Discipline Disproportionality and Policy Reform in Arkansas: Effects of State-Wide Limits on Exclusionary Discipline in Elementary School

Abstract

At least 16 states have limited exclusionary discipline for elementary age students, but little is known about the effects of these policies on discipline rates and disproportionality. This paper tests whether a limit on exclusionary discipline in grades K-5 in Arkansas impacted the likelihood of suspension and expulsion overall, for different student subgroups, and in different types of schools. Using a comparative interrupted time series approach, we find the policy—which included exceptions that could be interpreted subjectively—reduced the risk of suspension or expulsion in grades K-5, but that students who are not Black, students without disabilities, and more economically advantaged students experienced larger reductions in suspension and expulsion risk. Implications and limitations are discussed.

Disclosure statement

Any results, information, and opinions stemming from the research are solely those of the authors.

Notes

1 Notably, there are some schools that serve a mix of elementary (K-5) and middle (6-8) grades, with about 27% of student-year observations in these types of schools. Most of these schools (77%) end with grade 6, rather than expanding through grade 7 (6%), grade 8 (15%), or higher (1%). It is possible the effects of the policy could differ in these mixed grade schools, and that the grades 6-8 in these schools may not be completely unaffected by reform. For example, schools that contain both K-5 and 6-8 might be more likely to reduce exclusionary discipline in grades 6-8 in addition to K-5, even though it was not a direct requirement of the policy. To test whether this might be affecting the results, we conduct a set of tests dropping these schools with mixed grades, and the results are substantively and directionally the same with some slight differences in magnitude.

2 Notably, Hispanic/Latinx students are underrepresented both when controlling for and not controlling for infraction types, suggesting that there is neither differential selection nor differential processing for Hispanic/Latinx students in the state (Anderson & Ritter, 2017). Another study, also in Arkansas, more directly tests for differential selection v. differential processing, and finds that Hispanic students are less likely than White students to be referred for violence/weapon infractions or major nonviolent offenses (tobacco, vandalism, bullying), similarly likely to be referred for guns, drugs/alcohol, or minor nonviolent offenses (disorderly conduct, insubordination, etc.), and more likely to be referred for truancy infractions (Ritter & Anderson, 2018). This study also finds that, conditional on infraction type, Hispanic students are generally similarly or less likely to receive exclusionary discipline, except for in response to minor violence/weapon infractions, and when controlling for district, there is no longer evidence of differential processing.

3 Notably, there are some schools that serve a mix of elementary (K-5) and middle (6-8) grades, with about 27% of student-year observations in these types of schools. Most of these schools (77%) end with grade 6, rather than expanding through grade 7 (6%), grade 8 (15%), or higher (1%). It is possible the effects of the policy could differ in these mixed grade schools, and that the grades 6-8 in these schools may not be completely unaffected by reform. For example, schools that contain both K-5 and 6-8 might be more likely to reduce exclusionary discipline in grades 6-8 in addition to K-5, even though it was not a direct requirement of the policy. To test whether this might be affecting the results, we conduct a set of tests dropping these schools with mixed grades, and the results are substantively and directionally the same with some slight differences in magnitude.

4 “Other” non-specified infractions, which are not a researcher-created category, but rather, are reported in the state data as “other,” are also relatively common with about 91 incidents per 1,000 students. These “other” infraction types are about 30% of infractions reported for students in grades K-5 during the full panel, relative to almost 36% of infractions reported for students in grades 6-8. The types of students referred for “other” infractions are generally similar in grades K-5 and grades 6-8, except that more of these referrals in grades 6-8 are for Black or Hispanic students (46.8% and 7.3%, respectively, relative to 53.3% and 5.5% in grades K-5), and fewer in grades 6-8 are for male students (70%, relative to 77.2% in grades K-5). We conduct some tests in which these “other” infractions are also included with these subjective infraction categories, and the results are substantively and directionally similar, with some differences in magnitude or statistical significance. See Appendix Tables A1-A4.

5 In addition to our preferred CITS specification, results using a simple difference-in-difference analysis are available in Appendix Table A5. The results are directionally similar to the CITS results, with some differences in magnitude and statistical significance.

6 We thank a reviewer for this suggestion. The results are in Appendix Tables A6-A9 and generally indicate that the change in likelihood of receiving multiple referrals or multiple suspensions/expulsions is generally directionally similar to the change in likelihood receiving at least one referral or suspension/expulsion, but with generally attenuated magnitude.

7 For one cohort of students, they would normally return to “comparison” (grade 6) after previously being “treated” in grade 5. For these cases, we drop the grade 6 observations for this cohort only to ensure a clean (non-policy-exposed) counterfactual.

8 These models are a series of separate linear probability models with binary outcomes (receipt of a particular type of consequence), where the sample is restricted to a particular type of infraction. Using multinomial logistic regression would be a logical way to do this, but these models would not converge if including school fixed effects, so we used the LPM models with binary outcomes to include school fixed effects for comparability with the other preferred specifications.

9 The results are very robust to the use of a district fixed effect instead of a school fixed effect. The coefficients are very similar in magnitude, but lose statistical significance in column 6. Results available from the authors by request.

10 The results are very robust to the use of a district fixed effect instead of a school fixed effect. The coefficients aver very similar in magnitude, but lose statistical significance in column 3. Results available from the authors by request.

11 School percent non-White and school percent FRL are moderately positively correlated (r = 0.54), so comparisons by race and FRL status capture relatively similar comparisons. School enrollment is loosely correlated with percent non-White (r = 0.09) and negatively correlated with percent FRL (r = −.29).

12 White students, Hispanic students, and students of a race/ethnicity other than White, Black, or Hispanic, all experienced reductions in OSS/expulsion risk. However, it is notable that the other race/ethnicity group includes a variety of different less prevalent subgroups, and their experience may not have been uniform.