Affordable Housing and the Socioeconomic Integration of Elementary Schools

Also illustrative of the poor child’s educational deficiency is that the majority of high school dropouts come from families with incomes in the bottom fifth of the income distribution (Carnevale and Strohl 2010).

The eligibility criteria for the National School Lunch Program are described in detail below.

These plots were constructed using data from the Common Core of Data provided by the National Center for Education Statistics. When constructing the plots, a school was classified as an elementary school if it offered third grade. These counts exclude charter schools, magnet schools, and private schools.

Fry and Taylor (2012) document the severity and growth of residential segregation by income.

A leading proponent of school choice to achieve income integration is Kahlenberg (2013), who argues for magnet schools in high-poverty neighborhoods and the creation of incentives for non-poor schools to take poor students.

Many economists, starting with Friedman (1955), have long supported the movement away from neighborhood schools towards a choice-based model.

Tracking refers to the practice of sorting students into different classes based on their academic ability. Through various assessments, students would be labeled as above average (or gifted), average, or below average. This designation would then determine the student’s courses, and teachers and students are grouped only with similarly labeled classmates. Tracking can result from magnet programs because most of the classes offered by these programs coexist in the same public school as regular classes. For a review of the literature see Davis (2014).

Strong criticism of school choice comes from Rothstein (2013), who argues, “too many truly disadvantaged students live too geographically distant from middle-class schools for such schemes to be practical, and too many of their parents are too stressed to make the proactive choices necessary.”

For a review of the literature on the relationship between neighborhood quality and school performance visit Harvard’s Education Innovation Laboratory at http://edlabs.harvard.edu/neighborhoods-or-schools.

William Fischel’s Homevoter Hypothesis (Fischel 1985, 2001), which maintains that homeowners will go to great lengths to protect their property values, adds credence to the idea that poor families may be locked out of non-poor SAZs. Fischel shows that homeowners are aware of the positive effect that a good school can have on their home’s value and may believe that low-income students will tarnish the school’s reputation.

Ihlanfeldt (Forthcoming) provides some evidence on the extent to which improving housing affordability in neighborhoods zoned for high-performing schools might reduce racial segregation.

Kahlenberg also cites a review of 59 studies by Mickelson and Bottia (2009) that reached the same conclusion.

In this study, school performance was measured using test score data from Great Schools, and the housing cost measures were a weighted average of rental and owner monthly expenditures derived from the American Community Survey.

The districts excluded from the NCES and Maponics digital boundary files are mostly small, rural counties.

An admitted limitation of our data is that while school boundaries may change over time, our assignment to schools is based on the boundaries as they existed in 2013 in the case of the NCES files and 2011 in the case of the Maponics data. Because historical SAZ boundaries are not available, we cannot speak to how frequently SAZ boundaries have changed over the course of our sample. The geography that we ultimately use in our empirical work, however, is a grouping of schools based on socioeconomic characteristics and not the individual SAZs. That said, even if the school to which a home was assigned changed over the course of our study, as long as the post-SAZ-change school to which the home was assigned remained in the same socioeconomic grouping, the change in SAZ boundaries would have no impact on our housing stock measures. While we can only speculate, because households are spatially segregated by income and school boundary changes tend to be relatively minor amendments to existing boundaries, we believe that our housing stock measures are likely close to those that would be constructed using SAZs that are updated each year.

Tax roll data for recent years are available online at: http://​dor.​myflorida.​com/​dor/​property/​resources/​data.​html.​FDOR collects these rolls to monitor the performance of the county tax assessors.

See Appendix B of Andersson and Mayock (2014) for an analysis of the accuracy of the FDOR values during the 2008 – 2011 period.

In evaluating the county tax rolls, the FDOR computes price related differentials (PRD) to assess the vertical equity of the property tax within each county. The PRD is a statistic for measuring assessment regressivity or progressivity. If the PRD shows vertical inequity in the administration of the property tax, the tax roll may be rejected by the FDOR until the inequity is addressed by the county.

We exclude retirement homes and institutional housing, such as school dormitories and correctional facilities, from our analysis. We also exclude bank-owned properties, because such properties are not legally inhabitable.

The data can be accessed at http://​flhousingdata.​shimberg.​ufl.​edu/​a/​ahi_​basic. We include assisted units as a control variable and as a test of the validity of our models. An increase in the number of assisted housing units located in non-poor SAZs should increase the number of free lunch students in such schools because, by virtue of the fact that they qualified for housing assistance, the families occupying these likely also qualify for the free lunch program. If assisted units do not have this expected effect, it would suggest that our models may be misspecified.

To avoid simultaneity problems, we needed a school income typology in which an individual school remains in the same income group throughout the panel. We found that there was considerable persistence in the annual categorizations. Schools categorized as high (low) income in one year tended to be high (low) income in the other years.

Changes in our housing stock measures over time suggest that affordability declined in the high-income SAZs during the rise in housing prices and rose after the crash in housing prices. The differences in these housing stock measures over time, however, are not generally statistically significant, a finding that is largely attributable to very large variances about the means.

The FDOE data begins consistently reporting the pass rates on the mathematics proficiency portion of the Florida Comprehensive Assessment Test (FCAT) in the 2001-2002 school year. In 2011, the FCAT was replaced with a different assessment instrument (FCAT 2.0). Because of this change, we only use FCAT data from 2001 to 2010 in our analysis.

Ihlanfeldt and Mayock (2016a) review 12 studies, all of which provide evidence that a foreclosed upon property reduces the values of nearby homes. Although the focus of much of the literature has been on single-family homes, some studies have shown that foreclosures also lower the cost of condominiums (Campbell et al. 2011) and multifamily housing (Schuetz et al. 2008); hence, the evidence suggests that foreclosures bring about a general increase in housing affordability within a SAZ.

Unobservables that are time invariant pose no threat to identification because all of our models control for district-level time invariant heterogeneity.

Layoffs in a neighborhood could cause a reduction in local housing values if, in response to the negative income shocks, many households attempted to sell their properties at the same time, increasing inventory and depressing prices.

Even if incomes fell after the crash, within the higher income schools there may have been little if any free lunch switching as there may be relatively few families whose income would have fallen enough to become eligible for the free lunch program. Also, even if formerly high-income students became eligible for the program, their parents may have chosen not to apply to avoid their child possibly being stigmatized.

First-stage diagnostics for the IV models are strong in all cases. For both income proxies, Shea’s partial R-squared, which measures the strength of the correlation between the endogenous variable and its instrument, is in all cases respectable, with the lowest value equaling 0.31. For all of the endogenous variables, the first-stage F-statistics are all statistically significant at the 1% level and are, on average, greater than 10.

In the interest of brevity, the results tables only include the estimated effects of the affordable housing types. As expected, the unaffordable housing types are largely statistically insignificant.

The estimation of average partial effects from the PFP model is straightforward. If \(z_{kit}\) is an arbitrary element of the vector of regressors \(Z_{it}\) and the parameter associated with \(z_{ikt}\) is denoted Γ_k, then we simply differentiate the conditional mean with respect to \(z_{ikt}\) to get the partial effect for observation i in period t

where \(\phi \left ({}\right ) \) denotes the standard normal probability density function. Papke and Wooldridge (2008) show that the average partial effect (APE_k) of \(z_{ikt}\) on \(P_{it}\) can be consistently estimated by simply averaging over all observations of the above equation. That is,

There are 16 exogenous variables: the unaffordable housing types and assisted housing, broken down into 8 variables registering the proportion of each type found within high income and middle income SAZs.

A limitation of our affordable housing stock measures is that because we only observe the free lunch eligibility status and not the income levels of the students that we have classified as low-income, we cannot directly account for the impact of changes in the income levels on affordability. If incomes fell for low-income families whose children attended the schools that we have classified as low-income, then these families may have been unable to afford to relocate to housing units that we have classified as affordable. If none of the units that we classify as affordable could be purchased or rented by low-income households, then changes in the stock of affordable housing in the high-income SAZs should have no impact on the location of low-income households. The fact that we find that increases in the stock of affordable housing in high-income SAZs reduces the concentration of low-income students in low-income schools suggests that, if anything, our estimated effects are understated because our affordable housing stock variables are measured with error.

To investigate the extent to which our findings are sensitive to our choice of affordability cutoff, we reconstructed the affordable housing stock measures using two different cutoffs for the implied rent on the property: the one-bedroom and three-bedroom FMRs reported by HUD. The one-bedroom-FMR cutoff mechanically decreases the affordable housing stock relative to the stock measures based on the two-bedroom-FMR cutoff as the one-bedroom-FMR is lower than the two-bedroom-FMR. By similar logic, the affordable housing stock based on the three-bedroom-FMR cutoff is mechanically larger than the stock based on two-bedroom-FMR cutoff. We used these new affordable stock measures to re-estimate the models that are reported in the main text. The results are remarkably similar across the models based on the three different affordability measures. In all but a few cases, if the parameter associated with one of the affordable housing types is statistically significant (at the 10 percent level) for the two-bedroom-FMR definition of affordable housing, it is also statistically significant for the one- and three-bedroom FMR affordability definitions. The only notable exception is that the proportion of single-family rentals located in high-income SAZs, while significant across the board using the two-bedroom FMR to define affordability, is insignificant in the models using the one-bedroom-FMR cutoff and the exam scores to define the school groupings. The coefficient on the same single-family rental variable is also statistically insignificant (but only marginally, with p-values ranging between .11 and .12) in the models where affordability is based on the three-bedroom FMR and the school groups are defined by the Global Free Lunch Status.

To illustrate, we added 100 units of single-family rentals to the high-income group of SAZs and then calculated the effect this would have on the proportion of the district’s single-family rentals located in high-income SAZs.

In Florida, mobile homes are an important part of the housing stock. If we define the total single-family stock as the sum of detached single-family units, condominiums, and mobile homes, on average mobile homes constituted 14 percent of the total number of single-family units. Condominiums, on the other hand, comprised only 13 percent of this total.

The affordable housing types with relatively large elasticities and those yielding relatively large effects from an increase in their number differ because the percentage change is dependent on the base, which we measure as the median number of existing units found within the middle- or high-income SAZs.