Suspensions Suspended: Do Changes to High School Suspension Policies Change Suspension Rates?

ABSTRACT

In recent years, the frequent use of suspensions and the racial disparities in their application, particularly for nonviolent behaviors, has created a maelstrom of public pressure for schools to adjust their suspension practices. In an era of increasing institutional accountability for schools, there is evidence that schools may be responsive to policy shifts when they are under institutional pressure to do so. Several school districts have recently revised their out-of-school suspension policies, but researchers know little about (a) if these changes in policy actually change students’ odds of suspension and (b) if so, how these changes might shift racial disproportionality in suspensions. This analysis examines the recent removal of suspensions for low-level infractions from the formal school discipline policy of a large, urban district. I use student-level data to compare the frequency and disproportionality of suspensions before and after the discipline code change. Findings suggest that although suspension rates decrease overall, multiple suspensions per student are more likely. With regard to disproportionality, black girls and white boys are more likely than expected to receive a first suspension in the post period as well. These findings highlight the importance of exploring heterogeneity in outcomes resulting from potential unintended consequences of policy change.

Notes

1 See Appendix A1 and A2 for details on which codes were identified for removal from mandatory suspension.

2 There are also changes to wording for some infractions and some infractions change numbers across years (see Appendix A2 for details). These shifts have been adjusted such that all infractions retain the same number during the entire analytic period. Additional changes primarily include the addition of technology across the analytic period (as can be seen for code 19 in 2014 in Appendix A2).

3 The discipline code, across all years including the postperiod, allows for the suspension of a student at any level after the student is removed from any classroom by a teacher three times.

4 Engaging in persistent “Level 1” behavior (cutting class, being disruptive, etc.) is removed altogether in the 2013 policy change and replaced with inappropriate use of technology that also is thus never allowed as a “suspendable” infraction.

5 There is little reason to assume that behavior changes across time as the frequency of stable codes do not over time (contact author for more details).

6 I limit the analysis to ninth graders for several reasons. First, the compounding effect of suspensions across the high-school trajectory means experiencing a suspension in ninth grade would put a student at increasing risk for future suspensions in subsequent years. This increased likelihood would need a specific modeling strategy for students not in ninth grade. By limiting the analysis to first-time ninth graders, I also address concerns over differential attrition for students who are suspended and remove the possible association between repeating ninth grade and suspension likelihood. Also, this restriction allows a focus on suspensions for students with the same level of exposure to the school's disciplinary climate mediating concerns about differential school-student interactions due to repetition of ninth grade or mobility across high schools. Finally, given the associations between OSS negative outcomes, the potential of a policy to minimize OSS early in a high school trajectory may have the greatest impact.

7 To focus the analysis on comparable schools, I limited this analysis to students attending schools serving only ninth through 12th graders leaving an analytic sample of 348,956 students and approximately 1,900 school years.

8 To maintain continuity with other analyses of Department of Education data, I assigned overage to students who attend ninth grade at 15 years of age or older and identified students as chronically absent if they attended 89 percent or less of the school year.

9 Students missing attendance were assigned the school mean attendance rate for eighth graders within the year and assigned a flag for missing attendance (before being assessed for Chronic Absentee).

10 Students missing eighth-grade test scores were assigned their seventh-grade test scores and assigned a flag for missing these scores. If they were also missing seventh-grade test scores, they were assigned the mean unstandardized eighth-grade test score for their school in their eighth-grade year before standardization. Students who did not attend district schools until 9^th grade were assigned the average eighth grade value for their high school. Sensitivity tests were conducted assigning 0 instead of the mean for test scores with no meaningful variation in results.

11 Suspensions may have multiple infractions assigned but these comprised fewer than 5% of all suspensions. In these cases, I assigned the highest infraction level to the suspension.

12 Total suspensions and any suspensions were also examined but omitted for parsimony.

13 Rescinded codes are infraction codes 14 to 20 as can be seen in Appendix A and were the most minor infractions eligible for suspension preceding the 2012-13 policy change.

14 Attendance rates were calculated after removing suspended days from the denominator.

15 I conduct all analysis in a logistic regression framework.

16 Analyses, not shown here as they are beyond the scope of this analysis, have also been conducted with the standard errors clustered at the school-year and school-level characteristics represented.

17 For all analysis, I identify students suspended two or more times because the number of students suspended multiple times is very small (those suspended twice comprise 11 percent of all students suspended, those suspended three or more times six percent).

18 Models have also been explored using a static sample and clustering school-year standard errors – results are similar to those presented here.

19 Estimates are calculated separately for each suspension-type. Probabilities are displayed by race/gender but race/gender are not interacted with the post-period in this model.

20 Contact author for results of regressions by analytic period.

21 No affected codes (codes 35-45 and 50) are rescinded codes from the 2013 policy change.

22 Not shown here, contact author for more details.