Heterogeneous Effects of Charter Schools: Unpacking Family Selection and Achievement Growth in Los Angeles

ABSTRACT

Disparate findings on whether students attending charter schools outperform peers in traditional public schools (TPS) may stem from mixing differing types of charters or inadequately accounting for pupil background. To gauge prior family selection and heterogeneous effects, we distinguish between conversion and start-up charter schools, along with a third site-run model operating in the Los Angeles Unified School District (LAUSD). We find that TPS campuses converting to charter status (conversions) attracted more experienced and consistently credentialed teachers, and served relatively advantaged families, compared with newly created charter schools (start-ups), after tracking 66,000 students over 4 years, 2007–2011. Charters overall attracted pupils achieving at higher levels as they began a grade cycle (at baseline), relative to TPS peers, most pronounced among conversions that remained affiliated with the district. After matching students on their propensities to enter a charter school, we find that pupils attending charters outperformed TPS peers over the 4-year period. These benefits are most consistent and moderate in magnitude for middle school students. We observed significant though small effects in English language arts for pupils attending charter high schools. Latino students, mostly attending start-ups, enjoyed consistent benefits from attending a charter school.

KEYWORDS:

• achievement gains
• charter schools
• quasi-experimental methods

Acknowledgments

Hye Kyung Lee and William Welsh helped to frame our analysis. Much appreciation is expressed to the generous staff of the CALPADS data shop at the California Department of Education who compiled the information on charter school students. Rachel Bonkovsky, José Cole-Gutiérrez, Kathy Hayes, Michael Kirst, Cynthia Lim, Jason Mandell, Jane Patterson, Robert Perry, Elizabeth Robitaille, and Anisah Waite offered generous advice and support for the project, along with helpful feedback on earlier drafts. Three anonymous reviewers and the editor provided additional critiques to improve the article. Any errors of omission or interpretation are solely attributable to the authors.

Funding

Luke Dauter’s work was supported by the Spencer Foundation and the research network, Policy Analysis for California Education, thanks to the Hewlett Foundation.

Notes

1. The rigor of authorizing agencies may be another consequential element of the regulatory environment. But neither the count of authorized start-up and conversion charters, nor the selectivity of student admissions permitted under local policies, has been found to be predictive of achievement gains (Gleason et al., 2010; Zimmer et al., 2012).

2. Traditional schools may shed less effective teachers and raise compensation for more effective teachers when a charter school opens nearby, as Jackson (2012) found, drawing on North Carolina data. Cowen and Winters (2013) found higher turnover rates in Florida charter schools, compared with TPS, over the 2002–2008 period. Less effective teachers were more likely to exit than more effective teachers, but these rates did not differ significantly between sectors.

3. Focusing on the Milwaukee school voucher program, Fleming, Cowen, Witte, and Wolf (2015) found that parents drawing school vouchers earned less than TPS peers but the former displayed significantly higher levels of educational attainment.

4. In contrast, Hoxby and Murarka (2009) found no differences among ethnic groups for pupils attending New York City charter schools.

5. Most LAUSD students enter middle school at Grade 6 and high school at Grade 9.

6. Care must be taken in estimating a propensity score, because inclusion of too many variables, even though correlated with the treatment, can actually induce overt bias in the matched samples by reducing the overlaps between the treatment group and control group (Lesaffre & Albert, 1989).

7. The algorithm works by finding improvements in the most imbalanced variables, gradually improving balance over successive iterations during which each variable is weighed according to its relative importance for achieving the best balance. One may use the genetic algorithm by drawing from the propensity score and the covariates after they have been made orthogonal to it. If optimal balance is achieved by simply matching on the propensity score, then the other variables are given a zero weight and genetic matching will be equivalent to propensity score matching. One advantage of the genetic matching algorithm is that it directly optimizes covariate balance.

8. The Kolmogorov-Simirnov (KS) test is a nonparametric test for the equality of continuous, one-dimensional probability distributions that can be used to compare two samples, and the KS test is sensitive to imbalance across the empirical distribution.

9. Fewer differences for ESBMM schools emerged, except that the two high schools served higher shares of Asian students, perhaps another particular niche within this field of deregulated firms.