Econ 398: Public Economics Seminar

Kate LeMasters

Econ 398: Public Economics Seminar

Review Sheet: October 1, 2014

Currie & Yelowitz: Are Public Housing Projects Good for Kids?

1.  What is the primary research question?

The authors’ primary research question is how public housing participation affects housing quality and educational attainment, with the focus being on child’s educational attainment. They would like to also look at how public housing compares to other alternatives, such as private, unsubsidized housing or some other form of publicly subsidized housing, but the data doesn’t allow them to draw causal conclusions here. Due to constraints of the model, they must ask whether or not families with mixed-sex children are more likely to live in the projects, as this is their instrumental variable. They also look at whether marital status, age of the household held, ethnicity (black vs. white), and other factors affect the model.

2.  What are the author’s main findings?

The authors find that project households are less likely to suffer from overcrowding or live in high-density complexes (they have higher housing quality than they would otherwise). They also find that project children are less likely to have been held back in school, but this effect is much more significant for boys than girls. However, they also find that households in the projects are still less satisfied with their housing and neighborhood than other projects and that project children are more likely to have changed schools. There are no differences between project and other children in school ratings, extra-curriculars, and grade retention. Additionally, families with both a boy and girl are 24% more likely than families with same sex children to live in project housing, as these families get an additional room. Participation declines with age of household head, declines for married couples, and is highest among blacks.

3.  How do the questions, methodologies, and conclusions differ from previous closely related research?

There don’t seem to be other studies too similar to this one, but the authors do mention a few articles that they drew from. First, Aaron (1972) hypothesizes that any benefits inhabitants derive from physical housing amenities are offset by their inferior surroundings. So, his question is different, and, as it was written in 1972, may not be relevant for this study. Butcher and Case (1994) ask whether sex composition affects educational attainment, and they find that having multiple girls reduces educational attainment but there is no effect for boys. Because the authors’ primary question here is different, we cannot directly compare their findings, but the authors here found that the beneficial effects on school are confined primarily to boys, so boys still have ‘superior’ effects. This study and others related to sex and education attainment (Kaestner (1997); Kuo and Hauser (1996) look at completed educational attainment vs. being held back, so there is a different question asked here. Angrist and Kreuger (1992, 1995) use a similar methodology of instrumental variables when using multiple datasets with differing information. This study seems to be the first one using an instrumental variable in this way for the question of housing projects and educational attainment.

4.  Does the author set up a reasonable theoretical framework to justify their approach?

Yes, the authors begin with the statement that families would not move into public housing projects unless they were better in some respects than alternatives. This is reasonable because families must apply for public housing and there are incredibly long lists to get in, so there is likely demand for it. In opposition, prior work has hypothesized that any benefits for the individual are squandered by their surroundings, but there is little evidence that this is true. When creating an instrumental variable, they state that families eligible for larger apartments/higher subsidies should be more likely to live in public housing projects all else equal, which is a logical statement.

5.  What kind of assumption do the authors make – both that they admit to and they leave unsaid? Are these assumptions unreasonable?

It seems that the authors make more assumptions than most other papers we have seen due to their instrumental variables. While it may be part of their theory, they also do assume that families eligible for larger apartment due to the gender of their kids are more likely to live in public housing projects because they have a higher demand for them. By only looking at whether or not children have been held back, they assume that children that have been held back once are similar to those that have been held back more than once. While this is a reasonable assumption, children may be held back due to how many times they have changed schools, which they don’t consider. While it is not possible to look at neighborhood effects, the exclusion of these likely biases the results, as this could potentially proxy/control for school quality, etc. They do not include small towns in their sample, which also likely biases the results, although they do not adjust for it because they assume the most troubled projects are not in small urban areas. While this may be true, they do not provide additional support.

6.  Is the empirical strategy up to the task of identifying a clean answer to their research question?

The empirical strategy is unique and attempts to remove endogeneity between living in the projects and housing quality/education attainment of children, but, because an instrumental variable is used, it is obviously imperfect and still has many issues. The authors use a two-sample instrumental variable (TSIV) technique to combine information on the probability of living in a project with information on outcomes. This is appropriate because outcomes are in one data set and the endogenous regressor in another while both data sets contain the IV. The TSIV is an indicator equal to one if the household is entitled to a larger unit based on sex composition of children and zero if entitled to a standard unit, as families eligible for larger apartments should be more likely to live in public housing. However, the IV may be correlated with other forms of housing (section-8 vouchers), not only public housing, so it is likely biased. Additionally, the effect may be different for girl-girl vs. boy-boy even though they are both measured as a zero as an IV. They examine many measures for housing quality, as rent is capped at 30% of income for public housing. They also look at many measures for educational attainment, primarily if the child was held back in school. While this is a better indicator than completed education, it may largely be effected by school quality, participation/education of parents, etc. The model itself seems to be correct, as basic OLS is used without the IV, with controls for household head and child individual characteristics. To determine legitimacy of the IV, project housing is placed on the left-hand side and a dummy for gender makeup of kids on the right-hand side. Then, the IV is placed on the right-hand side and outcomes are again on the left. With robustness checks, how ages are grouped for being held back, and sensitivity analysis that removes some controls the results hold, so it is a sufficient model. However, the IV obviously still has many faults.

7.  Is the data up to the task?

The data is collected from the 1992 and 1993 waves of the Survey of Income and Program Participation (SIPP) to look at how living in a project is associated with outcomes, but they ultimately use US Census data to obtain outcomes and the March Current Population Survey (CPS) to obtain information on the probability of living in a project. SIPP’s information about project participation and child outcomes allows them to estimate the baseline OLS. However, the data is not specific enough to identify the separate effects of voucher programs, which is a critical fault. Additionally, they are not able to measure neighborhood effects, which is critically important. However, they are able to measure many controls and create an IV, so there are significant indicators present in the data. By combining multiple sources, they are able to carry this out.

8.  Given the theoretical framework and the assumptions made, do the authors properly interpret their own results?

Given the theory, assumptions, and restrictions made by IVs, they properly interpret their results. However, they draw conclusions that may be too broad. When running the regression separately for whites and blacks, they find that for white, the projects have no significant effect, but for blacks, living in the projects reduces the probability of grade repetition by 19%. However, the author cannot control for neighborhood effects, which may come into play here, especially in areas that are predominated by one ethnic makeup. Additionally, housing quality and educational attainment may be correlated, as the home environment likely effects the child’s ability to do homework, etc, so they should run a regression with household quality on the right side and education on the left. For educational achievement, the authors do not consider whether or not getting held back in earlier grades or later grades has larger effects on eventual educational attainment. So, their broad claim of education should be split up into measuring being held back at different ages. Given the assumptions made and models used, they correctly interpret their results, but their claim that projects have positive effects on housing quality and children’s academic achievement is too general and is not entirely supported by the empirics.