District-Level Achievement Gaps Explain Black and Hispanic Overrepresentation in Special Education

Abstract

To examine whether special education racial risk ratios reported by U.S. school districts are explained by district-level confounds, particularly, racial achievement gaps, we analyzed merged data ( N = 1,952 districts for Black–White comparisons; N = 2,571 districts for Hispanic–White comparisons) from the U.S. Department of Education’s Office of Civil Rights, Stanford Educational Data Archive, and Common Core data sets. Regression analysis results indicated that Black– and Hispanic–White district risk ratios were strongly related to Black– and Hispanic–White district achievement gaps. These results reconcile findings from district-level data with those from student-level data and support the finding that, when compared to otherwise similar White students by controlling for group differences in achievement, non-White students are on average underrepresented in special education. That is, non-White overrepresentation in special education in most districts is explained by racial achievement gaps in these districts. Residuals from the regressions provide a more accurate way to monitor for outlier districts than the current practice required in federal regulations of using unadjusted risk ratios.

Unmeasured Confounding and Racial or Ethnic Disparities in Disability Identification

“…Because these E- values assess for unmeasured confounding conditional on measured covariates, large unmeasured confounding resulting from factors other than those already included in recent studies would generally be necessary to explain recently reported disability under-identification of students who are Black or Hispanic. Weaker confounding than indicated by these E- values would not fully explain the recently reported disability under-identification of students who are Black or Hispanic (Farkas et al, 2020; Morgan et al, 2015, 2017). Although small-to-moderate unmeasured confounding might possibly result in null associations for some specific conditions, as mostly indicated by smaller LB 95% CIs that themselves are based on conservative E- value point estimates and conditional on each study’s measured covariates, large-to-very large unmeasured confounding would be necessary to result in levels of overidentification consistent with federal regulation.…”

“…Because findings of disability under-identification have been repeatedly replicated (e.g., Farkas et al, 2020; Morgan et al, 2015, 2017; Odegard et al, 2020) and, as indicated here, are largely robust to the possibility of unmeasured confounding, federal law and regulation may need to be redirected to instead monitor for systemic bias resulting in disability under-identification for students who are Black or Hispanic (Morgan et al, 2015, 2017). A recently proposed way to do so would for U.S. school districts to adjust the reported RRs for achievement gaps (Farkas et al, 2020; Morgan et al, 2017). Doing so would more accurately identify U.S. school districts where overidentification may be occurring as well as help ensure that students with disabilities who are Black or Hispanic are not being denied access to services based on their race or ethnicity.…”

“…Although findings of under-identification are characterized as in error and resulting from flawed analyses (Skiba et al, 2016), under-identification has been repeatedly replicated including in analyses of other nationally representative samples and data collection methods (e.g., Farkas et al, 2020; Morgan et al, 2017). Yet education researchers continue to dismiss evidence of under-identification (Artiles, 2019; Grindal et al, 2019; Whitford & Carrero, 2019).…”

“…Measured confounds included reading achievement, biological sex, free or reduced-price lunch eligibility, and English Language Learner status as well as school fixed effects. Study 2 (i.e., Farkas et al, 2020) reported adjusted RRs for disability identification generally based on analyses of district-level data ( N = 1,952 and 2,571 U.S. school districts) from the U.S. Department of Education’s Civil Rights Data Collection merged with the Stanford Education Data Archive. Measured confounds included the district’s Black- or Hispanic-White achievement gaps, enrollment size, percentage of students receiving free or reduced-price lunch, and the percentage of Black or Hispanic students as well as state fixed effects.…”

“…Replication studies continue to find evidence of under-identification including in analyses of other nationally representative samples (Morgan et al, 2017) as well as both student- and district-level data sets that indicate overidentification in unadjusted analyses, helping to address criticisms about data collection and sampling (Farkas et al, 2020; Morgan et al, 2017). The replication work has also statistically controlled for a wide range of student-, family-, and school-level confounds including family socioeconomic status (SES) and student achievement (Morgan et al, 2017).…”