Evaluating a Shelter-Based Program to Improve School Attendance Among Homeless Children in Shelter

Background and Purpose: The approximately 1.5 million U.S. public school students experiencing homelessness each year are chronically absent from school at double the rates of their housed counterparts. Chronic absenteeism can result in reduced academic achievement, lower socioemotional health, and lower graduation rates and lifetime earnings, and is therefore cause for policy intervention. In 2018, the Attendance Matters program, a shelter-based intervention to improve school attendance, was launched as a pilot in five family shelters in New York City, the municipality with the nation’s largest share of homeless families. Attendance Matters leveraged three core service tenets: interagency coordination, reliance on data, and the use of evidence-based practices. This study assesses the program’s impact on school attendance, as well as on standardized test scores and school stability, in its first year.

Methods: Our data are drawn from administrative records maintained by the NYC Department of Homeless Services (DHS) and the NYC Department of Education (DOE). Our sample consists of 2017 and 2018 school year and shelter records for K-12 public school students residing in DHS shelters , as linked using probabilistic techniques. There are some 24,000 student-school-years in total. The DHS data describes students' circumstances of shelter entry and the particulars of their shelter spells, while the DOE data details biographical information, enrollment, attendance, and standardized test scores.

The central empirical challenge is that treatment shelters are not randomly selected. To address this, we use two quasi-experimental analytical techniques. The first is linear regression conditioned on rich controls. The second is an elaboration: a difference-in-differences approach comparing treatment and control shelter pre- and post-treatment. We demonstrate that the treatment and control groups are similar on observables, and that the scope for biases arising from time-varying unobservables is small, lending credence to our chosen methods. Treatment is defined in terms of shelter-level exposure to the program, such that our estimates are lower-bound intent-to-treat effects.

Results: Compared to controls, the program was associated with a significant 3-day reduction in absences for primary school students in participating shelters. We also saw point estimate decreases in the absence rate and rates of chronic absence, though findings not consistently significant across model specifications. We did not see any significant decrease in days absent or the absence rate for high school students, though school stability did increase for students in participating shelters.

Conclusions and Implications: Attendance Matters appears to result in reduced absenteeism among primary school students. The findings are noteworthy given it was the first year of a minimally funded, low-infrastructure program, whose impact was assessed in an Intent-to-treat design. This evidence suggests that coordination across agencies, paired with reliance on data and evidence-based practices, can impact school attendance for highly vulnerable children. In addition, these results show promise for the forthcoming evaluation of the program’s second year and, given the program’s limited spending, a cost-benefit analysis.