long Beach city California 463,956 0.83 -1.294 2.12 25 42 Syracuse city New York 132,495 0.82 -0.563 1.39 55 Page 77
Table 27. Impact of Combining Needs and Fiscal Capacity for 50 Cities with the Highest Community Needs (continued) Equal Weight Index Rank City State Population Equal Weight Index Real Fiscal Capacity Adjusted Needs Index Adjusted Needs Index Rank 43 Irving city Texas 212,262 0.77 0.871 -0.10 115 44 Waco city Texas 107,146 0.75 -0.405 1.15 69 45 Jersey City New Jersey 246,335 0.74 -0.625 1.36 57 46 Miami Beach Florida 84,086 0.72 0.108 0.61 81 47 Oakland city California 373,910 0.72 -0.990 1.71 39 48 Memphis city Tennessee 642,251 0.71 -0.583 1.29 62 49 Fort Worth city Texas 604,538 0.68 -1.048 1.73 38 50 Escondido city California 133,017 0.67 -0.567 1.24 65 Looking at all 266 cities, some of the changes in ranking are substantial. West Palm Beach is ranked 65 th on the equal weight index but 182 nd on the adjusted needs index; Atlanta is ranked 95 th on the equal weight index but 211 th on the adjusted needs index. Other cities benefit substantially by the adjustment for fiscal needs. Peoria is ranked 190 th on the equal weight index but 89 th on the adjusted needs index; Henderson, NV is ranked 208 th on the equal weight index but 108 th on the adjusted needs index. 5.6. Summary This chapter explains why it is important to construct an index of fiscal capacity, develops a methodology to construct such an index, finds data to implement the index, and suggests a reasonable approach for combining information on community needs and real fiscal capacity to obtain an index of community needs adjusted for real fiscal capacity. The most important findings from the chapter are: • It is possible to construct an index of real fiscal capacity. • The index used is sensitive to both income and wage rates. Places with high income or lower government wages are more likely to have high scores on the index—that is, to have better-than-average fiscal capacity. • The index is negatively correlated with the equal weight index of community needs. Cities with high community needs are more likely to have low real fiscal capacity. • It is possible to combine a needs index and a fiscal capacity. The adjusted needs index developed in this chapter produced different rankings than the equal weight index of community needs. But, in general, the change in rankings was not great, probably because of the negative correlation between the two component indices. Page 78
6. Implications This chapter discusses the implications of this research for future analysis in areas of operational interest to HUD. The topics covered in successive sections are: 1. Feasibility of using ACS data to monitor community needs at the city and county levels. 2. Feasibility of constructing a single-valued index of community needs. 3. Feasibility of comparing community needs at different points in time. 4. Feasibility of extending the factor analysis developed in this report to different geographies when the ACS releases data for smaller places. 5. Feasibility of constructing a measure of neighborhood improvement to implement the Administration proposal to reward communities for successful community developments efforts. 6. Feasibility of measuring fiscal capacity. 7. The relevance of boundary changes and cost-of-living differences to an analysis of community needs. The last section identifies areas where more work needs to be done to improve HUD’s ability to monitor community needs. 6.1. ACS Data and Community Needs In previous studies, researchers at HUD used data from the long form of the decennial census and other sources to identify and measure community needs. The Census Bureau has replaced the long form with the American Community Survey, a monthly survey of 250,000 households that reports data using 5-year moving averages for all levels of geography, using 3-year moving averages for places with 20,000 or more persons, and using annual data for places with 65,000 or more persons. One objective of this research was to test whether the ACS data would support the same type of analysis that HUD had conducted using long-form data. The answer to this question is “yes.” In the future, HUD can depend on the ACS to monitor conditions in cities and counties. The report successfully uses ACS data to construct useful measures of community needs using factor analysis. Of the 24 needs indicators used in the final factor analysis, 16 used ACS data, one (POORPERS) used ACS data combined with long-form data, and four used long-form data. All five Page 79
Page 80 indicators that either used long-form data or a combination of ACS and long-form data should be available in the future from the ACS. The study did not use as a needs indicator the proportion of persons with a disability that limits employment. This variable is reported by both the ACS and the long form, but a change in the way the Census Bureau collects this information created an artificial shift in the published results. HUD should be able to use this variable as a need indicator in future analyses. There are some issues and open questions that HUD will have to keep in mind in future work using the ACS: • The reporting rules used in the ACS are similar to those used for the long form of the decennial census; but because the ACS sample size is smaller, the rules can result in more frequent suppression of data. Data on overcrowded housing was missing for so many places in the 2005 ACS data that we substituted 2000 long- form data for this need indicator. We also had to drop 36 cities from the analysis because the ACS suppressed data on the number of minorities in the cities. HUD should be able to work around suppression of data at the city or county level by using 3-year or 5-year moving average data. There will be no solution to suppression of data at the tract level. • As a general policy, the Census Bureau plans to release for the ACS all tables released for the 2000 long-form data. However, some special tabulations that were made public for long-form data have not yet been released for ACS data. The ones relevant to community needs analysis involve the intersection of information on poverty level, age of housing, and tenure. HUD should probably contact the Census Bureau to make sure that these tabulations are not forgotten. • The ACS has not released data on persons in group quarters yet, so we have no experience with the usefulness of the tabulations or the reliability of the data. • The Census Bureau will make revisions to the ACS questionnaire. Revisions always create the possibility of discontinuities in the data, such as the one that occurred with the disability questions. Users will have to be aware of possible problems. 43 43 The Census Bureau plans to restrict revisions to the ACS questionnaire to once every decade. All changes will be incorporated into the surveys beginning in the year ending in an “8,” e.g., 2008. Using this procedure, the first 5-year moving average using the new questionnaire for all 5 years will be centered on a year ending in “0,” e.g., 2010. This procedure ensures that the data centered on a decennial census year will all use the same questionnaire.
6. 2. A Single-Valued Index of Community Needs The explicit goal of this project was to develop a single-valued index of community needs. The research achieved this objective, but the outcome was only a qualified success. Nevertheless, the research resulted in two important insights: first, that an equal-weighted index correlates highly with most reasonable alternatives, and second, that changes in individual factor scores provide more useful information than do changes in an index, for understanding how conditions in individual cities have changed. The research also explored a new methodology that may prove useful in related future work and incorporated fiscal capacity into the analysis. Using data from the ACS and other sources, this report carried out a factor analysis that identified three dimensions of community needs: needs associated with poverty and structural problems, needs associated with immigration and lack of affordable housing, and needs arising from limited economic prospects. The report used these factors in an equal weight index that ranked cities on community needs. The equal weight index was one of six indices that the report considered. The report was unable to find any statistical, programmatic, or logical reasons that make a compelling case for choosing one index over any of the others. We had hoped that the hedonic analysis would provide definitive guidance in weighting the factors or even the needs indicators. However, the prominent role of housing affordability in Factor 2 and in two or three of the needs indicators undermined attempts to apply the hedonic results directly. Index 4, which provides a triple weight to the limited economic prospects factor, is derived from the hedonic analysis. Index 6 applies information from the hedonic analysis to derive weights. This approach is interesting, but we did not have the opportunity to examine its strengths and weaknesses. We consider the equal weight index to be only a qualified success because of our inability to justify the choice of weights. The efforts to develop a single-valued index did lead to two useful insights. First, comparing the six indices indicated that the rankings of cities across the other five indices were highly correlated with the ranking on the equal weight index. Statistically, the equal weight index produces results that are very similar to the other indices. Statistical closeness does not mean that the ranking of some cities are not substantially different depending upon the index used. If HUD were to use one of these indices to allocate funds to cities, the choice of index would be of great concern to individual cities. But, if HUD is interested primarily in analyzing the variation in needs across cities and over time, then the results from the equal weight index will be similar to those from any index that applies reasonable weights to the factor scores. Second, the analysis revealed that a single-valued index provides simplicity at the cost of concealing interesting information. If HUD is interested in how community needs vary across cities and over time, it should look at variation and changes in both the individual factors and a single-valued index. Page 81
The most important contribution of the research to the development of an index of community needs was the development of an index of real fiscal capacity and the development of a technique to combine an index of community needs with an index of fiscal capacity. From the perspective of the federal government, needs cannot be considered without reference to capacity. We discuss further the implications of the work on fiscal capacity in Section 6.6. 6.3. Intertemporal Comparisons of Needs This research developed a methodology for applying factor analysis to data on needs at two points in time and successfully implemented the methodology. The methodology allows HUD to compare factor scores from two different years to measure how conditions have changed in individual cities. Because the factor scores are computed from sample data, they are subject to sample variation and thus some care must be taken in interpreting the results. But the sample variation problem is no more serious than comparison involving sample data. There are two keys to carrying out intertemporal comparisons correctly. First, one must be sure that the dimensions of need identified in the base year are still relevant in the comparison year. Second, one must measure needs relative to conditions in the year in which the factor analysis is performed—that is, one must use the means and standard deviations from that year to standardize the needs indicators in both years. After reviewing the first draft of this report, HUD asked us to comment on whether we thought the factors identified in this report would be capable of comparing conditions in individual cities in 2005 and 2010. As noted above, one must always check to see if the dimensions of need are the same in both the base year and comparison year. To do this, one applies factor analysis to the needs indicators in both the base year and the comparison and compares the results. The factor analysis and the needs indicators should pass the Kaiser test for factor suitability in both years, and the test used to choose the number of factors should result in the same number of factors in both years. In addition, the factor loading should be similar in both years. To our knowledge, there is no statistical test to determine whether the loadings are similar, so this will be a judgment call. There are three reasons why HUD may not be able to apply the factor analysis developed for 2005 in this report to data from 2010. First, the underlying relationship between the needs indicators may change between 2005 and 2010. In our opinion, a change in the underlying relationships in only 5 years would be unusual but this is an empirical question. If there is a change in the underlying relationships, then the only solution is to recognize the break in pattern and try to interpret how needs have changed by comparing the results from using factors derived from both years. In other words, one would develop a set of factors using data from 2005 on needs indicators, apply the standardized scoring coefficients to standardized data from 2005 and 2010, and record how conditions changed. Then, one would repeat the steps using factors based on data from 2010. For Page 82
each city, one would compare the answers from the two sets of factor analyses and try to obtain a reasonable estimate of how conditions have changed in that city. In our view, the other two reasons for a break in factor patterns are more likely. First, as the ACS releases data on smaller cities, HUD will be applying the factor analysis to more cities and to urban counties. The addition of more observations may result in a change in factor pattern. Second, we believe that HUD should attempt to expand the needs indicators to provide more information on economic conditions and to cover needs related to education and public health. The addition of new indicators could also result in a change in factor pattern. In both cases, there is a simple solution. One would apply the factor pattern based on more observation or more needs indicators and derived in 2010 to both 2005 and 2010. 6.4. Factor Analysis Involving Different Geographies In 2008, the Census Bureau will release ACS data based on 3-year moving averages for counties and places with populations of 20,000 or more. The 2008 release will cover data collected in 2005, 2006, and 2007. The availability of new data raises three issues with respect to the factor analysis developed in this research: • Will HUD be able to use the ACS data to construct urban counties and will the addition of urban counties change the factor pattern? • Will the addition of c ities with populations between 20,000 and 65,000 change the factor pattern? • Should non-urban counties be included in the same factor analysis as cities and urban counties? A related issue is whether to use 3-year moving average data for all places, even those where annual data are available. The answers to all of these questions are inherently empirical. Researchers will have to wait for the 2008 ACS data and see how they change the analysis. The work in this project is most relevant to the issue of whether the factor patterns will change with the additions of observations on cities with populations between 20,000 and 65,000. Section 3.2.3 discusses the results we obtained when we split the sample into cities with 200,000 or more residents and cities with less than 200,000 residents and ran factor analysis on the two groups separately. (Table A.5 in Appendix A reports the results for the rotated factors.) The most important difference between the two factor analyses involved the Eigenvalue test that we used to determine the number of factors. The test identified four factors for the cities with 200,000 or more residents and only three factors for cities with less than 200,000 residents. There were some other noteworthy differences that are discussed in Section 3.2.3. Despite these differences, we applied factor analysis Page 83
to the combined database that includes both large and small cities. In our opinion, the fourth factor added little to the analysis, and the other differences were minor. Our greatest concern about combining large and small cities is that the test we applied in Section 3.2.3 uses a boundary point, a population of 200,000, which does not adequately distinguish large cities from small cities. The problem is that there are too few “large” cities to run a separate factor analysis with a boundary point much higher than 200,000. This problem will not lessen when the Census Bureau releases data on smaller cities. It may be that adding more cities to the under-200,000 group will result in a sharper difference in factor patterns between the two groups. However, we do not think this is the likely outcome. The factor pattern for the under-200,000 group is very close to the factor pattern for the entire group of cities. This leads us to believe that adding smaller cities will not greatly change the factor pattern. With respect to the other questions, we offer our opinions below, with the proviso that we think reliable answers are empirical and must await the 2008 data. • With respect to the urban counties issues, we believe that the 3-year moving average data will allow HUD to construct reliable needs indicators for urban counties and that adding urban counties to the factor analysis will not change factor pattern greatly. (In this discussion, we use the HUD definition of urban counties.) • With respect to non-urban counties, we believe that factor analysis applied separately to non-urban counties and to cities and urban counties combined would result in different factor patterns. At a minimum, combining the two groups would require eliminating important needs indicators, including MEDINCCBS2CITY, POVCON, and MINCON, because they are defined in the context of metropolitan areas. These needs indicators figured heavily in the interpretation of Factor 1. • With respect to using 3-year moving average data or annual data, we favor consistency—that is, using the data defined for the same period for all places. At the city and urban county level, we think 3-year average data are current enough to represent present needs. One possible problem is that the Census Bureau may not release some 3-year average data for smaller cities, just as it has not released annual data on minority populations and overcrowded housing for some cities with populations greater than 65,000. In this case, we favor using 5-year average data for the missing variables and relying on 3-year average data for most of the needs indicators. 6.5. Measuring Progress at the Tract Level HUD indicated early in the project that it was interested in the lessons that could be drawn from this research that are applicable to measuring needs at the tract level. The Page 84
Page 85 Administration has proposed creating a special fund within the CDBG program to award communities for making progress in reducing neighborhood distress. Such a proposal would require a community-needs measure at the neighborhood level. Since ACS data will be available at the census-tract level beginning in 2010, it was hoped that the experience gained here in constructing a city-level index using ACS data would be useful to HUD in developing a neighborhood-level index. In trying to apply our experience to this issue, we first considered the applicability at the tract level of the 27 needs indicators in Table 1. We think 7 or 8 of the 27 would not be applicable at the tract level. The two crime variables are not available because the FBI does not collect crime data at the tract level. The three long-run decline variables, EXCSINFRA, CHNGEMPLOYBASE, and CHGLOWINCCON, are relevant only in the city context. Finally, there are three variables linked to poverty rate of the tract; these are PCTPOPHIGHPOVNGHS, PCTPOPMODPOVNGHS, and PCTVACMODPOVCITY. The first two of these variables could not be integrated into a tract-level factor analysis because they become binary variables (100 percent or 0 percent) at the tract level. Conceptually, one could incorporate PCTVACMODPOVCITY into a tract-level factor analysis because this variable takes on more than two values. The other 19 needs indicators could be computed at the tract level using 5-year moving average data from the ACS. 44 Because of the substantial change in the number and type of indicators, a new factor analysis would have to be performed at the tract level. This factor analysis is likely to identify different dimensions of need than the three identified at the city level in this report. Finding appropriate weights for the factors will continue to be a problem if a single-value index of community needs is required. There are important data and conceptual considerations in measuring progress at the tract level. First, the ACS has a lower sampling rate over 5 years than the long-form survey in the decennial censuses. For this reason, the Census Bureau is more likely to suppress data used for some of the needs indicators at the tract level. Second, measurement error will be relatively high so that year-to-year changes on individual indicators may not reflect actual changes. The use of moving averages will dampen the effect of an unusual sample in any one year, but the dampened effect will persist for 5 years. Third, the use of 5-year moving average data probably determines the time frame to be used to measure progress. One is likely to want to compare conditions at 5-year intervals so that none of the samples overlap in the before and after comparison. On the conceptual side, a clear distinction needs to be made, if one plans to reward cities for alleviating needs, between measuring a change in needs and measuring how local government actions have reduced community needs. Even at the city level, needs can change independent of the efforts of the city to lessen needs. In the context of the factor analysis used in this report, a strong national economy can lower the values of important 44 The three dissimilarity indicators—MEDINCCBS2CITY, POVCON, and MINCON—could be computed at the tract level but may have a different meaning at the tract level. We included these variables as indicators of complicating conditions at the city level.
Page 86 needs indicators, such as the unemployment rate, the overall poverty rate, the pro portion of poor children and school-age children, and even the disparity in incomes between cities and metropolitan areas. The same consequences of a strong national economy apply to measured needs at the tract level. Conceptually, one would like to control for outside influences so that cities would not benefit from favorable external conditions or suffer from unfavorable external conditions. One possible way of doing this would be to measure progress against some national average. At the tract level, lower values for the needs indicators may or may not correspond to what HUD would consider improved conditions. Consider two different scenarios that result in a lower poverty rate for a single tract. In the first scenario, the city provides training to low-skilled workers, provides day care for working families, and encourages small business startups in the tract. These changes result in lower unemployment and higher incomes. In the second scenario, the city undertakes rigorous code enforcement resulting in demolition of low-rent structures, increases police protection, and improves neighborhood schools. These changes result in gentrification and, therefore, a lower poverty rate for the tract. The values of many of the needs indicators that can be computed at the tract level are sensitive to movement of households into and out of a tract. 45 The ACS tables on migration offer limited help in controlling for the impact of gentrification. The ACS provides tables on age, race, household type, education attainment, individual income, and poverty status separately for persons who lived in the same house one year ago and persons who moved into the house within the past year. All of these considerations suggest that, despite some useful insights gained from factor analysis at the city level, HUD will need to do a lot of conceptual and empirical work to develop a measure capable of implementing the Administration’s proposal. 6.6. Measuring Fiscal Capacity The report developed and implemented a measure of real fiscal capacity. This is a substantial advance in understanding the relative need for federal assistance among cities. In our opinion, capacity should be considered along with needs in determining how much federal assistance cities should receive. With respect to the measure of real fiscal capacity, more work needs to be done in two areas. First, because of data limitations, we were able to estimate relative real fiscal capacity for only 266 of our 473 cities. The data failure that had the largest impact on the analysis was the absence of data from the 2002 Economic Census for a large number of cities. We used the economic census data to estimate sales tax capacity and business tax 45 As noted earlier in the discussion of this topic, 19 of the 27 need indicators can be computed at the tract level. Fifteen of these depend on the characteristics of persons residing in the tract, and therefore are affected by movement of households into and out of the tract. Examples of these are: POORPERS, POORCHILD, LWINCHHDS, SGLPRNTFAM, RCNTIMMIG, and UNEDUCADULTS.
capacity. There are three possible strategies for increasing the number of cities for which real fiscal capacity can be computed. First, HUD could work with the Census Bureau to obtain data on more places from the economic census. Perhaps, HUD could construct the index at the Census Bureau using data not released to the public. Second, HUD could try to find alternative ways to estimate sales tax capacity and business tax capacity. Third, HUD could use the data from the 266 cities to develop a model of real fiscal capacity applicable to all cities. We think the modeling approach is imminently doable. The correlation between per capita income and fiscal capacity was 0.78, indicating the per capita income explains half of the variation in fiscal capacity. The additio n of other variables such as region, city size, principal city/suburb status, and proportion of working-age population should produce a model that fits the data well. Second, more thought needs to be given to how to combine an index of real fiscal capacity with an index of community needs. The approach we developed provides a consistent linking of the two indices and works reasonably well. Its main weakness is the assumption that the extent to which a city is below average on real fiscal capacity (measured in standard deviations) indicates the same level of need for federal assistance as an equal distance above average on community needs. 6.7. Changing City Boundaries and Cost-of-Living Differences In the Statement of Work, HUD asked us to consider the implications for an index of community needs of changing city boundaries and cost-of-living differences between cities. After selecting the 27 needs indicators in Table 1, we considered whether changes in city boundaries could affect the relevance or interpretation of any of the indicators. This question is particularly relevant to this research because an explicit goal was to measure, city-by-city, changes in community needs between 2000 and 2005. The indicators are defined so that they record conditions in a city using the boundaries in effect on the date the data were collected. Some indicators, such as the change in the concentration of low income families (CHGLOWINCCON), compare conditions at two points in time—for example, 1970 and 2005. These multiyear indicators are defined using the city boundaries in effect at each point in time. If a principal city annexed a high-income suburb in 2003, our changing-boundaries definition would result in the city having a lower measured poverty rate in 2005 than in 2000 even if it had the same number of poor persons in both years. In the case of the poverty rate variable (POORPERS), this result is completely consistent with the role of that need indicator in the analysis. We looked at each of the needs indicators in this way; in every case, we concluded that any change in city boundary between 2000 and 2005 should be incorporated into the measurement of that indicator. Page 87
Page 88 This conclusion applies to the three multiyear variables—EXCSINFRA, CHNGEMPLOYBASE, and CHGLOWINCCON—as well. However, we acknowledge that boundary changes could confound EXCSINFRA, but we believe that this possibility is remote. EXCSINFRA, as calculated in 2005, compares the maximum number of households recorded in 1970, 1980, 1990, 2000, and 2005 to the number of households in the city in 2005. If a city is steadily growing, then the maximum number of households would have occurred in 2005 and EXCSINFRA = 1. If a city is declining, then the maximum number of households would have occurred earlier than 2005 and EXCSINFRA > 1. The possibility exists that the original part of a city declined between 1970 and 2005, but the city annexed surrounding areas so that the number of households stayed constant or increased. In such a case, EXCSINFRA = 1, but the city would have excess infrastructure in the original part. We think that this possibility is remote and that, if a case like this did exist, holding original boundaries constant could produce an undesirable result as well. 46 Cost-of-living differences enter into the analysis of community needs in two ways. First, the poverty rate (POORPERS) and related variables are based on counts of persons living in households with incomes below poverty-level incomes. The assumption is that these persons have needs that cities have to provide for either because they have insufficient incomes or because they have other characteristics, such as an elderly age, that combines with insufficient income to create needs. Insufficient income is a function of both income level and the cost of necessities. Being below the national poverty lev el in Wichita poses fewer problems for a family than being below the national poverty level in San Francisco because of the cost of housing and other necessities. 47 Need indicators based on cost-adjusted poverty levels would be desirable. 48 Cost-of-living differences also enter the analysis in terms of the cost of responding to needs. A poor elderly person may need services such as meals on wheels, adult day care, and visiting nurse care. The costs of providing such services will differ across cities. We believe that the real fiscal capacity index adjusts for these differences directly and, therefore, no corresponding adjustments need to be made in the needs indicators used in constructing measures of community needs. 6.8. Areas for Future Work In our opinion, the most important area for future work is to expand and improve upon the list of needs indicators. We believe that the 27 variables in Table 1 are a well- conceived, broad-based, and carefully defined set of needs indicators. We believe these 46 For example, if the original part of the city was small in population relative to its size in 2005, then the decline in households measured using constant boundaries would be out of proportion to the problem. 47 The Census Bureau bases its poverty counts in the 48 contiguous states on the same income levels by household size; it uses higher income levels in Alaska and Hawaii. 48 The Government Accountability Office (GAO) has asked the Census Bureau to produce counts of poor persons using a poverty level adjusted for cost-of-living differences. The GAO used HUD fair market rents to adjust the poverty level. HUD should experiment with the special counts prepared for GAO.
indicators provided the basis for a useful factor analysis, and we would recommend the inclusion of all of these indicators in future work. However, we think the greatest payoff for understanding community needs is to improve these indicators and fill in some missing gaps. Here are our suggestions: • Carry out further analysis of the crime variables to determine whether there were any data errors, on our part or the part of the FBI, which caused the low correlations between the part 1 arrest data and the violent crime statistics. Since these data have many missing observations, the gains from solving this mystery may be limited. Future work may avoid crime statistics to keep from losing cities because of missing data. • Test whether there are any significant changes in the factor analysis results if one substitutes for POORPERS the counts that the Census Bureau is preparing for GAO using fair market rent (FMR) data to adjust for cost-of-living differences. • Test whether the employment disability counts can be used in the future. • Test whether the “poor housing appreciation in high poverty neighborhood” indicator based on HMDA data provides useful information. We agreed that this was a reasonable variable in concept but were not able to compute it within the resources available. • Explore with the Census Bureau the possibility of obtaining better information from the economic censuses to gauge the change in economic conditions in cities. In our opinion, this may be the single most important improvement to the set of needs indicators. It may be necessary to conduct some of the analysis at the Census Bureau to take advantage of data not released to the public. However, it is possible that HUD could obtain public release of useful data. For example, the Census Bureau does not release the count of total jobs in a city. HUD may need to know only the change in the number of total jobs. The Census Bureau may be willing to release the change in jobs but not the total number of jobs. • Investigate whether there are other useful measures of long-term trends that could be constructed. • Investigat e whether HUD could work with the Department of Education to construct useful measures of education needs defined at the city level, such as the number of children who are not fluent in English. • Investigate whether HUD could compile from HHS useful information at the city level on public health needs and conditions such as infant mortality rates. Page 89
Page 90 While we believe better data would produce the best payoff for measuring community needs, we also recommend some future conceptual work in the following areas: • Further testing of the regression approach. • Further analysis of the relationship between the regression approach and factor analysis. We believe that looking at the relationship between principal components analysis and the regression approach may be a useful way to proceed in this area. • Development of a model to estimate real fiscal capacity for the cities for which the needed data are not available. • Further consideration of how to combine a measure of real fiscal capacity with a measure of community needs.
Appendix A—Supplemental Tables Page A-1
Table A.1. Correlation Matrix for Needs Indicators PO P CHGLOWI NC CON CHN GEMPLB ASE DENIAL EXCSINFRA LACK AFF DR ENTALS LING ISO L LWINCHH DS MEDINCCBS 2 CITY MINCON OVERCR OW D_2000 PCTPO P HIGH PO VNG H S PCTPO P M O D PO VNG H S PCTVACM OD PO VCITY POP 1.00 0.03 -0.06 0.07 0.03 0.08 0.11 0.12 0.10 0.06 0.10 0.17 0.14 0.06 CHGLOWINCCON 0.03 1.00 0.15 0.44 0.26 0.28 0.24 0.42 0.47 0.34 0.18 0.12 0.29 0.24 CHNGEMPLBASE -0.06 0.15 1.00 0.08 -0.06 0.00 0.10 0.02 -0.01 -0.05 0.12 -0.01 0.04 0.01 DENIAL 0.07 0.44 0.08 1.00 0.52 0.28 -0.01 0.58 0.62 0.65 -0.03 0.51 0.60 0.76 EXCSINFRA 0.03 0.26 -0.06 0.52 1.00 0.21 -0.06 0.41 0.44 0.44 -0.07 0.33 0.38 0.64 LACKAFFDRENTALS 0.08 0.28 0.00 0.28 0.21 1.00 0.57 0.60 0.66 0.35 0.58 0.34 0.54 0.15 LINGISOL 0.11 0.24 0.10 -0.01 -0.06 0.57 1.00 0.32 0.30 0.09 0.77 0.10 0.34 -0.17 LWINCHHDS 0.12 0.42 0.02 0.58 0.41 0.60 0.32 1.00 0.90 0.70 0.29 0.51 0.74 0.52 MEDINCCBS2CITY 0.10 0.47 -0.01 0.62 0.44 0.66 0.30 0.90 1.00 0.73 0.23 0.46 0.71 0.52 MINCON 0.06 0.34 -0.05 0.65 0.44 0.35 0.09 0.70 0.73 1.00 0.00 0.45 0.57 0.61 OVERCROWD_2000 0.10 0.18 0.12 -0.03 -0.07 0.58 0.77 0.29 0.23 0.00 1.00 0.11 0.39 -0.15 PCTPOPHIGHPOVNGHS 0.17 0.12 -0.01 0.51 0.33 0.34 0.10 0.51 0.46 0.45 0.11 1.00 0.52 0.52 PCTPOPMODPOVNGHS 0.14 0.29 0.04 0.60 0.38 0.54 0.34 0.74 0.71 0.57 0.39 0.52 1.00 0.63 PCTVACMODPOVCITY 0.06 0.24 0.01 0.76 0.64 0.15 -0.17 0.52 0.52 0.61 -0.15 0.52 0.63 1.00 POORCHILD 0.12 0.36 0.05 0.71 0.39 0.45 0.14 0.80 0.72 0.68 0.14 0.62 0.79 0.69 POOROVER74 0.13 0.13 0.02 0.38 0.20 0.35 0.29 0.42 0.42 0.35 0.19 0.37 0.49 0.31 POORPERS 0.12 0.38 0.07 0.73 0.42 0.51 0.22 0.84 0.76 0.67 0.22 0.68 0.83 0.68 POVCON 0.11 0.34 -0.03 0.56 0.43 0.45 0.13 0.88 0.86 0.79 0.06 0.47 0.66 0.58 PR70RENTPOV 0.17 0.32 -0.05 0.59 0.51 0.49 0.20 0.82 0.80 0.71 0.16 0.57 0.77 0.63 VIOLCRIME 0.05 0.06 0.03 0.51 0.22 0.05 -0.16 0.42 0.36 0.39 -0.15 0.39 0.44 0.53 PT1CRIME -0.02 -0.04 0.06 0.04 -0.03 -0.03 -0.08 0.03 0.05 0.06 -0.05 0.08 0.02 0.09 PT2CRIME -0.03 -0.02 0.04 0.20 0.03 0.01 -0.12 0.12 0.20 0.20 -0.15 0.18 0.13 0.21 RCNTIMMIG 0.09 0.24 0.04 -0.17 -0.07 0.44 0.81 0.21 0.15 -0.02 0.60 -0.09 0.12 -0.27 SCHPOPPOOR 0.13 0.35 0.02 0.67 0.38 0.47 0.15 0.78 0.69 0.66 0.15 0.62 0.77 0.66 SGLPRNTFAM 0.07 0.35 0.01 0.68 0.44 0.47 0.02 0.74 0.76 0.69 0.05 0.55 0.69 0.64 UNDEREDWORKAGE 0.00 0.29 0.18 0.57 0.20 0.49 0.27 0.45 0.54 0.33 0.37 0.25 0.54 0.31 UNEDUCADULTS 0.10 0.35 0.13 0.48 0.20 0.66 0.65 0.65 0.64 0.41 0.69 0.39 0.71 0.28 UNEMPCEN 0.08 0.34 0.02 0.67 0.46 0.36 0.12 0.63 0.60 0.53 0.15 0.48 0.59 0.58 Page A-2
Page A-3 Table A.1. Correlation Matrix for Needs Indicators (continued) POORCHILD POOROVER74 POORPERS POVCON PR70RENTPOV VIOLCRIME PT1CRIME PT2CRIME RCNTIMMIG SCHPOPPOOR SGLPRNTFAM UND ERE DWOR KAGE UNEDU CAD U L TS UNEMPCEN POP 0.12 0.13 0.12 0.11 0.17 0.05 -0.02 -0.03 0.09 0.13 0.07 0.00 0.10 0.08 CHGLOWINCCON 0.36 0.13 0.38 0.34 0.32 0.06 -0.04 -0.02 0.24 0.35 0.35 0.29 0.35 0.34 CHNGEMPLBASE 0.05 0.02 0.07 -0.03 -0.05 0.03 0.06 0.04 0.04 0.02 0.01 0.18 0.13 0.02 DENIAL 0.71 0.38 0.73 0.56 0.59 0.51 0.04 0.20 -0.17 0.67 0.68 0.57 0.48 0.67 EXCSINFRA 0.39 0.20 0.42 0.43 0.51 0.22 -0.03 0.03 -0.07 0.38 0.44 0.20 0.20 0.46 LACKAFFDRENTALS 0.45 0.35 0.51 0.45 0.49 0.05 -0.03 0.01 0.44 0.47 0.47 0.49 0.66 0.36 LINGISOL 0.14 0.29 0.22 0.13 0.20 -0.16 -0.08 -0.12 0.81 0.15 0.02 0.27 0.65 0.12 LWINCHHDS 0.80 0.42 0.84 0.88 0.82 0.42 0.03 0.12 0.21 0.78 0.74 0.45 0.65 0.63 MEDINCCBS2CITY 0.72 0.42 0.76 0.86 0.80 0.36 0.05 0.20 0.15 0.69 0.76 0.54 0.64 0.60 MINCON 0.68 0.35 0.67 0.79 0.71 0.39 0.06 0.20 -0.02 0.66 0.69 0.33 0.41 0.53 OVERCROWD_2000 0.14 0.19 0.22 0.06 0.16 -0.15 -0.05 -0.15 0.60 0.15 0.05 0.37 0.69 0.15 PCTPOPHIGHPOVNGHS 0.62 0.37 0.68 0.47 0.57 0.39 0.08 0.18 -0.09 0.62 0.55 0.25 0.39 0.48 PCTPOPMODPOVNGHS 0.79 0.49 0.83 0.66 0.77 0.44 0.02 0.13 0.12 0.77 0.69 0.54 0.71 0.59 PCTVACMODPOVCITY 0.69 0.31 0.68 0.58 0.63 0.53 0.09 0.21 -0.27 0.66 0.64 0.31 0.28 0.58 POORCHILD 1.00 0.41 0.96 0.80 0.76 0.56 0.05 0.22 -0.05 0.98 0.79 0.49 0.57 0.65 POOROVER74 0.41 1.00 0.49 0.37 0.46 0.23 0.06 0.16 0.19 0.42 0.39 0.25 0.42 0.34 POORPERS 0.96 0.49 1.00 0.80 0.78 0.55 0.06 0.22 0.03 0.94 0.80 0.52 0.64 0.68 POVCON 0.80 0.37 0.80 1.00 0.82 0.47 0.07 0.23 0.04 0.77 0.74 0.32 0.45 0.55 PR70RENTPOV 0.76 0.46 0.78 0.82 1.00 0.40 0.03 0.17 0.06 0.73 0.73 0.30 0.51 0.59 VIOLCRIME 0.56 0.23 0.55 0.47 0.40 1.00 0.20 0.23 -0.24 0.53 0.50 0.27 0.19 0.40 PT1CRIME 0.05 0.06 0.06 0.07 0.03 0.20 1.00 0.53 -0.06 0.04 0.08 0.01 0.00 0.06 PT2CRIME 0.22 0.16 0.22 0.23 0.17 0.23 0.53 1.00 -0.19 0.19 0.27 0.09 0.04 0.17 RCNTIMMIG -0.05 0.19 0.03 0.04 0.06 -0.24 -0.06 -0.19 1.00 -0.01 -0.12 -0.02 0.36 0.03 SCHPOPPOOR 0.98 0.42 0.94 0.77 0.73 0.53 0.04 0.19 -0.01 1.00 0.76 0.45 0.55 0.63 SGLPRNTFAM 0.79 0.39 0.80 0.74 0.73 0.50 0.08 0.27 -0.12 0.76 1.00 0.49 0.49 0.63 UNDEREDWORKAGE 0.49 0.25 0.52 0.32 0.30 0.27 0.01 0.09 -0.02 0.45 0.49 1.00 0.74 0.50 UNEDUCADULTS 0.57 0.42 0.64 0.45 0.51 0.19 0.00 0.04 0.36 0.55 0.49 0.74 1.00 0.51 UNEMPCEN 0.65 0.34 0.68 0.55 0.59 0.40 0.06 0.17 0.03 0.63 0.63 0.50 0.51 1.00
Table A.2. Initial Factor Analysis, 2005 Data: Factor Loading for Unrotated Factors Factor1 Factor2 Factor3 CHGLOWINCCON 0.42991 0.16552 0.07619 CHNGEMPLBASE -0.10283 0.116 0.27252 DENIAL 0.7626 -0.23761 0.2932 EXCSINFRA 0.56431 -0.22377 -0.11208 LACKAFFDRENTALS 0.60586 0.55234 -0.07049 LINGISOL 0.21712 0.89532 -0.09646 LWINCHHDS 0.86956 0.11171 -0.25189 MEDINCCBS2CITY 0.90339 0.07241 -0.12628 MINCON 0.75356 -0.16559 -0.14176 OVERCROWD2000 0.17479 0.87566 0.05896 PCTPOPHIGHPOVNGHS 0.67931 -0.11091 -0.0474 PCTPOPMODPOVNGHS 0.84469 0.10956 0.08503 PCTVACMODPOVCITY 0.73709 -0.41366 0.05368 POORCHILD 0.91385 -0.1337 0.01959 POOROVER74 0.45471 0.0221 0.00709 POORPERS 0.94876 -0.07822 0.03881 POVCON 0.8429 -0.1709 -0.31292 PR70RENTPOV 0.8403 -0.11492 -0.28779 PT1CRIME 0.02003 -0.15022 0.09945 PT2CRIME 0.15692 -0.28168 0.11429 RCNTIMMIG 0.00871 0.82496 -0.32501 SCHPOPPOOR 0.89979 -0.11129 -0.00439 SGLPRNTFAM 0.85943 -0.1672 0.05814 UNDEREDWORKAGE 0.54942 0.27346 0.62907 UNEDUCADULTS 0.68537 0.57306 0.29624 UNEMPCEN 0.68953 -0.02331 0.1231 Page A-4
Page A-5 Table A.3. Initial Factor Analysis, 2005 Data: Factor Loading for Varimax Orthogonal Rotated Factors Factor1 Factor2 Factor3 CHGLOWINCCON 0.39005 0.15917 0.20137 CHNGEMPLBASE -0.15496 0.00839 0.27243 DENIAL 0.73597 -0.26953 0.33119 EXCSINFRA 0.59824 -0.13193 -0.07606 LACKAFFDRENTALS 0.53521 0.58412 0.2224 LINGISOL 0.11422 0.88724 0.24041 LWINCHHDS 0.87577 0.25161 -0.04226 MEDINCCBS2CITY 0.89575 0.17422 0.06764 MINCON 0.78093 -0.05402 -0.0509 OVERCROWD2000 0.05267 0.81275 0.37078 PCTPOPHIGHPOVNGHS 0.68753 -0.04021 0.04124 PCTPOPMODPOVNGHS 0.8029 0.13272 0.26546 PCTVACMODPOVCITY 0.76776 -0.35445 0.04696 POORCHILD 0.91089 -0.06806 0.13793 POOROVER74 0.44246 0.05011 0.09498 POORPERS 0.93538 -0.02019 0.18005 POVCON 0.89404 0.00574 -0.19561 PR70RENTPOV 0.8808 0.04943 -0.15453 PT1CRIME 0.02421 -0.17337 0.04706 PT2CRIME 0.17296 -0.29207 0.04254 RCNTIMMIG -0.04839 0.88482 -0.03193 SCHPOPPOOR 0.89774 -0.03985 0.12043 SGLPRNTFAM 0.85613 -0.11644 0.15311 UNDEREDWORKAGE 0.41377 0.07976 0.77123 UNEDUCADULTS 0.55759 0.48379 0.58389 UNEMPCEN 0.66185 -0.01566 0.22989
Table A.4. Factor Analysis Using 2000 Data: Unrotated and Rotated Factor Loading Unrotated Factors Orthogonal Rotated Factors Factor1 Factor2 Factor3 Factor1 Factor2 Factor3 CHGLOWINCCON_2000 0.49109 0.22375 0.1822 0.38067 0.3108 0.28796 CHNGEMPLBASE -0.09228 0.08752 0.28443 -0.14108 -0.00066 0.27779 DENIAL 0.70361 -0.32571 0.23506 0.74345 -0.15634 0.28151 EXCSINFRA_2000 0.5542 -0.25625 -0.07394 0.61016 -0.0709 -0.03083 LACKAFFDRENTALS_2000 0.5449 0.62309 -0.12124 0.33975 0.76166 0.0655 LINGISOL_2000 0.28082 0.89816 -0.15684 0.00947 0.95329 0.03615 LWINCHHDS_2000 0.85302 0.07528 -0.23753 0.8088 0.35672 -0.09126 MEDINCCBS2CITY_2000 0.89187 0.05764 -0.17376 0.84491 0.33822 -0.02614 MINCON_2000 0.74313 -0.22709 -0.21605 0.79404 0.03778 -0.13627 OVERCROWD2000 0.2676 0.87692 0.09703 -0.02082 0.87873 0.27822 PCTPOPHIGHPOVNGHS_2000 0.68746 -0.21556 -0.03037 0.72008 -0.00419 0.0383 PCTPOPMODPOVNGHS_2000 0.88635 0.09892 0.13002 0.79826 0.31452 0.27599 PCTVACMODPOVCITY_2000 0.69735 -0.47671 0.07489 0.79845 -0.26775 0.0997 POORCHILD_2000 0.95021 -0.07487 0.06862 0.91725 0.18149 0.19727 POOROVER74_200 0.71694 -0.09999 -0.08448 0.71821 0.12342 0.00887 POORPERS_2000 0.96514 0.02506 0.10894 0.89734 0.27141 0.25517 POVCON_2000 0.83176 -0.22906 -0.32327 0.88889 0.08221 -0.22784 PR70RENTPOV_2000 0.87912 -0.08413 -0.2715 0.88502 0.22114 -0.14654 PT1CRIME_2000 0.29437 -0.11921 0.03159 0.31223 -0.036 0.05548 PT2CRIME_2000 0.27073 -0.32528 -0.04544 0.35951 -0.22065 -0.05689 RCNTIMMIG_2000 0.05266 0.87531 -0.32979 -0.18353 0.90275 -0.17047 SCHPOPPOOR_2000 0.95062 -0.03845 0.05205 0.90821 0.21913 0.18714 SGLPRNTFAM_2000 0.90227 -0.23344 0.02561 0.92387 0.02783 0.1222 UNDEREDWORKAGE_2000 0.57969 0.24368 0.60105 0.41879 0.27028 0.71292 UNEDUCADULTS_2000 0.71862 0.55254 0.26904 0.48869 0.6657 0.46057 UNEMPCEN_2000 0.83467 0.06961 0.08606 0.76231 0.28137 0.22057 Page A-6
Table A.5. Factor Analysis Using 2005 Data, Rotated Factor Loadings for Large Cities and Small Cities Cities of 200,000 or More Cities of Less Than 200,000 Rotated Factor1 Factor2 Factor3 Factor4 Factor1 Factor2 Factor3 CHGLOWINCCON 0.32092 0.22554 0.09338 0.10486 0.25905 0.21509 0.10375 CHNGEMPLBASE -0.06154 0.13943 0.10658 - 0.06794 -0.11825 0.00499 0.08009 DENIAL 0.60478 -0.2097 0.32227 0.2362 0.53746 -0.14706 0.38518 EXCSINFRA 0.5062 -0.10094 0.12355 0.70666 0.36096 -0.06999 0.04443 LACKAFFDRENTAL S 0.60231 0.53932 0.35559 0.16933 0.37883 0.64542 0.29564 LINGISOL -0.04102 0.96264 0.04793 - 0.04789 0.0736 0.91129 0.06674 LWINCHHDS 0.87853 0.1837 0.03086 0.18513 0.82297 0.27014 0.08334 MEDINCCBS2CITY 0.83051 0.07524 0.16249 0.2488 0.73633 0.24373 0.21856 MINCON 0.62726 -0.15293 0.02349 0.46406 0.70944 0.02424 0.03982 OVERCROWD2000 0.02875 0.89844 0.23333 - 0.05285 0.00614 0.8519 0.28414 PCTPOPHIGHPOVN GHS 0.77314 0.07007 0.18185 0.03701 0.57861 0.01207 0.13945 PCTPOPMODPOVN GHS 0.85335 0.22304 0.30456 0.1798 0.72246 0.22209 0.32766 PCTVACMODPOVCI TY 0.68697 -0.3124 0.13567 0.42289 0.62676 -0.23827 0.14662 POORCHILD 0.9602 -0.05175 0.1048 - 0.01914 0.93392 0.0226 0.22342 POOROVER74 0.56129 0.14416 0.08524 0.20817 0.31713 0.09699 0.03095 POORPERS 0.93887 0.04466 0.20104 0.09484 0.91952 0.07197 0.24322 POVCON 0.86501 -0.07434 -0.07634 0.19655 0.89469 0.02452 -0.05989 PR70RENTPOV 0.82284 0.05715 0.05072 0.37423 0.80899 0.08035 -0.03507 PT1CRIME 0.07619 -0.00352 0.00198 0.06547 -0.01406 -0.04468 -0.02989 PT2CRIME 0.05836 -0.2747 0.01233 0.00065 0.16 651 -0.13149 0.03528 RCNTIMMIG -0.16047 0.89867 -0.22456 - 0.06925 -0.0769 0.87511 -0.2177 SCHPOPPOOR 0.96533 -0.03595 0.05385 - 0.00219 0.90835 0.05846 0.20762 SGLPRNTFAM 0.82503 -0.13778 0.34831 0.14783 0.75812 -0.02687 0.29724 UNDEREDWORKAG E 0.28883 0.15058 0.85108 0.10433 0.30405 0.19232 0.7888 UNEDUCADULTS 0.45417 0.67608 0.51839 0.01181 0.442 0.5655 0.56173 UNEMPCEN 0.68525 -0.06257 0.21859 0.43727 0.46274 0.09936 0.34128 Page A-7
Table A.6. 2005 Factor Analysis Using VIOLCRIME Instead of PT1CRIME, VARIMAX Rotated Factors Factor1 Factor2 Factor3 POORPERS 0.94261 LINGISOL 0.91454 UNEMPCEN 0.19566 POORCHILD 0.91804 RCNTIMMIG 0.87042 UNEDUCADULTS 0.49499 SCHPOPPOOR 0.90272 OVERCROWD_2000 0.85383 UNDEREDWORKAGE 0.72952 MEDINCCBS2CITY 0.88857 LACKAFFDRENTALS 0.6461 SGLPRNTFAM 0.12659 POVCON 0.88566 UNEDUCADULTS 0.58219 SCHPOPPOOR 0.10997 PR70RENTPOV 0.87071 LWINCHHDS 0.28366 RCNTIMMIG -0.13826 SGLPRNTFAM 0.86702 MEDINCCBS2CITY 0.23882 PT2CRIME 0.01392 LWINCHHDS 0.85881 CHGLOWINCCON 0.22623 PR70RENTPOV -0.20693 PCTPOPMODPOVNGHS 0.80524 UNDEREDWORKAGE 0.20319 POVCON -0.23032 PCTVACMODPOVCITY 0.78866 PCTPOPMODPOVNGHS 0.18899 POORPERS 0.1599 MINCON 0.7835 POOROVER74 0.07432 POOROVER74 0.07732 DENIAL 0.76327 PR70RENTPOV 0.07252 POORCHILD 0.13235 PCTPOPHIGHPOVNGHS 0.69193 UNEMPCEN 0.04013 PCTVACMODPOVCITY 0.05213 UNEMPCEN 0.6672 POORPERS 0.0348 PCTPOPMODPOVNGHS 0.23272 EXCSINFRA 0.59869 CHNGEMPLBASE 0.03249 PCTPOPHIGHPOVNGHS 0.02412 VIOLCRIME 0.55578 POVCON 0.01955 OVERCROWD_2000 0.27431 UNEDUCADULTS 0.54919 SCHPOPPOOR 0.0021 MINCON -0.09542 LACKAFFDRENTALS 0.50896 PCTPOPHIGHPOVNGHS -0.00745 MEDINCCBS2CITY -0.01553 UNDEREDWORKAGE 0.43965 MINCON -0.01292 LWINCHHDS -0.10573 POOROVER74 0.4368 POORCHILD -0.02401 LINGISOL 0.12639 CHGLOWINCCON 0.38837 SGLPRNTFAM -0.05827 LACKAFFDRENTALS 0.11336 PT2CRIME 0.18253 EXCSINFRA -0.08259 EXCSINFRA -0.12857 LINGISOL 0.06766 DENIAL -0.18631 DENIAL 0.31377 OVERCROWD_2000 0.0219 PT2CRIME -0.23332 VIOLCRIME 0.13291 RCNTIMMIG -0.09868 PCTVACMODPOVCITY -0.31302 CHNGEMPLBASE 0.29818 CHNGEMPLBASE -0.16497 VIOLCRIME -0.35229 CHGLOWINCCON 0.14697 Page A-8
Table A.7. Standardized Scoring Coefficients, Based on 2005 Data Without Crime Variables Factor1 Factor2 Factor3 CHGLOWINCCON -0.02266 -0.00619 0.06911 CHNGEMPLBASE -0.00733 0.00465 0.05042 DENIAL 0.05521 -0.1065 0.2061 EXCSINFRA 0.04954 -0.02226 -0.00265 LACKAFFDRENTALS -0.01665 0.11404 -0.01053 LINGISOL -0.03114 0.36138 -0.06993 LWINCHHDS 0.02825 0.09883 -0.10354 MEDINCCBS2CITY 0.20368 0.10156 -0.01419 MINCON 0.04264 -0.02485 -0.00641 OVERCROWD_2000 -0.01755 0.16406 0.05127 PCTPOPHIGHPOVNGHS 0.01425 -0.01001 -0.042 PCTPOPMODPOVNGHS 0.00878 0.04533 0.02058 PCTVACMODPOVCITY 0.09204 -0.13162 0.03791 POORCHILD 0.0883 -0.1326 0.12027 POOROVER74 0.00719 -0.00089 -0.02458 POORPERS 0.18723 0.00824 0.2283 POVCON 0.09751 0.00589 -0.429 PR70RENTPOV 0.15118 0.00913 -0.24679 RCNTIMMIG -0.00757 0.24664 -0.1858 SCHPOPPOOR 0.12067 0.08762 -0.1873 SGLPRNTFAM 0.06006 -0.02148 0.01251 UNDEREDWORKAGE -0.05684 -0.04957 0.41739 UNEDUCADULTS -0.04105 0.13016 0.41726 UNEMPCEN 0.01966 -0.02896 0.04813 Page A-9
Table A.8. Alternative Index Scores City State Index 1: Equal Weight Index 2: Triple Weight to Poverty and Structural Problems Index 3: Triple Weight to Immigration and Housing Affordability Factor Index 4: Hedonic Weights Index 5: Richardson Weights Index 6: Partial Hedonic Weights Birmingham city Alabama 0.54 1.30 -0.12 0.43 1.81 1.19 Huntsville city Alabama -0.43 -0.28 -0.71 -0.30 -0.21 -0.36 Mobile city Alabama 0.17 0.52 -0.35 0.36 0.68 0.37 Montgomery city Alabama -0.07 0.14 -0.54 0.21 0.20 -0.01 Anchorage municipality Alaska -0.47 -0.52 -0.57 -0.31 -0.60 -0.58 Avondale city Arizona 0.42 0.28 0.52 0.44 0.19 0.30 Chandler city Arizona -0.46 -0.68 -0.39 -0.30 -0.85 -0.70 Glendale city Arizona 0.05 -0.10 -0.03 0.28 -0.25 -0.16 Mesa city Arizona -0.03 -0.24 0.00 0. 14 -0.42 -0.27 Peoria city Arizona -0.44 -0.73 -0.50 -0.10 -0.99 -0.81 Phoenix city Arizona 0.33 0.19 0.48 0.34 0.09 0.21 Scottsdale city Arizona -0.85 -0.95 -0.81 -0.79 -1.03 -0.95 Tempe city Arizona -0.44 -0.54 -0.28 -0.51 -0.59 -0.50 Tucson city Arizona 0.17 0.21 0.15 0.15 0.23 0.20 Yuma city Arizona 0.28 0.02 0.11 0.70 -0.25 -0.10 Fayetteville city Arkansas -0.71 -0.43 -0.69 -1.00 -0.18 -0.37 Fort Smith city Arkansas 0.13 0.18 -0.21 0.43 0.14 0.05 Little Rock city Arkansas -0.37 -0.16 -0.65 -0.31 -0.04 -0.23 Alameda city California -0.56 -0.53 -0.29 -0.85 -0.43 -0.41 Alhambra city California 0.37 0.06 1.08 -0.03 -0.03 0.29 Anaheim city California 0.62 0.11 1.02 0.74 -0.24 0.16 Antioch city California -0.03 -0.21 0.01 0.11 -0.36 -0.23 Bakersfield city California 0.21 0.06 0.12 0.46 -0.10 -0.01 Baldwin Park city California 1.32 0.82 1.73 1.39 0.50 0.89 Bellflower city California 0.64 0.43 0.81 0.68 0.29 0.46 Berkeley city California -1.04 -0.59 -0.62 -1.91 -0.09 -0.33 Buena Park city California 0.21 -0.25 0.54 0.35 -0.57 -0.21 Burbank city California -0.25 -0.49 0.03 -0.29 -0.62 -0.42 Page A-10
Page A-11 City State Index 1: Equal Weight Index 2: Triple Weight to Poverty and Structural Problems Index 3: Triple Weight to Immigration and Housing Affordability Factor Index 4: Hedonic Weights Index 5: Index 6: Partial Richardson Hedonic Weights Weights Carlsbad city California -0.91 -1.07 -0.77 -0.89 -1.17 -1.04 Carson city California 0.26 -0.14 0.35 0.56 -0.46 -0.18 Chico city California -0.35 -0.06 -0.43 -0.56 0.17 -0.04 Chino city California 0.18 -0.41 0.20 0.75 -0.91 -0.52 Chula Vista city California 0.17 -0.17 0.37 0.32 -0.42 -0.16 Clovis city California -0.28 -0.48 -0.34 -0.01 -0.68 -0.55 Concord city California -0.11 -0.32 0.18 -0.19 -0.44 -0.25 Corona city California -0.02 -0.45 0.06 0.33 -0.80 -0.50 Costa Mesa city California 0.15 -0.17 0.56 0.05 -0.35 -0.07 Daly City California 0.22 -0.14 0.84 -0.04 -0.29 0.04 El Cajon city California 0.28 0.14 0.44 0.26 0.06 0.18 Escondido city California 0.37 -0.05 0.68 0.47 -0.34 -0.01 Fairfield city California -0.11 -0.28 0.01 -0.08 -0.39 -0.26 Fontana city California 0.62 0.13 0.82 0.91 -0.24 0.12 Fremont city California -0.34 -0.64 0.30 -0.68 -0.73 -0.44 Fresno city California 0.63 0.58 0.66 0.65 0.54 0.58 Fullerton city California -0.16 -0.50 0.15 -0.14 -0.71 -0.44 Garden Grove city California 0.83 0.23 1.52 0.75 -0.12 0.39 Glendale city California 0.25 -0.20 0.99 -0.05 -0.41 0.00 Hawthorne city California 0.97 0.78 1.26 0.88 0.68 0.86 Hayward city California 0.26 0.02 0.56 0.19 -0.11 0.10 Hemet city California 0.57 0.24 0.58 0.89 -0.05 0.18 Hesperia city California 0.39 0.09 0.30 0.79 -0.20 -0.01 Huntington Beach city California -0.58 -0.81 -0.48 -0.46 -0.99 -0.82 Indio city California 1.09 0.73 1.52 1.01 0.52 0.83 Inglewood city California 0.96 0.77 1.17 0.94 0.65 0.81 Lakewood city California -0.27 -0.65 -0.18 0.02 -0.97 -0.69 Livermore city California -0.63 -0.80 -0.48 -0.61 -0.92 -0.78 Long Beach city California 0.45 0.42 0.73 0.20 0.46 0.53
Page A-12 City State Index 1: Equal Weight Index 2: Triple Weight to Poverty and Structural Problems Index 3: Triple Weight to Immigration and Housing Affordability Factor Index 4: Hedonic Weights Index 5: Index 6: Partial Richardson Hedonic Weights Weights Los Angeles city California 0.69 0.56 1.16 0.33 0.57 0.73 Merced city California 0.76 0.68 0.77 0.84 0.61 0.67 Mission Viejo city California -0.82 -1.10 -0.78 -0.58 -1.34 -1.15 Modesto city California 0.15 -0.07 0.07 0.45 -0.28 -0.14 Mountain View city California -0.60 -0.72 -0.02 -1.08 -0.67 -0.50 Napa city California -0.14 -0.31 0.11 -0.22 -0.39 -0.24 Norwalk city California 0.72 0.09 1.05 1.01 -0.37 0.10 Oakland city California 0.39 0.59 0.73 -0.14 0.84 0.76 Oceanside city California -0.18 -0.50 0.02 -0.06 -0.73 -0.49 Ontario city California 0.82 0.36 1.07 1.03 0.03 0.37 Orange city California -0.10 -0.52 0.09 0.14 -0 .84 -0.53 Oxnard city California 0.95 0.63 1.45 0.77 0.48 0.77 Palmdale city California 0.64 0.35 0.65 0.93 0.10 0.29 Pasadena city California -0.28 -0.25 -0.03 -0.55 -0.16 -0.14 Pleasanton city California -0.93 -1.05 -0.72 -1.03 -1.09 -0.98 Pomona city California 1.09 0.55 1.56 1.14 0.21 0.64 Redding city California -0.31 -0.39 -0.48 -0.06 -0.50 -0.47 Redlands city California -0.36 -0.42 -0.41 -0.26 -0.48 -0.45 Redwood City California 0.12 -0.03 0.61 -0.22 -0.04 0.14 Rialto city California 0.66 0.13 0.84 1.02 -0.29 0.09 Richmond city California 0.34 0.36 0.56 0.09 0.43 0.45 Riverside city California 0.23 -0.06 0.39 0.36 -0.27 -0.05 Roseville city California -0.58 -0.81 -0.53 -0.39 -0.99 -0.83 Sacramento city California 0.26 0.37 0.36 0.06 0.48 0.43 Salinas city California 1.32 0.78 1.82 1.37 0.42 0.87 San Bernardino city California 1.11 1.18 1.29 0.87 1.29 1.27 San Buenaventura (Ventura) city California -0.44 -0.44 -0.38 -0.50 -0.43 -0.42 San Diego city California -0.25 -0.27 0.05 -0.53 -0.21 -0.15
Page A-13 City State Index 1: Equal Weight Index 2: Triple Weight to Poverty and Structural Problems Index 3: Triple Weight to Immigration and Housing Affordability Factor Index 4: Hedonic Weights Index 5: Richardson Weights Index 6: Partial Hedonic Weights San Francisco city California -0.30 -0.29 0.20 -0.81 -0.15 -0.08 San Jose city California 0.02 -0.21 0.46 -0.19 -0.31 -0.08 San Leandro city California 0.08 -0.24 0.36 0.13 -0.46 -0.20 San Marcos city California -0.09 -0.43 0.08 0.08 -0.68 -0.43 San Mateo city California -0.39 -0.61 0.01 -0.56 -0.70 -0.49 Santa Ana city California 1.59 0.84 2.31 1.62 0.37 0.98 Santa Barbara city California -0.37 -0.40 -0.17 -0.54 -0.38 -0.33 Santa Clara city California -0.31 -0.44 0.31 -0.78 -0.41 -0.23 Santa Maria city California 0.93 0.62 1.37 0.82 0.45 0.73 Santa Monica city California -0.91 -0.95 -0.65 -1.13 -0.92 -0.85 Santa Rosa city California -0.16 -0.29 -0.06 -0.13 -0.38 -0.28 Simi Valley city California -0.54 -0.84 -0.54 -0.23 -1.11 -0.90 Stockton city California 0.61 0.52 0.70 0.61 0.46 0.53 Sunnyvale city California -0.40 -0.54 0.26 -0.94 -0.49 -0.30 Thousand Oaks city California -0.74 -0.87 -0.60 -0.74 -0.96 -0.84 Torrance city California -0.53 -0.78 -0.19 -0.62 -0.92 -0.70 Tracy city California -0.07 -0.50 0.03 0.27 -0.86 -0.55 Turlock city California 0.34 -0.13 0.49 0.66 -0.51 -0.17 Tustin city California -0.13 -0.52 0.38 -0.23 -0.74 -0.40 Union City California 0.29 -0.15 0.87 0.16 -0.40 -0.01 Upland city California -0.28 -0.29 -0.28 -0.28 -0.29 -0.29 Vacaville city California -0.41 -0.69 -0.45 -0.09 -0.94 -0.76 Vallejo city California -0.01 -0.08 0.07 -0.03 -0.12 -0.06 Victorville city California 0.52 0.29 0.43 0.84 0.07 0.21 Visalia city California 0.23 0.06 0.10 0.53 -0.11 -0.02 Vista city California 0.42 0.22 0.70 0.33 0.13 0.30 West Covina city California 0.35 0.08 0.50 0.46 -0.12 0.08 Westminster city California 0.57 -0.09 1.11 0.69 -0.53 -0.01 Arvada city Colorado -0.51 -0.71 -0.66 -0.15 -0.93 -0.81
Page A-14 City State Index 1: Equal Weight Index 2: Triple Weight to Poverty and Structural Problems Index 3: Triple Weight to Immigration and Housing Affordability Factor Index 4: Hedonic Weights Index 5: Richardson Weights Index 6: Partial Hedonic Weights Aurora city Colorado 0.16 0.03 0.19 0.25 -0.07 0.02 Boulder city Colorado -1.03 -0.70 -0.65 -1.73 -0.32 -0.48 Colorado Springs city Colorado -0.45 -0.40 -0.54 -0.39 -0.39 -0.44 Denver city Colorado -0.04 0.10 0.09 -0.32 0.26 0.18 Fort Collins city Colorado -0.81 -0.57 -0.75 -1.11 -0.34 -0.50 Greeley city Colorado -0.04 -0.07 -0.08 0.02 -0.10 -0.09 Lakewood city Colorado -0.33 -0.37 -0.47 -0.14 -0.44 -0.43 Longmont city Colorado -0.16 -0.31 -0.07 -0.10 -0.42 -0.30 Pueblo city Colorado 0.31 0.53 -0.14 0.54 0.61 0.39 Thornton city Colorado -0.20 -0.51 -0.28 0.20 -0.81 -0.61 Westminster city Colorado -0.48 -0.69 -0.58 -0.17 -0.90 -0.77 Washington city District of Columbia -0.25 0.42 -0.10 -1.06 1.04 0.61 Boca Raton city Florida -0.62 -0.75 -0.51 -0.59 -0.84 -0.73 Cape Coral city Florida -0.17 -0.44 -0.17 0.09 -0.67 -0.49 Clearwater city Florida -0.18 -0.29 -0.21 -0.05 -0.40 -0.33 Davie town Florida -0.33 -0.50 -0.22 -0.28 -0.62 -0.49 Deerfield Beach city Florida 0.16 -0.04 0.54 -0.01 -0.12 0.07 Deltona city Florida -0.10 -0.48 -0.25 0.42 -0.84 -0.61 Fort Lauderdale city Florida -0.09 0.01 -0.16 -0.11 0.08 0.00 Gainesville city Florida -0.20 0.25 -0.23 -0.62 0.64 0.33 Hollywood city Florida 0.06 -0.14 0.31 0.01 -0.26 -0.08 Jacksonville city Florida -0.29 -0.25 -0.45 -0.17 -0.26 -0.31 Largo city Florida -0.18 -0.44 -0.31 0.23 -0.71 -0.55 Melbourne city Florida -0.25 -0.38 -0.36 -0.02 -0.52 -0.45 Miami city Florida 1.47 1.48 2.34 0.60 1.70 1.83 Miami Beach city Florida 0.39 0.28 1.22 -0.32 0.38 0.59 Miramar city Florida -0.16 -0.45 0.04 -0.06 -0.66 -0.43 Orlando city Florida 0.10 0.11 0.18 0.01 0.15 0.15
Page A-15 City State Index 1: Equal Weight Index 2: Triple Weight to Poverty and Structural Problems Index 3: Triple Weight to Immigration and Housing Affordability Factor Index 4: Hedonic Weights Index 5: Richardson Weights Index 6: Partial Hedonic Weights Palm Bay city Florida -0.11 -0.31 -0.16 0.14 -0.49 -0.37 Pembroke Pines city Florida -0.39 -0.63 -0.21 -0.33 -0.79 -0.60 Plantation city Florida -0.40 -0.65 -0.32 -0.24 -0.85 -0.67 Pompano Beach city Florida 0.25 0.04 0.37 0.35 -0.12 0.04 St. Petersburg city Florida -0.18 -0.14 -0.36 -0.03 -0.15 -0.20 Sunrise city Florida 0.04 -0.29 0.22 0.20 -0.54 -0.29 Tallahassee city Florida -0.65 -0.37 -0.66 -0.93 -0.12 -0.31 Tampa city Florida 0.02 0.16 -0.04 -0.05 0.26 0.16 West Palm Beach city Florida 0.27 0.35 0.41 0.06 0.45 0.42 Atlanta city Georgia 0.09 0.86 -0.19 -0.39 1.47 0.90 Roswell city Georgia -0.85 -1.04 -0.58 -0.92 -1.15 -0.97 Savannah city Georgia 0.18 0.54 -0.11 0.12 0.79 0.50 Honolulu CDP Hawaii -0.31 -0.38 0.05 -0.58 -0.37 -0.26 Boise City Idaho -0.53 -0.54 -0.64 -0.42 -0.57 -0.58 Nampa city Idaho 0.23 0.26 0.05 0.38 0.24 0.20 Aurora city Illinois 0.24 0.00 0.53 0.21 -0.15 0.06 Champaign city Illinois -0.59 -0.30 -0.53 -0.96 -0.02 -0.21 Chicago city Illinois 0.33 0.65 0.37 -0.01 0.93 0.72 Elgin city Illinois 0.29 -0.05 0.49 0.45 -0.31 -0.04 Evanston city Illinois -1.10 -0.91 -0.92 -1.47 -0.70 -0.80 Joliet city Illinois -0.05 -0.22 -0.18 0.25 -0.40 -0.30 Naperville city Illinois -1.15 -1.26 -0.94 -1.23 -1.32 -1.21 Peoria city Illinois -0.43 0.01 -0.80 -0.51 0.31 -0.05 Rockford city Illinois 0.24 0.41 -0.07 0.38 0.48 0.32 Schaumburg village Illinois -0.57 -0.79 -0.33 -0.58 -0.93 -0.74 Springfield city Illinois -0.42 -0.18 -0.75 -0.33 -0.06 -0.27 Waukegan city Illinois 0.68 0.40 0.93 0.70 0.22 0.44 Bloomington city Indiana -0.66 -0.24 -0.46 -1.28 0.17 -0.08 Evansville city Indiana 0.04 0.19 -0.44 0.35 0.21 0.04
Page A-16 City State Index 1: Equal Weight Index 2: Triple Weight to Poverty and Structural Problems Index 3: Triple Weight to Immigration and Housing Affordability Factor Index 4: Hedonic Weights Index 5: Richardson Weights Index 6: Partial Hedonic Weights Fort Wayne city Indiana -0.01 0.14 -0.46 0.30 0.16 -0.01 Hammond city Indiana 0.57 0.73 0.17 0.80 0.77 0.60 Indianapolis city (balance) Indiana -0.07 0.11 -0.41 0.09 0.18 0.01 South Bend city Indiana 0.29 0.61 -0.08 0.34 0.79 0.52 Cedar Rapids city Iowa -0.37 -0.24 -0.65 -0.23 -0.20 -0.33 Davenport city Iowa -0.26 -0.07 -0.55 -0.17 0.02 -0.15 Des Moines city Iowa -0.30 -0.24 -0.45 -0.23 -0.22 -0.28 Sioux City Iowa 0.05 0.03 -0.26 0.39 -0.07 -0.10 Kansas City Kansas 0.57 0.83 0.24 0.65 0.98 0.75 Lawrence city Kansas -0.81 -0.59 -0.63 -1.21 -0.36 -0.48 Olathe city Kansas -0.80 -0.84 -0.83 -0.72 -0.88 -0.86 Overland Park city Kansas -0.94 -1.05 -0.89 -0.87 -1.14 -1.06 Topeka city Kansas -0.17 -0.03 -0.56 0.08 0.00 -0.16 Wichita city Kansas -0.10 -0.05 -0.35 0.10 -0.06 -0.14 Lexington -Fayette Kentucky -0.51 -0.38 -0.61 -0.53 -0.30 -0.40 Baton Rouge city Louisiana 0.30 0.75 -0.13 0.27 1.04 0.67 Lafayette city Louisiana -0.35 -0.25 -0.56 -0.25 -0.22 -0.32 Shreveport city Louisiana 0.26 0.62 -0.35 0.51 0.78 0.45 Baltimore city Maryland 0.30 0.91 -0.02 0.01 1.36 0.90 Boston city Massachusetts 0.02 0.52 0.43 -0.90 1.06 0.78 Brockton city Massachusetts 0.37 0.38 0.38 0.34 0.39 0.39 Cambridge city Massachusetts -0.88 -0.51 -0.28 -1.84 -0.04 -0.20 Fall River city Massachusetts 0.51 0.55 0.44 0.54 0.57 0.53 Lawrence city Massachusetts 1.47 2.04 1.85 0.53 2.63 2.30 Lowell city Massachusetts 0.46 0.71 0.72 -0.06 1.00 0.87 Lynn city Massachusetts 0.49 0.73 0.71 0.02 0.99 0.86 New Bedford city Massachusetts 0.60 0.76 0.40 0.63 0.86 0.72 Newton city Massachusetts -1.21 -1.20 -0.96 -1.49 -1.12 -1.09
Page A-17 City State Index 1: Equal Weight Index 2: Triple Weight to Poverty and Structural Problems Index 3: Triple Weight to Immigration and Housing Affordability Factor Index 4: Hedonic Weights Index 5: Richardson Weights Index 6: Partial Hedonic Weights Quincy city Massachusetts -0.40 -0.58 -0.19 -0.44 -0.68 -0.53 Somerville city Massachusetts -0.13 0.12 0.34 -0.84 0.45 0.36 Springfield city Massachusetts 0.81 1.27 0.50 0.65 1.59 1.24 Worcester city Massachusetts 0.04 0.34 0.21 -0.43 0.65 0.47 Ann Arbor city Michigan -1.00 -0.63 -0.52 -1.84 -0.19 -0.37 Detroit city Michigan 0.99 1.85 0.24 0.89 2.41 1.72 Grand Rapids city Michigan 0.17 0.43 -0.02 0.11 0.60 0.40 Lansing city Michigan 0.19 0.54 -0.11 0.13 0.77 0.49 Livonia city Michigan -0.64 -0.96 -0.75 -0.23 -1.26 -1.06 Pontiac city Michigan 0.87 1.18 0.44 0.98 1.34 1.07 Southfield city Michigan -0.27 -0.39 -0.51 0.09 -0.55 -0.51 Sterling Heights city Michigan -0.20 -0.54 -0.24 0.17 -0.85 -0.62 Troy city Michigan -0.84 -0.96 -0.69 -0.86 -1.04 -0.93 Warren city Michigan 0.08 -0.14 -0.20 0.58 -0.41 -0.30 Westland city Michigan -0.10 -0.29 -0.39 0.37 -0.52 -0.44 Wyoming city Michigan 0.10 -0.10 -0.02 0.44 -0.32 -0.20 Bloomington city Minnesota -0.60 -0.70 -0.62 -0.49 -0.78 -0.72 Brooklyn Park city Minnesota -0.22 -0.30 -0.27 -0.10 -0.38 -0.34 Duluth city Minnesota -0.63 -0.37 -0.77 -0.75 -0.18 -0.37 Minneapolis city Minnesota -0.27 0.22 -0.11 -0.93 0.69 0.38 Plymouth city Minnesota -0.90 -1.00 -0.84 -0.87 -1.07 -1.00 Rochester city Minnesota -0.82 -0.78 -0.73 -0.94 -0.72 -0.73 St. Paul city Minnesota -0.21 0.17 -0.16 -0.65 0.52 0.27 Columbia city Missouri -0.76 -0.54 -0.69 -1.05 -0.32 -0.46 Independence city Missouri -0.04 -0.14 -0.33 0.36 -0.31 -0.28 Kansas City Missouri -0.13 0.15 -0.39 -0.14 0.33 0.10 Lee's Summit city Missouri -0.86 -0.95 -0.94 -0.70 -1.04 -1.00 St. Louis city Missouri 0.35 1.24 -0.20 0.01 1.88 1.20 Springfield city Missouri -0.15 0.01 -0.47 0.03 0.06 -0.09
Page A-18 City State Index 1: Equal Weight Index 2: Triple Weight to Poverty and Structural Problems Index 3: Triple Weight to Immigration and Housing Affordability Factor Index 4: Hedonic Weights Index 5: Richardson Weights Index 6: Partial Hedonic Weights Lincoln city Nebraska -0.57 -0.49 -0.59 -0.62 -0.43 -0.49 Omaha city Nebraska -0.18 -0.02 -0.33 -0.20 0.09 -0.05 Henderson city Nevada -0.53 -0.77 -0.64 -0.19 -1.01 -0.86 Las Vegas city Nevada 0.12 -0.16 0.19 0.34 -0.38 -0.19 North Las Vegas city Nevada 0.41 0.05 0.57 0.59 -0.21 0.05 Reno city Nevada -0.11 -0.16 -0.01 -0.17 -0.17 -0.12 Sparks city Nevada -0.13 -0.46 -0.08 0.15 -0.74 -0.51 Manchester city New Hampshire -0.24 -0.10 -0.21 -0.41 0.04 -0.06 Nashua city New Hampshire -0.52 -0.59 -0.41 -0.56 -0.62 -0.56 Camden city New Jersey 2.03 3.02 1.76 1.31 3.82 3.11 Clifton city New Jersey -0.08 -0.40 0.16 0.01 -0.63 -0.37 Elizabeth city New Jersey 0.97 0.91 1.35 0.64 0.96 1.06 Jersey City New Jersey 0.40 0.60 0.80 -0.20 0.88 0.80 Newark city New Jersey 1.15 1.50 1.35 0.59 1.86 1.65 Passaic city New Jersey 1.71 1.73 2.45 0.96 1.93 2.03 Paterson city New Jersey 1.20 1.28 1.56 0 .77 1.44 1.44 Trenton city New Jersey 1.07 1.46 0.92 0.83 1.76 1.47 Albuquerque city New Mexico -0.17 -0.20 -0.27 -0.03 -0.26 -0.25 Santa Fe city New Mexico -0.34 -0.25 -0.27 -0.50 -0.16 -0.21 Albany city New York -0.07 0.80 -0.38 -0.62 1.48 0.85 Buffalo city New York 0.55 1.45 -0.21 0.42 2.04 1.32 New Rochelle city New York -0.21 -0.30 0.05 -0.38 -0.31 -0.21 New York city New York 0.33 0.52 0.71 -0.23 0.77 0.71 Rochester city New York 0.77 1.62 0.30 0.40 2.25 1.60 Syracuse city New York 0.44 1.29 0.00 0.04 1.91 1.28 Yonkers city New York 0.14 0.14 0.22 0.04 0.17 0.18 Asheville city North Carolina -0.49 -0.29 -0.62 -0.55 -0.15 -0.30 Cary town North Carolina -0.90 -1.01 -0.69 -1.00 -1.05 -0.94 Charlotte city North Carolina -0.25 -0.22 -0.28 -0.27 -0.19 -0.22
Page A-19 City State Index 1: Equal Weight Index 2: Triple Weight to Poverty and Structural Problems Index 3: Triple Weight to Immigration and Housing Affordability Factor Index 4: Hedonic Weights Index 5: Richardson Weights Index 6: Partial Hedonic Weights Durham city North Carolina -0.21 -0.08 -0.15 -0.39 0.04 -0.03 Fayetteville city North Carolina 0.02 0.19 -0.45 0.34 0.21 0.03 Gastonia city North Carolina 0.40 0.42 0.08 0.69 0.36 0.30 Greensboro city North Carolina -0.08 -0.01 -0.21 0.00 0.01 -0.05 High Point city North Carolina 0.16 0.20 -0.12 0.40 0.17 0.10 Raleigh city North Carolina -0.51 -0.32 -0.43 -0.78 -0.14 -0.25 Wilmington city North Carolina -0.47 -0.30 -0.55 -0.56 -0.17 -0.30 Winston-Salem city North Carolina 0.07 0.24 -0.08 0.04 0.36 0.22 Fargo city North Dakota -0.65 -0.53 -0.76 -0.68 -0.44 -0.54 Akron city Ohio 0.12 0.52 -0.33 0.18 0.76 0.42 Cincinnati city Ohio 0.13 0.89 -0.36 -0.14 1.43 0.85 Cleveland city Ohio 0.96 1.83 0.17 0.87 2.40 1.69 Columbus city Ohio -0.08 0.19 -0.27 -0.16 0.38 0.17 Dayton city Ohio 0.64 1.41 -0.13 0.64 1.89 1.25 Lorain city Ohio 0.46 0.58 -0.09 0.89 0.55 0.39 Parma city Ohio -0.25 -0.52 -0.47 0.23 -0.81 -0.66 Toledo city Ohio 0.31 0.64 -0.30 0.60 0.78 0.46 Broken Arrow city Oklahoma -0.51 -0.63 -0.75 -0.16 -0.79 -0.75 Edmond city Oklahoma -0.90 -0.95 -0.93 -0.82 -1.00 -0.97 Lawton city Oklahoma 0.18 0.30 -0.32 0.57 0.28 0.12 Norman city Oklahoma -0.64 -0.49 -0.73 -0.68 -0.38 -0.50 Oklahoma City Oklahoma 0.04 0.17 -0.26 0.21 0.21 0.08 Tulsa city Oklahoma -0.01 0.14 -0.27 0.10 0.20 0.06 Beaverton city Oregon -0.47 -0.61 -0.26 -0.54 -0.68 -0.55 Eugene city Oregon -0.54 -0.38 -0.50 -0.75 -0.22 -0.33 Gresham city Oregon 0.32 0.24 0.33 0.39 0.17 0.22 Hillsboro city Oregon 0.09 0.13 0.20 -0.07 0.19 0.18 Medford city Oregon -0.21 -0.25 -0.37 0.00 -0.33 -0.32 Portland city Oregon -0.30 -0.10 -0.27 -0.53 0.08 -0.05
Page A-20 City State Index 1: Equal Weight Index 2: Triple Weight to Poverty and Structural Problems Index 3: Triple Weight to Immigration and Housing Affordability Factor Index 4: Hedonic Weights Index 5: Richardson Weights Index 6: Partial Hedonic Weights Salem city Oregon 0.15 0.10 0.09 0.25 0.04 0.06 Allentown city Pennsylvania 0.40 0.82 0.31 0.07 1.17 0.87 Bethlehem city Pennsylvania -0.25 -0.17 -0.35 -0.24 -0.12 -0.19 Erie city Pennsylvania 0.12 0.48 -0.40 0.28 0.66 0.34 Philadelphia city Pennsylvania 0.45 0.96 0.15 0.23 1.34 0.95 Pittsburgh city Pennsylvania -0.08 0.58 -0.48 -0.35 1.07 0.56 Reading city Pennsylvania 1.34 1.96 1.05 1.00 2.43 1.97 Scranton city Pennsylvania 0.21 0.55 -0.14 0.20 0.77 0.49 Cranston city Rhode Island -0.38 -0.59 -0.39 -0.16 -0.78 -0.63 Pawtucket city Rhode Island 0.47 0.65 0.42 0.33 0.79 0.66 Providence city Rhode Island 0.56 1.18 0.74 -0.25 1.77 1.38 Warwick city Rhode Island -0.52 -0.75 -0.64 -0.17 -0.98 -0.84 Charleston city South Carolina -0.46 -0.12 -0.59 -0.67 0.14 -0.11 Columbia city South Carolina -0.31 0.10 -0.48 -0.54 0.42 0.11 Sioux Falls city South Dakota -0.45 -0.39 -0.57 -0.39 -0.36 -0.42 Chattanooga city Tennessee -0.06 0.29 -0.53 0.06 0.48 0.17 Clarksville city Tennessee -0.35 -0.40 -0.60 -0.04 -0.51 -0.51 Knoxville city Tennessee -0.06 0.33 -0.42 -0.10 0.58 0.26 Memphis city Tennessee 0.39 0.68 -0.11 0.59 0.81 0.54 Murfreesboro city Tennessee -0.45 -0.31 -0.51 -0.53 -0.21 -0.31 Nashville-Davidson (balance) Tennessee -0.19 -0.05 -0.29 -0.24 0.05 -0.06 Abilene city Texas 0.07 0.12 -0.28 0.36 0.07 -0.01 Amarillo city Texas 0.09 0.05 -0.16 0.38 -0.05 -0.06 Arlington city Texas 0.06 -0.12 0.01 0.29 -0.29 -0.17 Austin city Texas -0.17 -0.08 -0.03 -0.39 0.03 -0.01 Baytown city Texas 0.50 0.28 0.26 0.96 0.03 0.14 Beaumont city Texas 0.29 0.56 -0.20 0.50 0.68 0.42 Bryan city Texas 0.54 0.56 0.43 0.63 0.55 0.52
Page A-21 City State Index 1: Equal Weight Index 2: Triple Weight to Poverty and Structural Problems Index 3: Triple Weight to Immigration and Housing Affordability Factor Index 4: Hedonic Weights Index 5: Richardson Weights Index 6: Partial Hedonic Weights Carrollton city Texas -0.14 -0.52 0.02 0.08 -0.82 -0.54 College Station city Texas -0.70 -0.45 -0.39 -1.27 -0.16 -0.28 Corpus Christi city Texas 0.25 0.18 -0.02 0.59 0.06 0.06 Dallas city Texas 0.80 0.78 0.97 0.64 0.81 0.85 Denton city Texas -0.41 -0.39 -0.33 -0.50 -0.36 -0.36 Fort Worth city Texas 0.37 0.38 0.30 0.44 0.36 0.35 Garland city Texas 0.54 0.21 0.61 0.79 -0.05 0.18 Grand Prairie city Texas 0.33 0.13 0.20 0.65 -0.07 0.04 Houston city Texas 0.82 0.75 0.94 0.75 0.73 0.79 Irving city Texas 0.42 0.09 0.71 0.46 -0.13 0.14 Killeen city Texas -0.07 -0.03 -0.41 0.24 -0.08 -0.16 Lewisville city Texas -0.21 -0.48 -0.16 -0.01 -0.69 -0.51 Lubbock city Texas 0.00 0.01 -0.30 0.28 -0.05 -0.11 McKinney city Texas -0.67 -0.83 -0.64 -0.53 -0.96 -0.85 Mesquite city Texas 0.28 -0.16 0.17 0.81 -0.57 -0.29 Midland city Texas 0.10 -0.05 -0.16 0.50 -0.25 -0.18 Odessa city Texas 0.45 0.30 0.01 1.03 0.06 0.09 Pasadena city Texas 0.95 0.62 0.99 1.23 0.35 0.57 Plano city Texas -0.71 -0.88 -0.55 -0.70 -0.98 -0.85 Richardson city Texas -0.58 -0.82 -0.41 -0.50 -0.99 -0.80 Round Rock city Texas -0.40 -0.54 -0.50 -0.17 -0.68 -0.60 San Angelo city Texas 0.08 0.04 -0.28 0.49 -0.09 -0.12 San Antonio city Texas 0.32 0.26 0.20 0.50 0.17 0.20 Tyler city Texas 0.19 0.31 -0.13 0.39 0.34 0.21 Waco city Texas 0.41 0.52 0.19 0.51 0.57 0.46 Wichita Falls city Texas -0.05 -0.01 -0.30 0.16 -0.04 -0.10 Ogden city Utah 0.21 0.53 0.15 -0.03 0.78 0.56 Orem city Utah -0.46 -0.56 -0.53 -0.31 -0.65 -0.60 Provo city Utah -0.41 -0.09 -0.17 -0.96 0.25 0.07
Page A-22 City State Index 1: Equal Weight Index 2: Triple Weight to Poverty and Structural Problems Index 3: Triple Weight to Immigration and Housing Affordability Factor Index 4: Hedonic Weights Index 5: Richardson Weights Index 6: Partial Hedonic Weights Salt Lake City Utah -0.18 0.05 -0.07 -0.54 0.29 0.15 Sandy city Utah -0.72 -0.89 -0.86 -0.42 -1.06 -0.97 West Jordan city Utah -0.40 -0.70 -0.48 -0.03 -0.98 -0.79 Alexandria city Virginia -0.72 -0.77 -0.18 -1.20 -0.68 -0.57 Chesapeake city Virginia -0.46 -0.68 -0.65 -0.05 -0.91 -0.80 Hampton city Virginia -0.21 -0.15 -0.48 -0.02 -0.16 -0.24 Newport News city Virginia -0.21 -0.12 -0.40 -0.11 -0.09 -0.18 Norfolk city Virginia -0.01 0.30 -0.18 -0.15 0.53 0.29 Portsmouth city Virginia 0.03 0.23 -0.27 0.14 0.33 0.15 Richmond city Virginia -0.02 0.38 -0.21 -0.23 0.68 0.39 Roanoke city Virginia -0.05 0.14 -0.45 0.16 0.21 0.02 Suffolk city Virginia -0.32 -0.35 -0.59 -0.01 -0.44 -0.46 Virginia Beach city Virginia -0.53 -0.66 -0.62 -0.30 -0.80 -0.72 Bellevue city Washington -0.75 -0.84 -0.30 -1.10 -0.81 -0.68 Bellingham city Washington -0.66 -0.44 -0.58 -0.97 -0.23 -0.36 Everett city Washington 0.02 0.03 0.06 -0.03 0.05 0.05 Kent city Washington 0.05 -0.05 0.20 -0.01 -0.10 -0.01 Seattle city Washington -0.71 -0.55 -0.44 -1.13 -0.34 -0.41 Spokane city Washington -0.22 -0.03 -0.45 -0.16 0.07 -0.09 Tacoma city Washington -0.03 0.16 -0.09 -0.17 0.32 0.18 Vancouver city Washington -0.01 0.06 -0.05 -0.04 0.11 0.06 Yakima city Washington 0.73 0.71 0.54 0.93 0.65 0.63 Green Bay city Wisconsin -0.09 -0.02 -0.23 -0.03 0.01 -0.06 Kenosha city Wisconsin -0.10 -0.21 -0.30 0.19 -0.34 -0.30 Madison city Wisconsin -0.79 -0.49 -0.61 -1.28 -0.18 -0.36 Milwaukee city Wisconsin 0.46 0.96 0.19 0.23 1.34 0.96 Waukesha city Wisconsin -0.68 -0.83 -0.77 -0.43 -0.98 -0.89
Table A.9. Factor and Equal Weight Index Scores in 2005 and 2000 for 370 Cities Page A-23 City State 2005 Population 2005 Factor 1 2005 Factor 2 2005 Factor 3 2005 Equal Weight Index 2000 Factor 1 2000 Factor 2 2000 Factor 3 2000 Equal Weight Index Birmingham city Alabama 222,154 2.46 -1.11 0.26 0.54 1.82 -1.26 0.82 0.46 Huntsville city Alabama 158,618 -0.05 -1.13 -0.11 -0.43 -0.10 -1.09 0.16 -0.34 Mobile city Alabama 193,332 1.03 -1.14 0.63 0.17 0.71 -1.17 0.83 0.12 Montgomery city Alabama 193,042 0.45 -1.26 0.61 -0.07 0.48 -1.25 0.75 -0.01 Anchorage municipality Alaska 266,281 -0.61 -0.73 -0.06 -0.47 -0.78 -0.66 0.04 -0.47 Avondale city Arizona 61,666 0.08 0.68 0.48 0.42 -0.42 0.45 0.83 0.29 Chandler city Arizona 225,725 -1.01 -0.30 -0.07 -0.46 -1.08 -0.28 0.11 -0.42 Glendale city Arizona 229,913 -0.33 -0.15 0.63 0.05 -0.54 -0.20 0.72 0.00 Mesa city Arizona 442,445 -0.56 0.05 0.41 -0.03 -0.83 -0.17 0.63 -0.12 Peoria city Arizona 141,941 -1.16 -0.58 0.41 -0.44 -1.19 -0.65 0.56 -0.43 Phoenix city Arizona 1,377,980 -0.04 0.70 0.34 0.33 -0.15 0.52 0.52 0.30 Scottsdale city Arizona 215,933 -1.10 -0.75 -0.70 -0.85 -1.26 -0.71 -0.48 -0.81 Tempe city Arizona 166,171 -0.69 -0.04 -0.61 -0.44 -0.60 -0.20 -0.64 -0.48 Tucson city Arizona 507,362 0.26 0.12 0.13 0.17 0.09 0.01 0.37 0.16 Yuma city Arizona 91,433 -0.37 -0.14 1.33 0.28 -0.52 -0.04 1.55 0.33 Fayetteville city Arkansas 58,839 -0.02 -0.67 -1.44 -0.71 -0.12 -0.62 -0.89 -0.54 Fort Smith city Arkansas 81,054 0.26 -0.73 0.87 0.13 -0.04 -0.72 1.20 0.15 Little Rock city Arkansas 176,924 0.16 -1.06 -0.21 -0.37 0.07 -0.99 0.13 -0.27 Alameda city California 77,058 -0.48 0.11 -1.30 -0.56 -0.62 0.31 -1.08 -0.46 Alhambra city California 76,309 -0.40 2.15 -0.64 0.37 -0.45 2.50 -0.03 0.68 Anaheim city California 329,483 -0.67 1.62 0.92 0.62 -0.61 1.73 0.99 0.70 Antioch city California 103,339 -0.48 0.07 0.31 -0.03 -0.64 -0.28 0.49 -0.14 Bakersfield city California 286,316 -0.18 -0.01 0.83 0.21 -0.22 -0.16 1.07 0.23 Baldwin Park city California 84,812 0.09 2.36 1.50 1.32 -0.25 2.82 1.99 1.52 Bellflower city California 78,198 0.12 1.07 0.73 0.64 -0.15 1.15 1.12 0.70 Berkeley city California 90,432 0.08 0.02 -3.22 -1.04 0.19 -0.01 -3.26 -1.03 Buena Park city California 76,062 -0.94 1.03 0.55 0.21 -0.82 1.16 0.88 0.40
City State 2005 Population 2005 Factor 1 2005 Factor 2 2005 Factor 3 2005 Equal Weight Index Page A-24 2000 Factor 1 2000 Factor 2 2000 Factor 3 2000 Equal Weight Index Burbank city California 100,053 -0.84 0.45 -0.35 -0.25 -0.78 0.47 0.08 -0.08 Carlsbad city California 92,998 -1.30 -0.56 -0.86 -0.91 -1.21 -0.54 -0.56 -0.77 Carson city California 92,156 -0.74 0.49 1.02 0.26 -0.96 0.66 1.56 0.42 Chico city California 71,298 0.37 -0.55 -0.87 -0.35 -0.01 -0.36 -0.93 -0.43 Chino city California 69,732 -1.29 0.23 1.59 0.18 -1.16 0.19 1.85 0.29 Chula Vista city California 212,954 -0.69 0.67 0.53 0.17 -0.61 0.52 0.65 0.19 Clovis city California 80,529 -0.79 -0.43 0.38 -0.28 -0.92 -0.65 0.85 -0.24 Concord city California 116,782 -0.64 0.62 -0.30 -0.11 -0.78 0.42 0.02 -0.12 Corona city California 162,410 -1.09 0.17 0.86 -0.02 -1.08 0.08 0.92 -0.02 Costa Mesa city California 105,333 -0.65 1.19 -0.10 0.15 -0.72 0.88 0.22 0.13 Daly City California 93,513 -0.67 1.78 -0.44 0.22 -0.85 1.50 -0.35 0.10 El Cajon city California 92,507 -0.06 0.67 0.23 0.28 0.06 0.49 0.58 0.38 Escondido city California 133,017 -0.69 1.16 0.63 0.37 -0.46 0.98 0.75 0.42 Fairfield city California 102,642 -0.52 0.21 -0.03 -0.11 -0.56 -0.13 0.27 -0.14 Fontana city California 158,235 -0.60 1.12 1.35 0.62 -0.54 0.92 1.79 0.72 Fremont city California 210,387 -1.08 1.26 -1.20 -0.34 -1.15 0.86 -0.86 -0.38 Fresno city California 477,251 0.51 0.70 0.68 0.63 0.57 0.55 0.97 0.70 Fullerton city California 142,064 -1.00 0.62 -0.10 -0.16 -0.84 0.65 0.18 0.00 Garden Grove city California 192,345 -0.66 2.55 0.62 0.83 -0.70 2.28 0.89 0.82 Glendale city California 194,620 -0.88 2.11 -0.49 0.25 -0.38 2.29 -0.62 0.43 Hawthorne city California 100,754 0.49 1.70 0.74 0.97 0.28 1.80 1.41 1.16 Hayward city California 135,474 -0.33 1.00 0.10 0.26 -0.49 1.31 0.35 0.39 Hemet city California 77,076 -0.26 0.60 1.37 0.57 0.01 0.11 1.34 0.48 Hesperia city California 79,714 -0.36 0.15 1.39 0.39 -0.47 -0.19 1.80 0.38 Huntington Beach city California 189,451 -1.16 -0.32 -0.27 -0.58 -1.23 -0.35 0.07 -0.50 Indio city California 65,091 0.19 2.18 0.89 1.09 0.15 2.00 1.65 1.26 Inglewood city California 120,204 0.48 1.48 0.91 0.96 0.63 1.47 1.17 1.09 Lakewood city California 88,253 -1.23 -0.04 0.46 -0.27 -1.18 -0.14 0.87 -0.15
City State 2005 Population 2005 Factor 1 2005 Factor 2 2005 Factor 3 2005 Equal Weight Index Page A-25 2000 Factor 1 2000 Factor 2 2000 Factor 3 2000 Equal Weight Index Livermore city California 87,054 -1.06 -0.26 -0.58 -0.63 -1.12 -0.43 -0.09 -0.55 Long Beach city California 463,956 0.38 1.15 -0.18 0.45 0.62 1.15 0.10 0.62 Los Angeles city California 3,731,437 0.38 1.88 -0.20 0.69 0.43 2.01 0.18 0.87 Merced city California 65,391 0.55 0.78 0.95 0.76 0.68 0.79 1.19 0.89 Mission Viejo city California 90,136 -1.53 -0.72 -0.21 -0.82 -1.55 -0.63 -0.07 -0.75 Modesto city California 202,971 -0.39 -0.06 0.90 0.15 -0.29 -0.07 1.18 0.27 Mountain View city California 69,427 -0.88 0.87 -1.79 -0.60 -0.89 1.11 -1.74 -0.51 Napa city California 73,085 -0.56 0.49 -0.33 -0.14 -0.71 0.34 0.21 -0.05 Norwalk city California 103,844 -0.84 1.54 1.46 0.72 -0.78 1.49 1.80 0.83 Oakland city California 373,910 0.88 1.24 -0.95 0.39 0.91 1.22 -0.79 0.45 Oceanside city California 162,259 -0.98 0.32 0.11 -0.18 -0.58 0.29 0.46 0.06 Ontario city California 156,679 -0.32 1.43 1.35 0.82 -0.39 1.44 1.57 0.87 Orange city California 137,994 -1.14 0.37 0.48 -0.10 -1.00 0.37 0.49 -0.05 Oxnard city California 178,871 0.15 2.21 0.50 0.95 -0.15 2.10 0.82 0.92 Palmdale city California 145,800 -0.09 0.66 1.36 0.64 -0.38 0.22 1.47 0.44 Pasadena city California 129,400 -0.21 0.34 -0.97 -0.28 -0.20 0.92 -0.53 0.07 Pleasanton city California 67,018 -1.22 -0.40 -1.19 -0.93 -1.45 -0.53 -0.68 -0.89 Pomona city California 161,257 -0.24 2.28 1.22 1.09 0.16 2.11 1.50 1.26 Redding city California 89,362 -0.51 -0.73 0.31 -0.31 -0.19 -0.73 0.65 -0.09 Redlands city California 73,548 -0.51 -0.47 -0.10 -0.36 -0.70 -0.38 0.04 -0.35 Redwood City California 81,195 -0.25 1.34 -0.73 0.12 -1.00 0.82 -0.36 -0.18 Rialto city California 93,284 -0.67 1.11 1.55 0.66 -0.26 0.65 1.99 0.79 Richmond city California 96,648 0.38 0.90 -0.27 0.34 0.45 1.01 -0.19 0.42 Riverside city California 294,059 -0.49 0.63 0.55 0.23 -0.37 0.34 0.88 0.28 Roseville city California 108,848 -1.15 -0.46 -0.12 -0.58 -1.21 -0.63 0.13 -0.57 Sacramento city California 445,287 0.52 0.50 -0.24 0.26 0.53 0.42 0.11 0.35 Salinas city California 156,950 -0.04 2.56 1.44 1.32 -0.31 2.14 1.71 1.18 San Bernardino city California 204,552 1.28 1.56 0.50 1.11 1.05 1.00 0.90 0.98
City State 2005 Population 2005 Factor 1 2005 Factor 2 2005 Factor 3 2005 Equal Weight Index Page A-26 2000 Factor 1 2000 Factor 2 2000 Factor 3 2000 Equal Weight Index San Buenaventura (Ventura) city California 100,154 -0.44 -0.29 -0.59 -0.44 -0.59 -0.19 -0.27 -0.35 San Diego city California 1,208,331 -0.30 0.51 -0.95 -0.25 -0.28 0.48 -0.52 -0.11 San Francisco city California 719,077 -0.26 0.95 -1.59 -0.30 -0.32 1.12 -1.44 -0.21 San Jose city California 887,330 -0.56 1.12 -0.50 0.02 -0.72 1.31 -0.35 0.08 San Leandro city California 77,631 -0.73 0.77 0.21 0.08 -0.82 0.58 0.27 0.01 San Marcos city California 77,445 -0.9 4 0.33 0.34 -0.09 -0.71 0.79 0.64 0.24 San Mateo city California 93,481 -0.94 0.59 -0.82 -0.39 -1.00 0.52 -0.57 -0.35 Santa Ana city California 302,302 -0.28 3.39 1.66 1.59 -0.28 3.81 1.96 1.83 Santa Barbara city California 90,708 -0.45 0.14 -0.79 -0.37 -0.49 0.63 -0.64 -0.17 Santa Clara city California 102,204 -0.65 1.23 -1.49 -0.31 -0.85 1.14 -1.33 -0.35 Santa Maria city California 88,817 0.14 2.02 0.65 0.93 0.09 1.91 1.19 1.06 Santa Monica city California 82,777 -1.00 -0.26 -1.47 -0.91 -0.74 -0.25 -1.22 -0.74 Santa Rosa city California 146,500 -0.49 0.08 -0.08 -0.16 -0.70 0.03 -0.04 -0.24 Simi Valley city California 116,722 -1.30 -0.54 0.23 -0.54 -1.18 -0.38 0.38 -0.39 Stockton city California 278,515 0.38 0.83 0.61 0.61 0.52 0.79 1.01 0.78 Sunnyvale city California 132,725 -0.75 1.26 -1.73 -0.40 -1.03 1.26 -1.49 -0.42 Thousand Oaks city California 127,895 -1.07 -0.39 -0.75 -0.74 -1.25 -0.42 -0.40 -0.69 Torrance city California 138,618 -1.16 0.32 -0.76 -0.53 -1.20 0.25 -0.36 -0.43 Tracy city California 82,218 -1.15 0.18 0.78 -0.07 -1.21 -0.20 1.14 -0.09 Turlock city California 74,883 -0.84 0.71 1.14 0.34 -0.38 0.32 1.04 0.33 Tustin city California 79,811 -1.11 1.13 -0.40 -0.13 -1.08 0.93 0.31 0.06 Union City California 65,239 -0.82 1.74 -0.04 0.29 -1.06 1.16 0.00 0.03 Upland city California 74,420 -0.29 -0.29 -0.27 -0.28 -0.70 -0.11 0.40 -0.14 Vacaville city California 81,117 -1.10 -0.50 0.38 -0.41 -1.05 -0.49 0.74 -0.27 Vallejo city California 115,657 -0.18 0.19 -0.05 -0.01 -0.41 0.08 0.15 -0.06 Victorville city California 93,042 -0.04 0.29 1.31 0.52 0.01 -0.04 1.33 0.43 Visalia city California 108,467 -0.18 -0.10 0.97 0.23 -0.40 -0.16 1.18 0.21 Vista city California 83,228 -0.06 1.12 0.19 0.42 -0.33 0.94 0.52 0.38
City State 2005 Population 2005 Factor 1 2005 Factor 2 2005 Factor 3 2005 Equal Weight Index Page A-27 2000 Factor 1 2000 Factor 2 2000 Factor 3 2000 Equal Weight Index West Covina city California 116,371 -0.33 0.74 0.63 0.35 -0.97 0.60 0.86 0.16 Westminster city California 97,946 -1.08 1.92 0.88 0.57 -0.77 1.85 0.71 0.60 Arvada city Colorado 104,766 -1.02 -0.88 0.38 -0.51 -1.04 -0.85 0.37 -0.51 Aurora city Colorado 291,317 -0.16 0.23 0.40 0.16 -0.44 -0.05 0.37 -0.04 Boulder city Colorado 83,432 -0.21 -0.08 -2.79 -1.03 -0.33 -0.15 -2.67 -1.05 Colorado Springs city Colorado 376,985 -0.34 -0.69 -0.31 -0.45 -0.56 -0.72 -0.32 -0.53 Denver city Colorado 545,198 0.31 0.28 -0.73 -0.04 0.25 0.26 -0.63 -0.04 Fort Collins city Colorado 122,297 -0.20 -0.67 -1.56 -0.81 -0.52 -0.63 -1.49 -0.88 Greeley city Colorado 82,836 -0.11 -0.14 0.11 -0.04 -0.09 -0.27 0.29 -0.03 Lakewood city Colorado 142,434 -0.43 -0.69 0.14 -0.33 -0.74 -0.57 0.05 -0.42 Longmont city Colorado 76,181 -0.54 0.08 -0.01 -0.16 -0.59 -0.14 0.03 -0.24 Pueblo city Colorado 101,302 0.86 -0.81 0.88 0.31 0.35 -0.69 1.06 0.24 Thornton city Colorado 102,331 -0.98 -0.41 0.80 -0.20 -1.00 -0.61 1.09 -0.17 Westminster city Colorado 99,305 -1.01 -0.72 0.29 -0.48 -1.08 -0.54 0.29 -0.44 Washington city District of Columbia 515,118 1.41 0.12 -2.28 -0.25 1.31 0.10 -1.74 -0.11 Boca Raton city Florida 74,361 -0.95 -0.36 -0.55 -0.62 -1.26 -0.47 -0.48 -0.73 Cape Coral city Florida 134,388 -0.84 -0.16 0.48 -0.17 -1.01 -0.58 0.79 -0.27 Clearwater city Florida 108,382 -0.46 -0.26 0.16 -0.18 -0.35 -0.34 0.14 -0.18 Davie town Florida 88,683 -0.75 -0.05 -0.20 -0.33 -0.98 -0.26 0.56 -0.23 Deerfield Beach city Florida 71,599 -0.34 1.11 -0.27 0.16 -0.57 0.27 0.67 0.12 Deltona city Florida 85,979 -1.04 -0.47 1.21 -0.10 -0.86 -0.61 1.27 -0.07 Fort Lauderdale city Florida 141,307 0.15 -0.26 -0.15 -0.09 0.14 0.14 0.12 0.13 Gainesville city Florida 100,879 0.93 -0.27 -1.26 -0.20 0.24 -0.56 -0.90 -0.41 Hollywood city Florida 138,412 -0.44 0.68 -0.07 0.06 -0.42 0.29 0.46 0.11 Jacksonville city Florida 768,537 -0.19 -0.69 0.01 -0.29 -0.23 -0.73 0.38 -0.19 Largo city Florida 71,269 -0.84 -0.52 0.83 -0.18 -0.75 -0.52 0.83 -0.15 Melbourne city Florida 76,373 -0.58 -0.51 0. 33 -0.25 -0.45 -0.64 0.34 -0.25 Miami city Florida 361,701 1.48 3.64 -0.70 1.47 1.42 3.61 -0.07 1.65
City State 2005 Population 2005 Factor 1 2005 Factor 2 2005 Factor 3 2005 Equal Weight Index Page A-28 2000 Factor 1 2000 Factor 2 2000 Factor 3 2000 Equal Weight Index Miami Beach city Florida 84,086 0.10 2.46 -1.38 0.39 0.84 2.31 -1.32 0.61 Miramar city Florida 115,444 -0.90 0.35 0.09 -0.16 -0.93 0.41 0.66 0.04 Orlando city Florida 221,299 0.14 0.29 -0.13 0.10 0.18 0.07 -0.21 0.01 Palm Bay city Florida 90,102 -0.60 -0.25 0.51 -0.11 -0.65 -0.56 0.92 -0.09 Pembroke Pines city Florida 159,422 -0.99 0.07 -0.24 -0.39 -1.25 -0.08 0.18 -0.39 Plantation city Florida 88,859 -1.02 -0.19 0.00 -0.40 -1.24 -0.32 0.13 -0.48 Pompano Beach city Florida 94,892 -0.28 0.54 0.49 0.25 -0.02 0.18 0.65 0.27 St. Petersburg city Florida 232,960 -0.08 -0.63 0.18 -0.18 -0.11 -0.62 0.40 -0.11 Sunrise city Florida 86,586 -0.80 0.48 0.44 0.04 -0.83 0.22 0.52 -0.03 Tallahassee city Florida 141,148 0.06 -0.67 -1.36 -0.65 0.07 -0.82 -0.97 -0.57 Tampa city Florida 312,855 0.36 -0.14 -0.15 0.02 0.40 -0.17 0.20 0.15 West Palm Beach city Florida 86,804 0.46 0.62 -0.25 0.27 0.29 0.44 0.11 0.28 Atlanta city Georgia 394,929 2.02 -0.63 -1.11 0.09 1.82 -0.58 -0.69 0.18 Roswell city Georgia 98,137 -1.33 -0.18 -1.03 -0.85 -1.24 -0.19 -0.93 -0.79 Savannah city Georgia 117,478 1.08 -0.55 0.01 0.18 0.92 -0.57 0.49 0.28 Honolulu CDP Hawaii 362,252 -0.50 0.57 -0.99 -0.31 -0.54 0.71 -0.65 -0.16 Boise City Idaho 191,667 -0.54 -0.80 -0.26 -0.53 -0.75 -0.79 -0.16 -0.57 Nampa city Idaho 67,112 0.31 -0.22 0.60 0.23 -0.27 -0.28 0.88 0.11 Aurora city Illinois 170,490 -0.38 0.96 0.15 0.24 -0.75 0.46 0.51 0.07 Champaign city Illinois 65,600 0.15 -0.43 -1.51 -0.59 -0.18 -0.47 -1.62 -0.76 Chicago city Illinois 2,701,926 1.12 0.42 -0.54 0.33 0.97 0.47 -0.24 0.40 Elgin city Illinois 93,412 -0.58 0.78 0.68 0.29 -0.80 0.63 0.78 0.20 Evanston city Illinois 62,258 -0.63 -0.65 -2.02 -1.10 -0.58 -0.47 -1.79 -0.94 Joliet city Illinois 128,090 -0.47 -0.37 0.69 -0.05 -0.41 -0.39 0.91 0.04 Naperville city Illinois 147,779 -1.44 -0.64 -1.35 -1.15 -1.48 -0.74 -1.06 -1.10 Peoria city Illinois 102,136 0.68 -1.35 -0.63 -0.43 0.86 -1.21 -0.25 -0.20 Rockford city Illinois 139,173 0.67 -0.54 0.59 0.24 0.26 -0.69 0.72 0.10 Schaumburg village Illinois 77,817 -1.13 0.03 -0.60 -0.57 -1.37 -0.23 -0.18 -0.60
City State 2005 Population 2005 Factor 1 2005 Factor 2 2005 Factor 3 2005 Equal Weight Index Page A-29 2000 Factor 1 2000 Factor 2 2000 Factor 3 2000 Equal Weight Index Springfield city Illinois 110,262 0.17 -1.24 -0.20 -0.42 -0.02 -1.13 -0.07 -0.41 Waukegan city Illinois 82,355 -0.02 1.30 0.75 0.68 -0.08 1.12 1.04 0.69 Bloomington city Indiana 55,406 0.38 -0.16 -2.21 -0.66 0.09 -0.47 -2.34 -0.90 Evansville city Indiana 110,708 0.43 -1.16 0.83 0.04 0.19 -1.19 0.88 -0.04 Fort Wayne city Indiana 219,346 0.36 -1.14 0.76 -0.01 0.18 -1.14 0.82 -0.05 Hammond city Indiana 72,507 0.97 -0.43 1.15 0.57 0.35 -0.61 1.33 0.36 Indianapolis city (balance) Indiana 765,310 0.38 -0.91 0.32 -0.07 0.14 -0.99 0.39 -0.15 South Bend city Indiana 97,070 1.08 -0.63 0.41 0.29 0.65 -0.85 0.74 0.18 Cedar Rapids city Iowa 119,670 -0.04 -1.07 -0.01 -0.37 -0.62 -0.95 -0.23 -0.60 Davenport city Iowa 95,382 0.21 -0.99 -0.02 -0.26 0.03 -0.99 0.35 -0.20 Des Moines city Iowa 196,917 -0.14 -0.66 -0.11 -0.30 -0.10 -0.61 0.28 -0.14 Sioux City Iowa 78,395 0.00 -0.74 0.90 0.05 -0.31 -0.58 0.77 -0.04 Kansas City Kansas 142,341 1.22 -0.26 0.76 0.57 0.83 -0.46 0.85 0.41 Lawrence city Kansas 74,951 -0.27 -0.35 -1.81 -0.81 -0.37 -0.61 -1.44 -0.81 Olathe city Kansas 107,710 -0.90 -0.88 -0.61 -0.80 -1.18 -0.82 -0.19 -0.73 Overland Park city Kansas 161,901 -1.22 -0.82 -0.77 -0.94 -1.35 -0.79 -0.68 -0.94 Topeka city Kansas 117,326 0.19 -1.15 0.45 -0.17 0.03 -1.02 0.31 -0.23 Wichita city Kansas 354,582 0.04 -0.73 0.39 -0.10 -0.18 -0.75 0.36 -0.19 Lexington-Fayette Kentuck y 255,389 -0.20 -0.76 -0.56 -0.51 -0.29 -0.81 -0.48 -0.53 Baton Rouge city Louisiana 205,442 1.44 -0.77 0.23 0.30 0.79 -0.85 0.35 0.10 Lafayette city Louisiana 108,175 -0.10 -0.88 -0.08 -0.35 -0.22 -0.82 0.39 -0.22 Shreveport city Louisiana 192,531 1.16 -1.26 0.89 0.26 0.91 -1.24 1.10 0.26 Baltimore city Maryland 608,481 1.82 -0.50 -0.41 0.30 1.63 -0.52 0.20 0.44 Boston city Massachusetts 520,702 1.27 1.04 -2.27 0.02 0.98 0.88 -1.86 0.00 Brockton city Massachusetts 91,938 0.40 0.41 0.29 0.37 0.40 0.16 0.28 0.28 Cambridge city Massachusetts 81,260 0.04 0.61 -3.28 -0.88 -0.03 0.55 -2.67 -0.72 Fall River city Massachusetts 97,612 0.61 0.34 0.58 0.51 0.52 0.30 1.10 0.64 Lawrence city Massachusetts 82,191 2.89 2.41 -0.88 1.47 1.88 2.05 -0.24 1.23
Page A-30 City State 2005 Population 2005 Factor 1 2005 Factor 2 2005 Factor 3 2005 Equal Weight Index 2000 Factor 2 2000 Factor 3 2000 Equal Weight Index 2000 Factor 1 Lowell city Massachusetts 96,876 1.10 1.10 -0.83 0.46 0.65 0.64 -0.33 0.32 Lynn city Massachusetts 83,419 1.09 1.04 -0.67 0.49 0.66 0.78 -0.33 0.37 New Bedford city Massachusetts 84,898 1.01 0.10 0.68 0.60 0.97 0.30 0.98 0.75 Newton city Massachusetts 82,383 -1.18 -0.57 -1.89 -1.21 -1.25 -0.43 -1.68 -1.12 Quincy city Massachusetts 84,080 -0.85 0.12 -0.48 -0.40 -0.69 0.08 -0.49 -0.37 Somerville city Massachusetts 74,869 0.48 1.05 -1.91 -0.13 -0.12 0.87 -1.61 -0.28 Springfield city Massachusetts 146,948 1.96 0.04 0.42 0.81 1.32 -0.18 0.42 0.52 Worcester city Massachusetts 154,398 0.80 0.47 -1.14 0.04 0.67 0.12 -0.50 0.09 Ann Arbor city Michigan 98,743 -0.08 0.19 -3.10 -1.00 -0.33 0.01 -2.99 -1.10 Detroit city Michigan 836,056 3.13 -0.88 0.73 0.99 2.59 -1.00 0.95 0.85 Grand Rapids city Michigan 193,568 0.81 -0.32 0.03 0.17 0.42 -0.38 0.13 0.06 Lansing city Michigan 119,675 1.06 -0.55 0.04 0.19 0.57 -0.61 0.25 0.07 Livonia city Michigan 103,497 -1.42 -0.91 0.40 -0.64 -1.44 -0.95 0.62 -0.59 Pontiac city Michigan 59,472 1.64 -0.20 1.16 0.87 1.41 -0.33 1.13 0.74 Southfield city Michigan 75,053 -0.56 -0.87 0.62 -0.27 -0.71 -0.67 0.43 -0.32 Sterling Heights city Michigan 123,368 -1.05 -0.29 0.73 -0.20 -1.27 -0.47 0.83 -0.30 Troy city Michigan 83,958 -1.15 -0.47 -0.89 -0.84 -1.48 -0.44 -0.68 -0.87 Warren city Michigan 134,901 -0.48 -0.62 1.33 0.08 -0.85 -0.55 1.64 0.08 Westland city Michigan 80,284 -0.57 -0.82 1.08 -0.10 -0.90 -0.68 1.33 -0.08 Wyoming city Michigan 68,960 -0.41 -0.21 0.93 0.10 -0.75 -0.52 1.00 -0.09 Bloomington city Minnesota 80,055 -0.84 -0.64 -0.32 -0.60 -1.12 -0.62 -0.08 -0.61 Brooklyn Park city Minnesota 66,408 -0.42 -0.34 0.08 -0.22 -0.79 -0.44 0.10 -0.38 Duluth city Minnesota 76,918 0.02 -0.98 -0.93 -0.63 -0.04 -1.00 -0.57 -0.54 Minneapolis city Minnesota 350,260 0.95 0.14 -1.91 -0.27 0.75 0.07 -1.75 -0.31 Plymouth city Minnesota 68,978 -1.15 -0.74 -0.81 -0.90 -1.38 -0.78 -0.67 -0.94 Rochester city Minnesota 88,338 -0.72 -0.61 -1.12 -0.82 -0.62 -0.57 -0.92 -0.70 St. Paul city Minnesota 261,559 0.75 -0.09 -1.30 -0.21 0.61 0.01 -1.24 -0.21 Columbia city Missouri 82,103 -0.20 -0.59 -1.49 -0.76 -0.17 -0.69 -1.43 -0.76
Page A-31 City State 2005 Population 2005 Factor 1 2005 Factor 3 2005 Equal Weight Index 2000 Factor 1 2000 Factor 2 2000 Factor 3 2000 Equal Weight Index 2005 Factor 2 Independence city Missouri 111,842 -0.30 -0.76 0.95 -0.04 -0.55 -0.85 1.02 -0.12 Kansas City Missouri 440,885 0.57 -0.79 -0.16 -0.13 0.39 -0.79 0.04 -0.12 Lee's Summit city Missouri 86,357 -1.08 -1.05 -0.46 -0.86 -1.20 -1.07 -0.13 -0.80 St. Louis city Missouri 333,730 2.57 -1.03 -0.50 0.35 2.43 -1.02 -0.09 0.44 Springfield city Missouri 139,600 0.24 -0.96 0.28 -0.15 0.01 -0.98 0.43 -0.18 Lincoln city Nebraska 226,062 -0.38 -0.62 -0.70 -0.57 -0.55 -0.70 -0.51 -0.59 Omaha city Nebraska 373,215 0.23 -0.55 -0.23 -0.18 -0.15 -0.65 -0.18 -0.33 Henderson city Nevada 223,776 -1.13 -0.79 0.31 -0.53 -1.20 -0.72 0.55 -0.46 Las Vegas city Nevada 538,653 -0.58 0.29 0.66 0.12 -0.57 0.18 0.83 0.15 North Las Vegas city Nevada 165,061 -0.47 0.82 0.87 0.41 -0.35 0.99 1.33 0.66 Reno city Nevada 204,478 -0.22 0.15 -0.26 -0.11 -0.39 0.11 0.02 -0.09 Sparks city Nevada 76,405 -0.96 0.00 0.57 -0.13 -0.90 0.01 0.77 -0.04 Manchester city New Hampshire 109,308 0.12 -0.17 -0.67 -0.24 -0.12 -0.22 -0.29 -0.21 Nashua city New Hampshire 84,632 -0.69 -0.25 -0.62 -0.52 -0.69 -0.48 -0.22 -0.46 Camden city New Jersey 73,305 4.51 1.36 0.23 2.03 3.62 0.77 0.77 1.72 Clifton city New Jersey 72,667 -0.89 0.51 0.14 -0.08 -1.02 0.46 0.44 -0.04 Elizabeth city New Jersey 121,137 0.83 1.93 0.14 0.97 0.45 2.32 0.59 1.12 Jersey City New Jersey 246,335 0.91 1.40 -1.11 0.40 0.71 1.32 -0.49 0.52 Newark city New Jersey 254,217 2.02 1.65 -0.24 1.15 2.21 1.27 0.44 1.30 Passaic city New Jersey 68,422 1.76 3.56 -0.18 1.71 0.93 3.35 0.18 1.49 Paterson city New Jersey 148,353 1.40 2.09 0.13 1.20 1.17 1.92 0.61 1.23 Trenton city New Jersey 77,471 2.04 0.69 0.47 1.07 1.84 0.15 0.59 0.86 Albuquerque city New Mexico 488,133 -0.26 -0.42 0.18 -0.17 -0.31 -0.53 0.26 -0.19 Santa Fe city New Mexico 66,453 -0.12 -0.15 -0.75 -0.34 -0.40 -0.14 -0.22 -0.25 Albany city New York 78,402 2.10 -0.84 -1.45 -0.07 1.53 -0.82 -1.07 -0.12 Buffalo city New York 256,492 2.79 -1.35 0.23 0.55 2.66 -1.36 0.33 0.55 New Rochelle city New York 75,961 -0.43 0.44 -0.64 -0.21 -0.65 0.29 -0.30 -0.22
Page A-32 City State 2005 Population 2005 Factor 1 2005 Factor 3 2005 Equal Weight Index 2000 Factor 1 2000 Factor 2 2000 Factor 3 2000 Equal Weight Index 2005 Factor 2 New York city New York 7,956,113 0.79 1.28 -1.08 0.33 0.88 1.38 -0.60 0.55 Rochester city New York 189,312 2.90 -0.41 -0.17 0.77 2.42 -0.70 0.09 0.60 Syracuse city New York 132,495 2.55 -0.66 -0.56 0.44 2.02 -0.74 -0.33 0.32 Yonkers city New York 193,327 0.15 0.35 -0.10 0.14 0.17 0.43 0.08 0.23 Asheville city North Carolina 74,889 0.01 -0.81 -0.65 -0.49 0.07 -0.75 -0.27 -0.32 Cary town North Carolina 107,446 -1.17 -0.37 -1.16 -0.90 -1.33 -0.49 -1.26 -1.03 Charlotte city North Carolina 601,598 -0.16 -0.32 -0.28 -0.25 -0.38 -0.36 -0.15 -0.30 Durham city North Carolina 191,731 0.11 -0.06 -0.67 -0.21 0.03 -0.33 -0.40 -0.23 Fayetteville city North Carolina 128,777 0.43 -1.17 0.82 0.02 0.17 -1.10 0.72 -0.07 Gastonia city North Carolina 72,183 0.46 -0.40 1.14 0.40 0.25 -0.49 0.91 0.23 Greensboro city North Carolina 208,552 0.08 -0.42 0.11 -0.08 -0.24 -0.63 0.11 -0.25 High Point city North Carolina 101,852 0.26 -0.55 0.76 0.16 -0.15 -0.51 0.65 -0.01 Raleigh city North Carolina 315,249 -0.04 -0.31 -1.19 -0.51 -0.36 -0.30 -0.87 -0.51 Wilmington city North Carolina 91,207 -0.05 -0.67 -0.69 -0.47 0.30 -0.67 -0.45 -0.27 Winston-Salem city North Carolina 183,467 0.50 -0.30 0.00 0.07 0.17 -0.54 0.08 -0.09 Fargo city North Dakota 88,809 -0.33 -0.92 -0.71 -0.65 -0.60 -0.79 -0.61 -0.67 Akron city Ohio 200,181 1.12 -1.00 0.26 0.12 0.84 -1.03 0.51 0.11 Cincinnati city Ohio 287,540 2.03 -1.09 -0.54 0.13 1.75 -1.02 -0.49 0.08 Cleveland city Ohio 414,534 3.14 -1.00 0.73 0.96 2.52 -0.97 0.72 0.75 Columbus city Ohio 693,983 0.59 -0.56 -0.27 -0.08 0.23 -0.70 -0.14 -0.20 Dayton city Ohio 132,679 2.56 -1.28 0.64 0.64 1.92 -1.41 0.55 0.35 Lorain city Ohio 65,476 0.76 -0.91 1.53 0.46 0.62 -0.76 1.42 0.43 Parma city Ohio 79,708 -0.93 -0.80 0.96 -0.25 -1.10 -0.76 1.25 -0.21 Toledo city Ohio 285,937 1.14 -1.23 1.03 0.31 0.72 -1.18 1.00 0.18 Broken Arrow city Oklahoma 85,039 -0.80 -1.11 0.37 -0.51 -1.19 -1.02 0.46 -0.58 Edmond city Oklahoma 71,658 -1.02 -0.98 -0.70 -0.90 -1.12 -0.94 -0.56 -0.87 Lawton city Oklahoma 79,486 0.48 -1.09 1.15 0.18 0.31 -1.21 0.96 0.02 Norman city Oklahoma 97,484 -0.27 -0.88 -0.76 -0.64 -0.56 -0.75 -0.73 -0.68
Page A-33 City State 2005 Population 2005 Factor 1 2005 Factor 3 2005 Equal Weight Index 2000 Factor 1 2000 Factor 2 2000 Factor 3 2000 Equal Weight Index 2005 Factor 2 Oklahoma City Oklahoma 515,751 0.37 -0.71 0.46 0.04 0. 20 -0.75 0.59 0.01 Tulsa city Oklahoma 370,447 0.36 -0.66 0.27 -0.01 -0.04 -0.80 0.46 -0.13 Beaverton city Oregon 83,447 -0.81 0.05 -0.65 -0.47 -0.85 -0.04 -0.44 -0.44 Eugene city Oregon 142,716 -0.13 -0.44 -1.06 -0.54 -0.28 -0.61 -0.92 -0.60 Gresham city Oregon 95,334 0.11 0.35 0.50 0.32 -0.43 -0.09 0.45 -0.02 Hillsboro city Oregon 82,732 0.19 0.37 -0.30 0.09 -0.74 0.32 -0.10 -0.17 Medford city Oregon 73,782 -0.31 -0.62 0.31 -0.21 -0.27 -0.59 0.56 -0.10 Portland city Oregon 513,627 0.20 -0.22 -0.88 -0.30 -0.14 -0.18 -0.63 -0.31 Salem city Oregon 142,006 0.02 0.00 0.42 0.15 -0.17 -0.32 0.66 0.06 Allentown city Pennsylvania 105,231 1.46 0.17 -0.44 0.40 0.89 -0.06 0.21 0.35 Bethlehem city Pennsylvania 68,144 -0.04 -0.49 -0.23 -0.25 0.20 -0.43 -0.17 -0.13 Erie city Pennsylvania 91,423 1.02 -1.18 0.53 0.12 0.93 -1.13 0.79 0.20 Philadelphia city Pennsylvania 1,406,415 1.74 -0.29 -0.09 0.45 1.49 -0.37 0.33 0.48 Pittsburgh city Pennsylvania 284,366 1.58 -1.07 -0.76 -0.08 1.26 -1.14 -0.22 -0.03 Reading city Pennsylvania 81,302 2.89 0.62 0.50 1.34 2.04 0.12 0.63 0.93 Scranton city Pennsylvania 67,314 1.07 -0.65 0.20 0.21 0.28 -0.95 0.62 -0.02 Cranston city Rhode Island 77,025 -0.91 -0.40 0.17 -0.38 -0.90 -0.62 0.75 -0.26 Pawtucket city Rhode Island 72,896 0.92 0.34 0.13 0.47 0.50 0.49 0.53 0.50 Providence city Rhode Island 160,264 2.12 1.01 -1.46 0.56 2.04 0.86 -1.00 0.63 Warwick city Rhode Island 85,804 -1.09 -0.83 0.36 -0.52 -1.10 -0.81 0.62 -0.43 Charleston city South Carolina 109,151 0.38 -0.78 -0.98 -0.46 0.24 -0.85 -0.65 -0.42 Columbia city South Carolina 88,450 0.72 -0.75 -0.89 -0.31 0.93 -0.85 -0.43 -0.12 Sioux Falls city South Dakota 132,358 -0.29 -0.75 -0.30 -0.45 -0.61 -0.73 -0.24 -0.52 Chattanooga city Tennessee 139,158 0.81 -1.24 0.24 -0.06 0.57 -1.07 0.56 0.02 Clarksville city Tennessee 107,130 -0.48 -0.98 0.43 -0.35 -0.51 -0.99 0.86 -0.21 Knoxville city Tennessee 168,744 0.91 -0.96 -0.15 -0.06 0.66 -0.90 -0.04 -0.09 Memphis city Tennessee 642,251 1.12 -0.86 0.90 0.39 0.89 -0.93 1.17 0.38 Murfreesboro city Tennessee 83,822 -0.10 -0.59 -0.65 -0.45 -0.42 -0.65 -0.23 -0.44
Page A-34 City State 2005 Population 2005 Factor 1 2005 Factor 3 2005 Equal Weight Index 2000 Factor 1 2000 Factor 2 2000 Factor 3 2000 Equal Weight Index 2005 Factor 2 Nashville-Davidson (balance) Tennessee 522,662 0.17 -0.45 -0.30 -0.19 -0.04 -0.58 -0.09 -0.24 Abilene city Texas 105,165 0.19 -0.79 0.80 0.07 -0.11 -0.90 1.01 0.00 Amarillo city Texas 176,999 -0.01 -0.54 0.82 0.09 -0.11 -0.63 0.93 0.06 Arlington city Texas 348,965 -0.39 -0.06 0.63 0.06 -0.62 -0.18 0.48 -0.11 Austin city Texas 678,457 0.05 0.17 -0.73 -0.17 -0.25 0.24 -0.58 -0.20 Baytown city Texas 61,504 -0.05 -0.09 1.65 0.50 -0.12 -0.03 1.69 0.51 Beaumont city Texas 107,876 0.97 -0.92 0.81 0.29 0.64 -1.01 1.04 0.22 Bryan city Texas 56,277 0.59 0.27 0.76 0.54 0.09 -0.02 0.95 0.34 Carrollton city Texas 122,699 -1.10 0.26 0.42 -0.14 -1.11 -0.01 0.12 -0.33 College Station city Texas 65,370 -0.08 0.08 -2.12 -0.70 -0.40 -0.14 -1.97 -0.84 Corpus Christi city Texas 280,002 0.09 -0.43 1.09 0.25 0.00 -0.43 1.38 0.32 Dallas city Texas 1,144,946 0.76 1.22 0.41 0.80 0.38 0.97 0.44 0.60 Denton city Texas 87,766 -0.37 -0.22 -0.63 -0.41 -0.32 -0.17 -0.43 -0.31 Fort Worth city Texas 604,538 0.38 0.20 0.54 0.37 0.20 0.17 0.78 0.38 Garland city Texas 235,750 -0.27 0.72 1.17 0.54 -0.73 0.26 1.14 0.22 Grand Prairie city Texas 148,677 -0.16 0.02 1.13 0.33 -0.42 -0.04 1.27 0.27 Houston city Texas 1,941,430 0.65 1.13 0.66 0.82 0.34 0.91 0.77 0.67 Irving city Texas 212,262 -0.41 1.16 0.51 0.42 -0.57 0.85 0.35 0.21 Killeen city Texas 98,434 0.03 -0.91 0.69 -0.07 -0.13 -0.65 0.62 -0.05 Lewisville city Texas 81,484 -0.87 -0.08 0.30 -0.21 -1.08 -0.37 0.36 -0.36 Lubbock city Texas 199,789 0.03 -0.74 0.70 0.00 -0.11 -0.69 0.68 -0.04 McKinney city Texas 92,337 -1.07 -0.60 -0.33 -0.67 -0.95 -0.27 0.07 -0.38 Mesquite city Texas 126,895 -0.82 0.02 1.62 0.28 -0.93 -0.58 1.47 -0.02 Midland city Texas 100,79 9 -0.27 -0.54 1.11 0.10 -0.38 -0.68 1.02 -0.01 Odessa city Texas 94,329 0.08 -0.65 1.91 0.45 0.05 -0.62 2.13 0.52 Pasadena city Texas 150,180 0.14 1.05 1.66 0.95 -0.13 0.70 1.85 0.81 Plano city Texas 251,648 -1.13 -0.31 -0.68 -0.71 -1.30 -0.49 -0.72 -0.84 Richardson city Texas 107,892 -1.19 -0.17 -0.39 -0.58 -1.20 -0.18 -0.38 -0.59
Page A-35 City State 2005 Population 2005 Factor 1 2005 Factor 3 2005 Equal Weight Index 2000 Factor 1 2000 Factor 2 2000 Factor 3 2000 Equal Weight Index 2005 Factor 2 Round Rock city Texas 81,639 -0.74 -0.65 0.18 -0.40 -1.14 -0.51 0.33 -0.44 San Angelo city Texas 82,293 -0.03 -0.83 1.11 0.08 -0.04 -0.71 1.23 0.16 San Antonio city Texas 1,202,223 0.17 0.02 0.77 0.32 0.05 -0.04 1.05 0.35 Tyler city Texas 87,687 0.50 -0.62 0.70 0.19 0.09 -0.55 0.71 0.08 Waco city Texas 107,146 0.69 -0.14 0.67 0.41 0.69 -0.35 0.89 0.41 Wichita Falls city Texas 88,861 0.04 -0.67 0.48 -0.05 -0.12 -0.84 0.88 -0.03 Ogden city Utah 79,171 1.00 0.04 -0.40 0.21 0.64 -0.20 0.06 0.17 Orem city Utah 85,616 -0.69 -0.63 -0.07 -0.46 -0.85 -0.58 0.07 -0.46 Provo city Utah 101,164 0.39 0.18 -1.79 -0.41 0.24 -0.04 -1.62 -0.47 Salt Lake City Utah 182,670 0.41 0.10 -1.06 -0.18 0.23 0.14 -0.97 -0.20 Sandy city Utah 88,189 -1.13 -1.07 0.04 -0.72 -1.29 -0.96 0.10 -0.71 West Jordan city Utah 101,626 -1.14 -0.59 0.52 -0.40 -1.10 -0.77 0.72 -0.38 Alexandria city Virginia 133,479 -0.85 0.64 -1.93 -0.72 -0.51 0.60 -1.72 -0.54 Chesapeake city Virginia 214,835 -1.00 -0.94 0.56 -0.46 -0.92 -0.91 0.63 -0.40 Hampton city Virginia 133,584 -0.05 -0.88 0.28 -0.21 -0.31 -0.88 0.59 -0.20 Newport News city Virginia 176,591 0.01 -0.68 0.05 -0.21 -0.01 -0.73 0.32 -0.14 Norfolk city Virginia 206,172 0.76 -0.44 -0.35 -0.01 0.70 -0.49 -0.01 0.07 Portsmouth city Virginia 95,183 0.53 -0.73 0.30 0.03 0.35 -0.73 0.80 0.14 Richmond city Virginia 180,757 0.98 -0.49 -0.55 -0.02 1.22 -0.52 -0.51 0.06 Roanoke city Virginia 90,074 0.43 -1.06 0.48 -0.05 0.50 -0.91 0.75 0.12 Suffolk city Virginia 77,922 -0.39 -1.00 0.45 -0.32 -0.26 -0.84 0.96 -0.05 Virginia Beach city Virginia 430,856 -0.86 -0.75 0.04 -0.53 -0.98 -0.74 0.27 -0.48 Bellevue city Washington 114,748 -0.98 0.37 -1.64 -0.75 -1.12 0.12 -1.27 -0.76 Bellingham city Washington 69,057 -0.11 -0.44 -1.44 -0.66 -0.19 -0.50 -0.94 -0.54 Everett city Washington 88,850 0.04 0.13 -0.11 0.02 -0.07 -0.01 0.15 0.03 Kent city Washington 84,979 -0.20 0.43 -0.09 0.05 -0.32 0.09 0.08 -0.05 Seattle city Washington 536,946 -0.31 -0.04 -1.77 -0.71 -0.32 -0.11 -1.55 -0.66 Spokane city Washington 192,777 0.25 -0.81 -0.09 -0.22 0.20 -0.74 -0.11 -0.22
Page A-36 City State 2005 Population 2005 Factor 1 2005 Factor 2 2005 Factor 3 2005 Equal Weight Index 2000 Factor 1 2000 Factor 2 2000 Factor 3 2000 Equal Weight Index Tacoma city Washington 191,934 0.45 -0.17 -0.38 -0.03 0.35 -0.24 -0.23 -0.04 Vancouver city Washington 155,488 0.17 -0.11 -0.08 -0.01 -0.31 -0.09 0.06 -0.11 Yakima city Washington 79,517 0.68 0.26 1.24 0.73 0.61 0.15 1.55 0.77 Green Bay city Wisconsin 94,242 0.09 -0.44 0.06 -0.09 -0.29 -0.55 0.10 -0.24 Kenosha city Wisconsin 95,440 -0.36 -0.58 0.63 -0.10 -0.47 -0.64 0.80 -0.10 Madison city Wisconsin 203,704 -0.04 -0.32 -2.02 -0.79 -0.27 -0.36 -2.04 -0.89 Milwaukee city Wisconsin 556,948 1.72 -0.21 -0.12 0.46 1.29 -0.34 0.12 0.36 Waukesha city Wisconsin 62,690 -1.05 -0.91 -0.06 -0.68 -1.09 -0.72 0.21 -0.53
Table A.10. Adjusted Needs Index for 234 Cities City State Population Standardized Equal Weight Index Standardized Real Fiscal Capacity Index Adjusted Needs Index Equal Weight Index Rank Adjusted Needs Index Rank Birmingham city Alabama 222,154 1.00 0.71 0.29 34 98 Huntsville city Alabama 158,618 -0.87 1.40 -2.27 188 214 Mobile city Alabama 193,332 0.30 0.60 -0.30 80 130 Anchorage municipality Alaska 266,281 -0.95 -0.85 -0.10 198 114 Chandler city Arizona 225,725 -0.93 -0.98 0.05 194 109 Me sa city Arizona 442,445 -0.10 -1.29 1.18 117 67 Peoria city Arizona 141,941 -0.90 -1.32 0.43 190 89 Phoenix city Arizona 1,377,980 0.61 -0.96 1.57 55 45 Scottsdale city Arizona 215,933 -1.69 1.54 -3.23 227 227 Tempe city Arizona 166,171 -0.90 -0.05 -0.85 191 162 Tucson city Arizona 507,362 0.29 -1.07 1.36 83 56 Fort Smith city Arkansas 81,054 0.22 2.04 -1.82 88 200 Little Rock city Arkansas 176,924 -0.76 2.02 -2.78 182 221 Anaheim city California 329,483 1.17 -0.93 2.10 26 26 Bakersfield city California 286,316 0.37 -1.10 1.48 76 51 Berkeley city California 90,432 -2.05 0.07 -2.12 233 207 Burbank city California 100,053 -0.52 0.90 -1.42 165 193 Chula Vista city California 212,954 0.29 -0.75 1.05 82 72 Concord city California 116,782 -0.25 -0.66 0.40 135 92 Corona city California 162,410 -0.08 -0.48 0.40 116 93 El Cajon city California 92,507 0.50 -0.85 1.35 63 58 Escondido city California 133,017 0.67 -0.57 1.24 50 65 Fairfield city California 102,642 -0.26 -1.31 1.05 137 71 Fullerton city California 142,064 -0.35 -0.65 0.30 143 96 Garden Grove city California 192,345 1.58 -1.44 3.01 17 11 Glendale city California 194,620 0.44 -0.61 1.05 71 70 Hayward city California 135,474 0.46 -0.82 1.28 68 63 Hemet city California 77,076 1.06 -1.49 2.55 30 17 Inglewood city California 120,204 1.81 -1.81 3.62 14 7 Long Beach city California 463,956 0.83 -1.29 2.12 41 25 Los Angeles city California 3,731,437 1.29 -0.96 2.25 24 22 Modesto city California 202,971 0.25 -0.93 1.18 85 68 Oakland city California 373,910 0.72 -0.99 1.71 47 39 Oceanside city California 162,259 -0.39 -0.70 0.31 150 95 Ontario city California 156,679 1.55 -0.73 2.28 18 21 Oxnard city California 178,871 1.81 -1.13 2.93 15 12 Pasadena city California 129,400 -0.58 0.34 -0.92 171 168 Pomona city California 161,257 2.07 -1.69 3.76 10 5 Redwood City California 81,195 0.19 1.02 -0.83 93 159 Richmond city California 96,648 0.61 -1.04 1.65 53 42 Riverside city California 294,059 0.41 -0.99 1.40 74 54 Roseville city California 108,848 -1.16 0.29 -1.45 212 195 Sacramento city California 445,287 0.47 -0.80 1.26 66 64 Page A-37
City State Population Standardized Equal Weight Index Standardized Real Fiscal Capacity Index Adjusted Needs Index Equal Weight Index Rank Adjusted Needs Index Rank Salinas city California 156,950 2.52 -1.24 3.76 6 4 San Bernardino city California 204,552 2.12 -1.53 3.65 9 6 San Buenaventura (Ventura) city California 100,154 -0.90 0.18 -1.08 189 179 San Diego city California 1,208,331 -0.52 0.32 -0.84 164 161 San Francisco city California 719,077 -0.62 0.26 -0.88 173 164 San Jose city California 887,330 0.00 -0.47 0.46 107 87 Santa Ana city California 302,302 3.04 -1.24 4.29 2 1 Santa Barbara city California 90,708 -0.76 1.45 -2.20 181 212 Santa Clara city California 102,204 -0.63 -0.63 -0.01 177 110 Santa Monica city California 82,777 -1.81 2.52 -4.33 229 231 Santa Rosa city California 146,500 -0.35 0.48 -0.83 142 160 Stockton city California 278,515 1.13 -1.34 2.48 27 19 Vallejo city California 115,657 -0.07 -0.86 0.79 114 77 Arvada city Colorado 104,766 -1.02 -0.92 -0.10 203 116 Aurora city Colorado 291,317 0.26 -1.07 1.33 84 60 Boulder city Colorado 83,432 -2.03 2.54 -4.57 232 232 Colorado Springs city Colorado 376,985 -0.90 -0.31 -0.59 192 146 Denver city Colorado 545,198 -0.13 0.35 -0.48 120 139 Fort Collins city Colorado 122,297 -1.61 -0.41 -1.20 225 185 Greeley city Colorado 82,836 -0.12 -0.54 0.42 119 91 Lakewood city Colorado 142,434 -0.67 -0.23 -0.44 178 137 Longmont city Colorado 76,181 -0.35 -0.11 -0.24 141 129 Westminster city Colorado 99,305 -0.97 -0.78 -0.19 200 123 Washington city District of Columbia 515,118 -0.52 0.72 -1.24 163 186 Cape Coral city Florida 134,388 -0.37 -0.50 0.12 147 107 Clearwater city Florida 108,382 -0.40 1.24 -1.64 153 197 Fort Lauderdale city Florida 141,307 -0.21 1.58 -1.79 130 199 Gainesville city Florida 100,879 -0.42 -0.07 -0.36 155 134 Hollywood city Florida 138,412 0.07 -0.48 0.55 99 84 Jacksonville city Florida 768,537 -0.61 -0.56 -0.05 172 112 Largo city Florida 71,269 -0.38 -0.22 -0.16 148 119 Melbourne city Florida 76,373 -0.53 -0.34 -0.19 166 124 Miami Beach city Florida 84,086 0.72 0.11 0.61 46 81 Miami city Florida 361,701 2.82 -0.12 2.93 3 13 Orlando city Florida 221,299 0.15 1.28 -1.13 94 181 Pompano Beach city Florida 94,892 0.45 0.16 0.28 69 99 St. Petersburg city Florida 232,960 -0.38 0.42 -0.80 149 155 Tallahassee city Florida 141,148 -1.31 0.08 -1.38 215 191 Tampa city Florida 312,855 0.01 1.04 -1.03 104 176 Page A-38
City State Population Standardized Equal Weight Index Standardized Real Fiscal Capacity Index Adjusted Needs Index Equal Weight Index Rank Adjusted Needs Index Rank West Palm Beach city Florida 86,804 0.49 1.64 -1.15 65 182 Atlanta city Georgia 394,929 0.14 2.32 -2.18 95 211 Savannah city Georgia 117,478 0.31 1.34 -1.03 79 175 Honolulu CDP Hawaii 362,252 -0.63 0.29 -0.92 176 169 Boise City Idaho 191,667 -1.08 1.41 -2.49 209 218 Aurora city Illinois 170,490 0.43 -1.12 1.56 72 46 Chicago city Illinois 2,701,926 0.61 -0.63 1.24 54 66 Evanston city Illinois 62,258 -2.17 0.92 -3.09 234 226 Joliet city Illinois 128,090 -0.14 -1.08 0.94 121 75 Rockford city Illinois 139,173 0.42 0.18 0.24 73 102 Springfield city Illinois 110,262 -0.86 0.19 -1.05 187 177 Bloomington city Indiana 55,406 -1.32 0.83 -2.15 217 210 Evansville city Indiana 110,708 0.03 0.63 -0.60 103 147 Fort Wayne city Indiana 219,346 -0.05 -0.25 0.19 110 104 Indianapolis city (balance) Indiana 765,310 -0.18 0.26 -0.44 125 136 South Bend city Indiana 97,070 0.52 0.26 0.26 62 100 Cedar Rapids city Iowa 119,670 -0.76 1.19 -1.95 183 204 Davenport city Iowa 95,382 -0.55 0.77 -1.32 169 189 Des Moines city Iowa 196,917 -0.63 0.26 -0.89 175 165 Kansas City Kansas 142,341 1.07 -1.38 2.45 29 20 Overland Park city Kansas 161,901 -1.86 1.17 -3.03 230 224 Topeka city Kansas 117,326 -0.37 0.44 -0.82 146 157 Wichita city Kansas 354,582 -0.23 -0.02 -0.21 132 127 Lexington-Fayette Kentucky 255,389 -1.02 0.63 -1.65 202 198 Baton Rouge city Louisiana 205,442 0.54 0.40 0.14 61 106 Lafayette city Louisiana 108,175 -0.73 0.54 -1.27 180 187 Shreveport city Louisiana 192,531 0.46 -0.27 0.73 67 79 Baltimore city Maryland 608,481 0.54 0.37 0.17 60 105 Boston city Massachusetts 520,702 -0.01 1.42 -1.43 108 194 Brockton city Massachusetts 91,938 0.67 -0.35 1.02 51 74 Cambridge city Massachusetts 81,260 -1.74 3.17 -4.91 228 233 Fall River city Massachusetts 97,612 0.95 -0.78 1.74 35 36 Lawrence city Massachusetts 82,191 2.81 -1.01 3.82 4 3 Lowell city Massachusetts 96,876 0.85 -0.70 1.54 39 47 Lynn city Massachusetts 83,419 0.90 -0.51 1.41 36 53 New Bedford city Massachusetts 84,898 1.12 -0.78 1.90 28 30 Quincy city Massachusetts 84,080 -0.82 0.14 -0.97 185 171 Somerville city Massachusetts 74,869 -0.28 -0.24 -0.04 138 111 Springfield city Massachusetts 146,948 1.52 -0.41 1.93 20 28 Worcester city Massachusetts 154,398 0.04 0.25 -0.21 101 126 Ann Arbor city Michigan 98,743 -1.98 0.84 -2.82 231 222 Detroit city Michigan 836,056 1.89 -1.46 3.34 11 9 Grand Rapids city Michigan 193,568 0.29 -0.17 0.46 81 86 Page A-39
City State Population Standardized Equal Weight Index Standardized Real Fiscal Capacity Index Adjusted Needs Index Equal Weight Index Rank Adjusted Needs Index Rank Sterling Heights city Michigan 123,368 -0.43 -0.08 -0.35 156 133 Warren city Michigan 134,901 0.11 -0.45 0.56 96 83 Westland city Michigan 80,284 -0.24 -1.11 0.87 133 76 Bloomington city Minnesota 80,055 -1.21 2.60 -3.81 213 230 Duluth city Minnesota 76,918 -1.26 0.78 -2.05 214 205 Minneapolis city Minnesota 350,260 -0.57 0.85 -1.42 170 192 Rochester city Minnesota 88,338 -1.62 0.87 -2.49 226 217 St. Paul city Minnesota 261,559 -0.45 0.16 -0.62 159 148 Columbia city Missouri 82,103 -1.52 0.85 -2.37 223 215 Independence city Missouri 111,842 -0.12 -0.41 0.29 118 97 Kansas City Missouri 440,885 -0.29 0.50 -0.79 139 153 Springfield city Missouri 139,600 -0.32 1.28 -1.61 140 196 St. Louis city Missouri 333,730 0.63 0.19 0.44 52 88 Lincoln city Nebraska 226,062 -1.14 -0.50 -0.64 211 149 Omaha city Nebraska 373,215 -0.40 0.17 -0.57 151 144 Henderson city Nevada 223,776 -1.08 -0.96 -0.12 208 118 Las Vegas city Nevada 538,653 0.20 -1.11 1.31 91 61 Reno city Nevada 204,478 -0.26 -0.07 -0.19 136 122 Manchester city New Hampshire 109,308 -0.50 1.41 -1.92 162 202 Nashua city New Hampshire 84,632 -1.05 2.25 -3.30 206 228 Clifton city New Jersey 72,667 -0.20 0.31 -0.50 128 140 Elizabeth city New Jersey 121,137 1.84 -0.73 2.56 12 16 Jersey City New Jersey 246,335 0.74 -0.62 1.36 45 57 Newark city New Jersey 254,217 2.18 -0.70 2.89 8 14 Passaic city New Jersey 68,422 3.28 -0.96 4.24 1 2 Paterson city New Jersey 148,353 2.29 -0.98 3.28 7 10 Albuquerque city New Mexico 488,133 -0.36 -0.20 -0.17 144 120 Santa Fe city New Mexico 66,453 -0.70 1.44 -2.14 179 208 Albany city New York 78,402 -0.17 0.63 -0.80 124 154 Buffalo city New York 256,492 1.03 -1.18 2.21 32 23 New York city New York 7,956,113 0.60 -1.03 1.64 56 43 Rochester city New York 189,312 1.46 -0.28 1.74 22 37 Syracuse city New York 132,495 0.82 -0.56 1.39 42 55 Yonkers city New York 193,327 0.22 -1.29 1.51 87 48 Asheville city North Carolina 74,889 -0.98 1.85 -2.84 201 223 Charlotte city North Carolina 601,598 -0.53 0.36 -0.89 168 166 Durham city North Carolina 191,731 -0.45 0.13 -0.58 157 145 Fayetteville city North Carolina 128,777 0.01 -0.34 0.34 105 94 Greensboro city North Carolina 208,552 -0.19 0.84 -1.03 126 174 Raleigh city North Carolina 315,249 -1.04 0.32 -1.36 204 190 Wilmington city North Carolina 91,207 -0.95 0.99 -1.94 199 203 Winston-Salem city North Carolina 183,467 0.09 0.20 -0.11 97 117 Page A-40
City State Population Standardized Equal Weight Index Standardized Real Fiscal Capacity Index Adjusted Needs Index Equal Weight Index Rank Adjusted Needs Index Rank Fargo city North Dakota 88,809 -1.31 1.72 -3.03 216 225 Akron city Ohio 200,181 0.20 -0.45 0.65 90 80 Cincinnati city Ohio 287,540 0.21 0.72 -0.50 89 141 Cleveland city Ohio 414,534 1.81 -0.39 2.21 13 24 Columbus city Ohio 693,983 -0.19 -0.14 -0.05 127 113 Dayton city Ohio 132,679 1.20 -0.24 1.44 25 52 Toledo city Ohio 285,937 0.57 -0.46 1.03 59 73 Oklahoma City Oklahoma 515,751 0.03 -0.22 0.26 102 101 Tulsa city Oklahoma 370,447 -0.06 0.49 -0.55 113 143 Eugene city Oregon 142,716 -1.09 -0.13 -0.96 210 170 Gresham city Oregon 95,334 0.58 -1.09 1.67 57 41 Portland city Oregon 513,627 -0.62 0.12 -0.74 174 152 Salem city Oregon 142,006 0.24 -0.52 0.77 86 78 Bethlehem city Pennsylvania 68,144 -0.53 0.15 -0.69 167 150 Erie city Pennsylvania 91,423 0.19 -0.01 0.20 92 103 Philadelphia city Pennsylvania 1,406,415 0.83 -0.74 1.58 40 44 Pittsburgh city Pennsylvania 284,366 -0.20 0.99 -1.20 129 183 Reading city Pennsylvania 81,302 2.55 -0.28 2.83 5 15 Scranton city Pennsylvania 67,314 0.36 0.30 0.06 77 108 Cranston city Rhode Island 77,025 -0.77 -0.47 -0.31 184 131 Pawtucket city Rhode Island 72,896 0.86 -1.12 1.98 37 27 Providence city Rhode Island 160,264 1.04 -0.83 1.87 31 32 Warwick city Rhode Island 85,804 -1.05 -0.14 -0.90 205 167 Charleston city South Carolina 109,151 -0.93 4.35 -5.28 195 234 Sioux Falls city South Dakota 132,358 -0.91 1.36 -2.27 193 213 Chattanooga city Tennessee 139,158 -0.16 0.91 -1.07 122 178 Knoxville city Tennessee 168,744 -0.17 1.03 -1.20 123 184 Memphis city Tennessee 642,251 0.71 -0.58 1.29 48 62 Nashville- Davidson (balance) Tennessee 522,662 -0.42 0.12 -0.53 154 142 Arlington city Texas 348,965 0.08 -1.27 1.35 98 59 Austin city Texas 678,457 -0.36 -0.04 -0.33 145 132 Corpus Christi city Texas 280,002 0.44 -1.06 1.50 70 49 Dallas city Texas 1,144,946 1.51 -0.24 1.74 21 35 Fort Worth city Texas 604,538 0.68 -1.05 1.73 49 38 Garland city Texas 235,750 1.00 -1.55 2.55 33 18 Houston city Texas 1,941,430 1.54 -0.36 1.91 19 29 Irving city Texas 212 ,262 0.77 0.87 -0.10 43 115 Laredo Texas 199,789 -0.05 -1.88 1.84 109 33 Mesquite city Texas 126,895 0.49 -1.39 1.88 64 31 Pasadena city Texas 150,180 1.79 -1.56 3.36 16 8 San Antonio city Texas 1,202,223 0.58 -1.24 1.82 58 34 Tyler city Texas 87,687 0.33 0.51 -0.18 78 121 Waco city Texas 107,146 0.75 -0.40 1.15 44 69 Page A-41
Page A-42 City State Population Standardized Equal Weight Index Standardized Real Fiscal Capacity Index Adjusted Needs Index Equal Weight Index Rank Adjusted Needs Index Rank Ogden city Utah 79,171 0.38 -0.09 0.47 75 85 Orem city Utah 85,616 -0.94 -0.12 -0.83 197 158 Provo city Utah 101,164 -0.83 -0.01 -0.82 186 156 Salt Lake City Utah 182,670 -0.40 1.66 -2.06 152 206 Alexandria city Virginia 133,479 -1.43 1.20 -2.63 221 219 Chesapeake city Virginia 214,835 -0.94 0.07 -1.01 196 173 Hampton city Virginia 133,584 -0.46 -0.23 -0.23 160 128 Newport News city Virginia 176,591 -0.45 0.24 -0.69 158 151 Norfolk city Virginia 206,172 -0.06 0.41 -0.47 111 138 Richmond city Virginia 180,757 -0.08 1.02 -1.10 115 180 Virginia Beach city Virginia 430,856 -1.06 0.24 -1.30 207 188 Bellevue city Washington 114,748 -1.49 2.29 -3.77 222 229 Bellingham city Washington 69,057 -1.33 0.81 -2.14 218 209 Everett city Washington 88,850 0.00 0.85 -0.85 106 163 Kent city Washington 84,979 0.05 0.26 -0.21 100 125 Seattle city Washington 536,946 -1.41 1.28 -2.69 220 220 Spokane city Washington 192,777 -0.46 -0.09 -0.37 161 135 Vancouver city Washington 155,488 -0.06 -0.48 0.42 112 90 Yakima city Washington 79,517 1.37 -0.32 1.69 23 40 Green Bay city Wisconsin 94,242 -0.22 0.76 -0.98 131 172 Kenosha city Wisconsin 95,440 -0.24 -0.81 0.57 134 82 Madison city Wisconsin 203,704 -1.58 0.86 -2.44 224 216 Milwaukee city Wisconsin 556,948 0.86 -0.64 1.49 38 50 Waukesha city Wisconsin 62,690 -1.35 0.49 -1.84 219 201
Appendix B—Using Regression Analysis to Develop a Community Needs Index This report uses factor analysis to identify common themes in the needs of cities and to measure the needs represented by those common themes. Chapter 3 points out that factor analysis has several conceptual weaknesses. The main flaws of factor analysis are: • It assumes the existence of unobservable factors that cause the observed needs. The existence of these factors cannot be proved. • There are no definite ways to determine how many factors are at work or to select among all the possible “rotations” of the factors. We resorted to some common rules of thumb and judgment to select the factors used to analyze cities needs in Chapter 4. • Judgment must be used to choose weights to combine the factors into a single- value index of need. • The individual factor scores and any index created by combining them are unitless measures of need—that is, a factor score of 1.0 says that that city is one standard deviation above the average of all cities on that factor; it does not provide any information on how serious a 1.0 score is. At the project’s Orientation Meeting on October 12, 2006, George Galster suggested an alternative to factor analysis that would avoid all of these weaknesses. This appendix describes this alternative approach, explains how we implemented it, and presents the results. The alternative approach produced some interesting findings, but we were unable to translate the results into a comprehensive assessment of community needs. The problem was that some community needs, such as the lack of affordable housing, cannot be measured using the alternative approach. For this reason, we used factor analysis to measure needs and relegated our experience with the alternative approach to this appendix. The concluding section of the appendix contains a summary of the insights we gained from the alternative approach. B.1. General Concept, Model Specification, and Difficulties B.1.1. General Concept The alternative a pproach is derived from, but is not the same as, two methods employed in economics. The first is a methodology used to explain price differences across similar products with different characteristics. This “hedonic” model was originally developed to explain the variation in prices of cars that have different features. The model has since Page B-1
Page B-2 been applied frequently to explain differences in the price of individual houses based on such features as house size; the condition of the house; lot size; location with respect to employment centers, good schools, and shopping; amenities such as air conditioning, garage, and decks; and the quality of the neighborhood. These applications use data on the price and the features of specific houses. The second method uses variations in median house prices across metropolitan areas to measure the market valuation of amenities associated with living in these various places, such as climate, culture, and sports teams. This method uses housing prices and amenities measured at the metropolitan scale. The proposed methodology draws from the intuitions and conclusions of both prior strands of work, but at an intermediate spatial scale. It attempts to explain variation in the median price of houses across census tracts in various cities. It uses census information on various characteristics of the census tracts and the information produced by this study on the characteristics of the cities in which the tracts are located. Drawing upon the central insight of both prior strands of work—that the housing market effectively capitalizes the value placed by consumers on characteristics of the geography surrounding the dwelling—this approach seeks to find how aspects of the political jurisdiction that have been associated traditionally with “distress” are capitalized, controlling for characteristics of the dwellings, neighborhoods, and larger metropolitan area. The “hedonic” model starts from recognizing two facts: (1) house prices vary significantly across the country, and (2) the value of a house depends upon the characteristics of the house, the characteristics of the immediate neighborhood in which the house is located, and the characteristics of the political jurisdiction and the broader metropolitan market in which the house is located. 49 To the extent that a city has problems, the value of houses in that city should be lower than the value of comparable housing in comparable neighborhoods in a city without such problems. For example, consider two identical neighborhoods in two different cities in the same metropolitan area. Because the neighborhoods have similar housing and are located in the same housing market, one would expect that the median price of houses in the two neighborhoods would be equal unless one city was a more desirable place to live than the other city. The goal is to use regression analysis to determine how much effect particular city-level problems have on the value of homes in those cities. If one can successfully isolate these effects, then one has a direct measure of the impact of the problems we have been considering. The measure is a “dollars and cents” measure, and therefore the effects of different problems can be added together. In this way, one could determine the combined effects of a number of city-level problems and could also assess which problems have the greatest negative impact. The key advantage here is that the respective coefficients measuring these effects provide guidance about how the market evaluates the various problematic characteristics of the city, without the imposition of value judgments from the researcher. 49 While it is a misnomer to call the techniques employed in this appendix hedonic models, we shall use the “hedonic” label for convenience and because the reasoning behin d the model draws on the hedonic literature.
Page B-3 From the 2000 decennial census, we have information on owners’ assessments of the value of their homes. 50 In particular, we know the median value (as reported by owners) of the owner-occupied homes in 26,287 census tracts in the 370 cities for which we have data on our 24 needs indicators. B.1.2. Model Specification 51 The house prices (P ijk ) in the ith neighborhood located in the jth political jurisdiction in the kth metro area is a function of the myriad characteristics (C) of these various scales of geography. For example, C i includes housing characteristics, C j includes municipal tax rate, and C k includes climate and housing supply elasticity. Symbolically: [1] P ijk = f(C i , C j , C k ) But because many (not all) of the characteristics of jurisdictions (demographic, economic, social, etc.) are mathematical summations of the corresponding characteristics of the constituent neighborhoods, we can write: [2] C j = g(C i ) Other characteristics of the jurisdiction (like tax rate) can be symbolized C j* . Thus, [1] can be written: [3] P ijk = f(C i , [Cj * , g(C i )] , C k ) From the perspective of a community needs index, the only level of geography that is directly relevant is the jurisdictional j level. Certainly, variations among neighborhoods in j may matter for needs, but such variation should be captured with variables measured at the community j scale to be operational, e.g., percentage of tracts in a community exceeding 40 percent poverty. If we were to run a regression using median census tract housing prices as the dependent variable for a large set of tracts across j jurisdictions and k different metropolitan areas, C k could be measured by a dummy variable for each metropolitan area. [C j* , g(C i )] will be measured with the variables used for input into the factor analysis; coefficients of these variables would be the key item of interest that weight the various attributes of the jurisdiction. The Ci term is trickiest. We certainly can measure many tract characteristics using census data, but others of interest (like crime rates) we cannot for the entire nation. So, there will be some omitted variables of interest. It is likely, however, that many of these will be highly correlated with others that we can measure. 50 The ACS will provide same information at the census-tract level beginning in 2010. 51 George Galster drafted this formal presentation of the hedonic-like model.
Page B-4 The measure of community needs would be the sum of the product of [C j* , g(C i )] and its coefficients. Its interpretation is straightforward. Community conditions in city A reduce the median price of houses in city A by X dollars. B.1.3. Difficulties The statistical problem is how to relate the variation in median home values to the information we have about the tract and the local market. 52 Because city-level problems are likely to have less impact on the values of the houses in a tract than the characteristics of the houses themselves and their neighborhood, it is very important to have good information on the houses and the tract. This is the first problem that must be overcome. A second problem is the possibility that not all city-level problems will be associated with lower house values. An example of such a situation is housing affordability. The needs indicators in Table 1 include several measures of possible housing market problems. While the lack of affordable rental housing (LACKAFFDRENTALS) given the city’s income level is definitely a legitimate city-level problem, it is likely to be associated with high housing prices, not low housing prices. There are also situations where the functioning of private markets may run counter to public perception of what is desirable and what is not. For example, city-level diversity in terms of income and racial or ethnic composition is considered desirable from the perspective of public policy, but some homebuyers may pay a premium to live in cities that are homogenous in terms of income or race or ethnicity. If enough homebuyers value racial and income homogeneity, then the implicit prices derived from the hedonic model will represent the market value but not the social value of diversity. B.2. Implementing the “Hedonic” Approach for City-Level Needs Indicators B.2.1. Choice of Database As noted above, we have data on median housing values for 26,287 census tracts. After eliminating tracts with missing values for various variables, this left 21,375 for our first regression. In reviewing earlier versions of this model, some HUD staff members expressed concern about using median values in the regression because housing values can vary significantly within a tract. At HUD’s suggestion, we experimented with eliminating tracts where housing values vary greatly within the tract. We dropped roughly 25 percent of the tracts—those with the greatest variation relative to the 52 The Final Work Plan for this project contains the formal presentation of this model.
Page B-5 median. 53 This resulted in a database of 20,485 tracts, of which we used 16,096 in the second regression. The two regressions produced similar results, and neither set of results was clearly superior to the other. 54 Table B.10 at the end of this appendix contains the complete results from the regression using the full set of tracts, while Table B.11 contains the complete results from the regression using the tracts with less dispersion in house values. The discussion in this chapter will focus on the regression based on the tracts with less dispersion—that is, the regression reported in Table B.11. Both models fit the data well. The regression using the restricted data set explains 77 percent of the variation in median house prices, and the probability that the reduction in variance is due to chance is less than 1 in 10,000. B.2.2. The Tract-Level Variables To isolate the effects of city-level needs indicators on median house prices, we used 20 control variables defined at the tract level. Table B.1 lists the tract-level variables. In choosing these variables, we looked first for information that might describe the characteristics of the houses whose median value we are trying to explain. The decennial census has two tables that contain relevant information: a distribution of owner-occupied homes by the number of rooms in the home and a distribution of owner-occupied housing by the year the unit was built. We use the median number of rooms and create five variables to characterize the age of the owner-occupied housing in the tract. 55 Next, we use the decennial census to provide information about the housing and people in the tract that might relate to the value of owner-occupied housing. Table B.1 contains variables that describe the tenure pattern and the type of structures in the tract and a variable that measures the proportion of overcrowded units in the tract. The decennial census provides information on the poverty rate in the tract and the racial and ethnic composition of the tract. It also provides information at the tract level on populations that we include in the city-level needs indicators—namely, single-parent families, recent immigrants, and linguistically isolated households. 53 We calculated the ratio of (3 rd quartile value – 1 st quartile value)/median and eliminated all tracts where this ratio equaled or exceeded 0.56. The distribution of this ratio had a first quartile of 0.30, a median of 0.41, and a third quartile of 0.56. 54 Both regressions use the natural log of the median house value as the dependent variable. This means that independent variables have a multiplicative effect on median house price—that is, if a variable has a value of 1.0 and its coefficient is 1.04, then the impact of that variable is to increase the median value of the homes in that tract by 4 percent. Those with positive coefficients increase median house price, while those with negative coefficients decrease median house price. 55 Our age-of-housing variables segment age into six periods. We use only five of them, because in regression analysis if there is a group of mutually exclusive categories, one must category must serve as the point of comparison. With a set of categorical variables, the coefficients tell how much impact each of the included variables has compared to the excluded variable, for example, by what percentage the median value of a units built after 1989 is greater or less the median value of a units built in the 1970s, the omitted age category.
Page B-6 Table B.1. Tract-Level Variables Used in Regressions Variable Name Explanation Characteristics of owner-occupied housing in the tract ROOMS Median number of rooms in owner-occupied houses. BUILTB50 Percent of owner-occupied was built before 1950 BUILT50S Percent of owner-occupied was built the 1950s BUILT60S Percent of owner-occupied was built the 1960s BUILT80S Percent of owner-occupied was built the 1970s BUILTA89 Percent of owner-occupied was built after 1989 Characteristics of the tract ORATE Percent of the occupied units that are owner-occupied PCTSFDETACHED Percent of the units that are in single-family detached structures PCT5PLUSUNITS Percent of the units that are in structures containing 5 or more units OVERCROWD2000_TR Percent of households in the tract living in units where the number of person per room is 1.01 or greater. HISPAN Percent of the population that is Hispanic NHBLACK Percent of the population that is non-Hispanic Black NHOTHER Percent of the population that is non-Hispanic and not White alone or Black alone POV10_19 The poverty rate in the tract is greater than or equal to 10 percent and less than 20 percent. POV20_29 The poverty rate in the tract is greater than or equal to 20 percent and less than 30 percent. POV30_39 The poverty rate in the tract is greater than or equal to 10 percent and less than 40 percent. POV40PLUS The poverty rate in the tract is greater than or equal to 40 percent. SGLPRNTFAM_TR Percent of families in the tract that are single parent-headed with own children under 18. RCNTIMMIG_TR Percent of household population in the tract that is foreign born and entered the United States within the last 15 years. LINGISOL_TR Percent of households in the tract that are “linguistically isolated” according to the definition in Table 1. Table B.2 shows that the tract-level variables perform very well. Median housing value is positively and significantly related to the median number of rooms. All the year-built coefficients are statistically significant, except the coefficient for houses built in the 1960s. The sign and size of the coefficients suggest that older housing is less valuable than more recently built housing, with the exception that houses built before 1950 are more valuable than those built in the 1950s. 56 56 One possible explanation for this result is that “being built before 1950” proxies for other features of a housing unit that are not picked up by the other variables, for example, being located close to the central business district where land prices are high.