Association for Public Policy Analysis & Management

Open-enrollment and choice systems are designed to remove the role of school location and residential settlement patterns from school attendance. Families are encouraged to consider all schools, not just those nearby, to pursue a rewarding school match. The challenge of getting to school each day, however, may wear on students just as with adult commuters. Previous research has shown that difficult commutes (Stein & Grigg, 2019) and the walking portion of commutes (Burdick-Will, Stein, & Grigg, 2019) are related to increased school absence. Among adults, a long and difficult commute can encourage an employee to reduce transit strain by seeking out a different workplace (Koslowsky, Kluger & Reich, 1995; Amponsah-Tawiah, Annor & Arthur, 2016). In this paper, we show that students who attend high schools that are difficult to get to are more likely to transfer schools during their freshman year. Moreover, we find that when these students change schools, their second school is substantially closer to home, requires fewer vehicle transfers, and is less likely to have been included among their initial set of school choices. Student mobility frameworks (e.g. NRC & IM, 2010; Welsh, 2017) do not explicitly account for geography as an influence in mobility, and this analysis to our knowledge is the first to directly examine geographic influences on school transfer, much less in the context of school choice.

The data for this study comes from the administrative records for approximately 4,000 residentially-stable first-time ninth grade students during the 2014-15 school year in the Baltimore City Public Schools, along with open source transportation scheduling data (General Transit Feed Specification [GTFS]) for Baltimore City in 2015. Using suite of tools that allow the utilization of GTFS data with the Network Analyst suite in ArcGIS, we estimate routes from home to school for each student in the dataset (see Stein & Grigg, 2019; Burdick-Will, Stein, & Grigg, 2019 for examples of this methodology). From each estimated route we create metrics of commute difficulty that include total transit time and number of connections need. We then use these route metrics to predict whether a student will change schools during the ninth-grade year using logistic regression models with high school fixed effects, controlling for student demographic characteristics. For students who changed schools, we also examine the characteristics and location of the school to which the student transferred.

We find that students with longer commutes are more likely to change schools. Specifically, students who subsequently transferred schools had commutes that were on approximately 5 minutes longer (effect size = .32) than students who did not transfer schools (commute times of 40.3 and 35.6 minutes respectively). Students’ who transferred had new commutes that were 7.5 minutes shorter than their original school (effect size = .51). A statistically significant effect on total travel time was estimated from high school fixed effect model such that 10 minutes of commute time is related to an increase in the odds of transferring (OR = 1.30; p<.01).