Association for Public Policy Analysis & Management

Research has shown minority students and students with disabilities receive suspensions and expulsions at disproportionate rates, but the majority of prior work fails to account for the infraction committed or school attended. This study will contribute to the literature by exploiting infraction-level data on both infractions and consequences, as well as student-level demographics and achievement data. These data represent all students in Arkansas K-12 public schools for three years from 2010-11 to 2012-13.

This study will add to the existing knowledge base by more carefully considering the extent to which discipline is delivered discriminatorily. Finding disproportionalities is one thing, but these could be the result of disproportionate misbehavior rather than discrimination. These infraction-level data allow us to uncover differential treatment for the same offenses, which would be a more troubling result.

This study will focus on three research questions:

Research Question 1: What student demographic characteristics (gender, race, etc.) predict the type and frequency of discipline referral(s) a student receives, if any?

We will utilize logit models to predict the likelihood of being referred for certain types of infractions (fighting, insubordination, etc.) for various student subgroups. The sample includes all students attending Arkansas public schools, regardless of whether they appear in the disciplinary database. These relative likelihoods will indicate disproportionalities in the referral process.

Research Question 2: Holding infraction constant, and controlling for school-level characteristics, what student characteristics predict the type of consequence received (In-School-Suspension, Out of School Suspension, Corporal Punishment, Expulsion, etc.)?

Multinomial logit modeling with student-level data will be used to predict the consequence received, by subgroup, holding infraction constant. Multinomial logit is preferred in this case, because it is not clear how to assign even an ordinal level of severity to consequences. To account for differences among schools and districts, we will control for a variety of school- and district-level factors such as total enrollment, percent minority, percent eligible for free-and reduced-lunch, and region.

Research Question 3: What student characteristics predict the severity of consequence (number of days of suspension, for example) for the same infraction and within the same school?

Pooled ordinary least squares (OLS) regression will be used to estimate the relationship between race and consequences, conditional on being referred for some infraction, while controlling for race, gender, grade, Special Education status, and the order or frequency of infractions. Here, the unit of analysis is an infraction, so students can be present multiple times per year, and clustered standard errors allows for correlation between an individual student’s outcomes.

Previous findings from these data have indicated that holding infraction and school constant, African-American students receive more severe (longer) consequences, resulting in more missed instructional time. In this study, the model will be improved by including a vector of consequence dummies as well as a vector of dummies indicating if that infraction was the student’s first, second, third, etc. offense that year.