Appendix C
Estimates of the Employability of the Public Housing Population Subject to Welfare Reform Using Bureau of the Census Public Use Microdata Samples (PUMS)1
This Appendix provides information on how estimates of probable employment for mandated public housing residents were developed using 1990 Census Public Use Microdata Samples. These estimates are presented in Section IV of this report. The methods presented here are replicable in any community containing public housing and represented in PUMS samples. The primary steps shown in this Appendix concern the kinds of data necessary to develop PUMS employment estimates, the modifications to the data necessary for analysis, and the criteria for choosing independent variables (those characteristics likely to predict employability). The results of logistic regressions on PUMS populations are weighted by the demographic distribution of the mandated public housing residents. Adjustments in the estimates of work participation are made to take into account lack of information on education from some housing authorities and the trends in employment in each city between 1990 and 1997.
Background. The five percent sample of the 1990 Census of Population and Housing from the Public Use Microdata Samples (PUMS) was used to estimate the long-term probability that certain heads of household in public housing would be employed if they sought work and had the same degree of success as similar persons in PUMS. For each of the eight study Housing Authorities (HAs), these residents of public housing are those mandated by Temporary Assistance to Needy Families (TANF) to find employment. The estimates were done in support of the report's analysis of rent revenue impacts on the Public Housing Program (Section IV), which centers on changes in rent revenues paid to HAs by those households mandated to find jobs under the rules of TANF.
Important household factors affecting housing authority rent revenues include residents' household income, adjustments to income, and utility allowances. Incomes are expected to change dramatically for many households as they reach the end of their welfare assistance under TANF. The key unknown in determining a household's long-term income is its potential wage income.
This appendix describes how PUMS data and logistic regressions were used to estimate the probability of being employed some time after TANF benefits end. In Section IV of the report, these estimates were combined with current HA program data to complete the needed calculation of wage income.2 Finally, rent revenues could then be determined.
In the first step of estimating job participation, data for the Public Use Microdata Areas (PUMAs) which make up each of the eight study cities were extracted from PUMS. For the smaller sites, one PUMA encompasses the entire city. The largest cities are made up of more than one PUMA. Table C-1 gives the PUMS identification numbers of the Metropolitan Statistical Areas (MSAs) and PUMAs used for analysis.
Key Demographic Characteristics. To estimate the probability of having a job in each of the eight study cities, subsamples of households which closely resemble the public housing residents with respect to potentially key demographic characteristics were drawn from PUMS. The subsamples were restricted to households with no children under 2 years old, to simulate the mandated households who are exempted because of the presence of very young children.3 The subsamples were further restricted to heads of household who were nonelderly, single females, not in school, and who were in the civilian labor force in 1990. Finally, in Norfolk, Richmond and Cleveland, the PUMS subsamples also selected only Black heads of household, reflecting the very small proportion of non-Black mandated public housing residents in those cities.
Demographic data available both from the PUMS and each HA, which could be used as potential predictors of job participation, included ages of children and age, sex, marital status, race and ethnicity of head of household. What likely would be regarded as the most important characteristic, educational attainment, is part of the PUMS data for each city, but was available from only the Norfolk, Richmond and Dallas HAs. Education was also provided by California Social Services for San Francisco welfare recipients. A match with about 45 percent of the mandated households in public housing allowed education to be included as a variable for estimation of the probability of mandated households working in San Francisco.
Logistic Regressions: Logistic regressions were run from PUMS data, utilizing SPSS statistical software, to estimate job participation for public housing residents mandated by TANF to find a job. The dependent variable for the regressions was employment, defined as working in 1989 at least 30 hours per week for 26 weeks or more. The independent variables used were presence of children under six years of age and education, age and race/ethnicity of head of household as given in Table C-2. Presence of children under six and age of head were treated as a combined independent variable with six categories (none or some children under six by three age categories). All independent variables were defined as indicator variables, with zero being the value of the variables for the excluded categories.
The forward likelihood-ratio method was used to determine which demographic characteristics were of importance in each site with respect to employment. Independent variables were added if their entry significance level was sufficiently small. The regression coefficients are given in Table C-4. Estimated work participation rates, applying the demographic distribution of the mandated residents in public housing, are given in Table C-5. The distribution for the each of the eight HAs is given in Table C-6.
Except for Richmond, education was the first independent variable to be entered. Surprisingly, the education data did not enter into the regression at all for Richmond.
Work Participation Estimates and Adjustments for Education. Regression coefficients were applied to the distribution of mandated public housing residents to estimate how many would be employed, assuming they would have the same probability of being employed as the data show for their counterparts in the PUMS. In addition to the distribution of mandated public housing residents for each city, Table C-5 gives the estimated probabilities of employment for each applicable regression equation.
In the three cities where education data was available and entered in as an independent variable, estimates of employment were made both with and without inclusion of education. As seen in Table C-5, removing education in the regressions for Norfolk, Dallas and San Francisco reduced the estimated employment by at least ten percentage points. For those cities, the estimate used in the Section IV of the report was that obtained including education. In each of the four cities (Cleveland, Columbus, Toledo and Los Angeles), where education was unavailable, the estimated employment was reduced by ten percentage points to compensate for what might be the result if data on education had been available. It is recognized that the correctness of this may be questionable, but given the consistency with which education played a role in the estimates for other sites, it seems to be a reasonable correction to make. For Richmond, no adjustment was made since education did not enter into the regressions there using the forward likelihood ratio method.
In the three cities where education data was available and entered in as an independent variable, estimates of employment were made both with and without inclusion of education. As seen in Table C-5, removing education in the regressions for Norfolk, Dallas and San Francisco reduced the estimated employment by at least ten percentage points. For those cities, the estimate used in the Section IV of the report was that obtained including education. In each of the four cities (Cleveland, Columbus, Toledo and Los Angeles), where education was unavailable, the estimated employment was reduced by ten percentage points to compensate for what might be the result if data on education had been available. It is recognized that the correctness of this may be questionable, but given the consistency with which education played a role in the estimates for other sites, it seems to be a reasonable correction to make. For Richmond, no adjustment was made since education did not enter into the regressions there using the forward likelihood ratio method.
Table C-3 gives the adjusted estimates of employment as used in Section IV of the report. In addition to the adjustments considering the effect of educational data, there was an adjustment in the employment estimate for Norfolk taking into account changes over time, as described in the next section. The work participation rates in Table C-3 vary from 40 percent in Toledo, to 72 percent in Dallas.
Potential Change in Employment Over Time. The work participation rates reflect the 1989 economy. Under a contract with Standard & Poor's DRI, trends in employment since 1989 were analyzed in each of the sites using the Census Bureau's Current Population Survey (CPS) for March in each of the years from 1990 through 1997. Six target populations were examined for trends over the seven years using restricted populations of different configurations to overcome the generally very small sample sizes obtained when all constraints are imposed which are similar to those for HA mandated residents. These variations are given below:
Target 1: Black or Hispanic females in the civilian work force, single, age 18-544, not in school, with no more than a high school education, and with no children under age 2. (Closest to HA mandated population.)
Target 2: Females in the civilian work force, not in school, with no more than a high school education. (Least restrictive -- does not include constraints with respect to race, marital status, age, and children<2.)
Target 3: Black or Hispanic females in the civilian work force, single, not in school, with no more than a high school education, and with no children under age 2. (Omits age constraint.)
Target 4: Black or Hispanic females in the civilian work force, single, age 18-54, not in school, and with no more than a high school education. (Omits children under 2 constraint.)
Target 5: Females in the civilian work force, single, age 18-54, not in school, with no more than a high school education, and with no children under age 2. (Omits race constraint.)
Target 6: Black or Hispanic females in the civilian work force, age 18-54, not in school, with no more than a high school education, and with no children under age 2. (Omits marital status constraint.)
Note that even when the population is constrained to resemble the HA mandated population, the CPS sample still has a different distribution with respect to variables such as education achieved, number of children under six, age, and other important variables potentially making a difference in job participation. Sample sizes for the 1990 data, which comes from PUMS, vary between 4 and 176 for Target 1 and between 70 and 1,205 for Target 2. For the CPS in years 1991-1997 the sample sizes are roughly one-half those of the 1990 data. Table C-7 presents the sample sizes by MSA and target population for 1990 and averaged over the years 1991 through 1997.
The charts at the end of this Appendix show the trends in the estimated proportion employed at least 30 hours per week for a minimum of 26 weeks. The trends are given for Targets 1, 2, 5 and 6 for the MSA and central city for each site.5 The trends shown are based on weighted data.
For the CPS population most similar to the public housing TANF mandated population (Target 1), trends in the estimated proportion employed showed the following: Richmond, Cleveland and Dallas were virtually unchanged; Norfolk dropped to a level about 40 percent less employed; employment in Columbus rose about five percentage points; Toledo's sample was extremely small for the target group and employment was very volatile, but for larger groups seemed to stay steady; employment in Los Angeles dropped about five percentage points; and, there is a mixed picture in San Francisco, where employment seemed to drop about ten percentage points in the MSA as a whole, but may have increased in the central city based on a fairly small sample size (for a larger group -- Target 2 -- without Target 1's age, race, ethnicity, children under six and marital status constraints, the employment in the central city decreased slightly).
With the possible exception of Norfolk, the trends do not indicate a rationale for substantial adjustments to the work participation estimates. For the analyses in Section IV, the estimates in Table C-3 were not adjusted further for cities other than Norfolk.
In the case of Norfolk, employment decreases about 40 percent in the central city and 45 percent in the MSA. Although not shown in Table C-7, sample sizes can be considerably smaller for central cities than for MSAs, although they are fairly close in Norfolk. For Norfolk, the Target 1 MSA sample is all in the central city for 1990 and 88 percent are in the central city for 1991 through 1997. For Targets 2, 5 and 6, the Norfolk central city sample is 78 to 100 percent of the sample for the MSA as a whole, depending on target population and year. Given the forty percent decrease shown for Target 1 in the central city, and taking into account the differences between PUMS, CPS and HA populations, it was thought reasonable to adjust Norfolk's employment estimate downward by about one-third. Thus Table C-3 shows 42 percent employed for Norfolk, which is one-third less than the 63 percent obtained by applying the logitistic regression coefficients to the distribution of mandated public housing residents in Norfolk.
Charts
1 Terrence L. Connell, Division of Policy Studies, Office of Policy Development and Research, U.S. Department of Housing and Urban Development, Washington, D.C.
2 The analysis in the report's Section IV, "The Financial Exposure of the Housing Authorities," relies on the wage income of other public housing heads of household, who are working but not receiving welfare assistance, for what might be expected as a wage for those coming off of welfare.
3 Under two years was selected instead of one year, the requirement in most places, because employment was determined based on job experience in the previous year (1989), not 1990 when the Census was actually taken.
4 The upper age limit was 64 for sites other than Richmond and Norfolk.
5 Targets 3 and 4 were not included here because they did not seem to contribute any additional information useful in assessing what may have occurred between 1990 and 1997.
|
| 
