«The Social Control of Childhood Behavior via Criminalization or Medicalization: Why Race Matters DISSERTATION Presented in Partial Fulfillment of the ...»
cover many commonly diagnosed behavior disorders, including oppositional defiant and conduct disorders and over 80% of all children covered under the “other health impairment” have been clinically diagnosed with ADHD (Frick and Nigg 2012; Holler and Zirkel 2008). Importantly, teachers and administrators making decisions about disciplining a child need only consider whether the misbehavior is due to the child’s diagnosed disorder and not whether the child was diagnosed with any disorder (Kim, Losen, and Hewitt 2010). Moreover, if children are diagnosed with learning disorders or physical, yet have untreated or undiagnosed psychological or behavior problems, they may be at an increased risk of school failure and potential disciplinary problems (Kim, Losen, and Hewitt 2010; Schifrer 2013). Consequently, I chose to focus on how schools construct their students’ primary behavior problems through coverage under IDEA for only those problems related to behavior. Additionally, because IDEA and Section 504 require different criteria for services and rely on different funding mechanisms, I chose to run all analyses on the two variables separately.
Independent Variables The goal of this paper is to examine the relationship between racial composition and rates of criminalized and medicalized of school discipline at the school- and district-level influence. Therefore, the central school-level independent variable captures the proportion of the school student body that is African-American (percent AfricanAmerican). The central district-level independent variable captures the proportion of the district that is African-American (percent African-American). Finally, to examine whether the association between school-level racial composition and school discipline varies across districts with varying racial compositions, cross-level interactions of schooland district-level percent African-American are included for all four dependent variables.
School-Level Control Variables In addition to school racial and ethnic composition, several school-level control variables are included. Following prior school-level research (Welch and Payne 2010), I measure socioeconomic status using the percentage of students in the school receiving free or reduced lunches (Percent free and reduced lunch). To control for Latino immigrant composition at the school-level, I include an index composed of the average of the summed z-scores for two variables that measure the percent of the school that either Latino or considered limited-English proficient (α =.73). To control for serious or criminal behavior on school groups, I include a dummy variable equal to one if the police had to remove or arrest a student on school grounds during the school year (studentpolice contact). To control for other school-level demographic and organization factors, I include variables measuring the percentage of the student body that is male (percent male) and the student-teacher ratio, which is logged to handle issues of skewness.
Finally, I include dummy variables equal to one if the school is either a charter or a magnet school (Charter/Magnet school) and if the school has a gifted and talented program. To capture school locality, I include a series of dummy variables equal to one if the school is located in a large urban area, small to medium urban area, small town, or rural area (suburb was reference category).
District-Level Control Variables To account for district-level socioeconomic status, I created a measure of district disadvantage using an index composed of the average of the summed z-scores for five variables that measure the percent of the school-district that: holds less than a high-school degree; is out of the labor force; living in single-mother households; living in households on public assistance; and living in households receiving SNAP benefits (α =.78). To control for population turnover at the district level, I include a measure of residential instability using an index composed of the average of the summed z-scores for the percent of the school-district that are renter occupied and the percentage of the district that lived in a different school district prior to 2005 (α =.78). To control for Latino immigrant composition at the district-level, I include an index composed of the average of the summed z-scores for three variables that measure the percent of the school-district that: are Latino; are foreign-born; arrived in the United States since 2005; and speak English “less than well” (α =.93).
Finally, I include three variables capturing the occupations of adults living in the school district. To control for professional/managerial occupations, I include an index composed of the average of the summed z-scores for four variables that measure the percent of the school-district that work in finance, information, professional, or managerial employment (α =.81). To control for service sector employment, I include an index composed of the average of the summed z-scores for the percent of the schooldistrict that are employed in sales and the percentage of the district that are employed in retail (α =.46). Additionally, I include a variable indicating the percentage of the school district employed in the manufacturing sector (percent manufacturing). I include measures of federal funding (IDEA funding and Safe Schools Act funding), state funding (special education funding), and district level funding measures (local funding and median home value). Each variable is logged to handle issues of skewness. Finally, I include dummy variables to capture geographic location of the school-district. I also include a series of dummy variables to capture Census region (South was reference category).
Analytic strategy To examine criminalized and medicalized school discipline as a function of school and district-level racial composition, this project employs random intercept models with schools at level-1 and districts at level-2. Because the variances of the count dependent variables are all considerably larger than the means, I control for overdispersion by running negative binomial models (Cameron and Trivedi 1998). By specifying these counts with variable exposure by school student body, the analysis becomes one of rates of discipline across schools (Osgood 2000). Additionally, because this project is interested in the association between racial composition and school discipline at both the school- and district-level, percent African-American at the school-level is group-mean centered (Enders and Tofighi 2007; Kreft and De Leeuw 1998; Raudenbush and Bryk 2002). In group-mean centering, the values of level-1 explanatory variables are centered around the mean value for each level-2 group. For this project, school-level percent African-American is centered around the mean value for all schools in a given district. In the example, all values of percent African-American at the school level are centered around the mean percent African-American for all schools a given district.
In this model, η ij logged expected count of students who were either suspended or expelled or provided services under IDEA or Section 504 for school i in district j.
Looking at the central level-1 independent variable in this model, %AA ij represents the percentage of African-American students in school i in district j and %AA. j represents the average percentage of African-American students among all schools in district j. As a result, β 1 represents the within-district relationship between school-level percent AfricanAmerican and school discipline, or the expected difference between two schools in the same district that vary by 1 percent African-American and β 2 represents the betweendistrict association between percent African-American and school discipline, or the expected difference in mean counts of the dependent variable across two different districts that vary by 1 percent African-American.
Unlike level-1 variables that are measured in their original metric or centered on the overall mean (grand-mean centering), group-mean centered variables are uncorrelated with all level-2 variables (Enders and Tofighi 2007; Kreft and De Leeuw 1998;
Raudenbush and Bryk 2002). In the case of highly correlated level-1 and level-2 variables that are measured using the same construct, coefficients for variables in their original metric or grand-mean centered represent difficult to interpret effects of the combination of both level-1 and level-2 variables (Enders and Tofighi 2007; Kreft and De Leeuw 1998; Raudenbush and Bryk 2002). This is particularly true when level-1 and level-2 measures represent slightly different concepts (Enders and Tofighi 2007).
Because group-mean centered level-1 variables are uncorrelated with level-2 variables, their coefficients represent the “pure” estimate of the effect of level-1 variables (Enders and Tofighi 2007; Kreft and De Leeuw 1998; Raudenbush and Bryk 2002). Further, cross-level interactions between level-1 and level-2 predictors measured in their original metric or grand-mean centered variables may confound correlation with moderation, presenting statistically significant findings when there are none (Enders and Tofighi 2007; Hoffman and Gavin 1998). On the other hand, because there is no correlation between level-1 and level-2 variables using the group-centered approach, models including cross-level interaction represent the true moderating influence of district-level racial composition on the association between school-level racial composition and school discipline (Enders and Tofighi 2007).
Results Table E.1 presents descriptive statistics for all variables used in the analysis for elementary schools, middle schools, and high schools in the United States. These descriptive statistics reveal several telling patterns regarding school discipline in the United States. Nearly 87 out of every 1,000 American elementary and middle schools suspended at least one student during the 2009-2010 school-year. Conversely, only about 9 out of 1,000 American elementary and middle schools reported covering used IDEA or Section for emotional or behavior problems, respectively. Turning to the school- and district-level independent variables, elementary and middle school students in the United States have student bodies that are, on average, almost 18 percent African-American. At the same time, the average school district in the United States is around 12 percent African-American.
This project controls for a number of important social and economic variables at both the school- and district-level. Notably, slightly more than 52 percent of the students attending American elementary and middle schools are covered under a free and reduced lunch program. Furthermore, seven percent of schools reported that a police officer had removed or arrested a student on school grounds. Turning to important district-level variables, there is a greater amount of state and federal funding going toward medicalization than supporting punitive disciplinary policies. While school districts receive just $1.10 per student in Safe Schools Act funding, they receive, on average, $51.59 in federal IDEA funding and $55.91 in additional state funding for services.
These differences speak to the relative affordability of school punishment compared with medicalization and the importance of human and social resources capable of actually spending such funds appropriately.
Table E.2 presents the coefficients and standard errors for random intercept negative binomial models (with variable exposure) of school punishment (suspension and expulsion) and medicalized school discipline (IDEA or Section 504 enrollment) for elementary and middle schools in the United States. For each dependent variable, Model 1 includes all school and district-level predictors and Model 2 includes a cross-level interaction between percent African-American at the school and district level. To examine whether an association between racial composition and different types of school discipline exists at both the school- and district-level, I begin with a discussion of Model 1 for each dependent variable. Turning to rates of suspension and expulsion, there is evidence of an association between racial composition and school punishment at both the school- and district-levels. At the school-level, elementary and middle schools with larger African-American student bodies relative to other schools in their districts are more likely to use some form of school suspension during the school year. On average, schools with one standard deviation larger percent African-American (13.3 percent) than other schools in their district have 12 percent [100*(e.008*13.3)-1] higher suspension and expulsion rates, controlling for other relevant school and district-level factors. Holding the racial composition of the schools constant, districts with one standard deviation larger percent African-American (15.3 percent) have 12 percent [100*(e.004*15.5)-1] higher average school rates of suspension and expulsion than other districts with relatively smaller African-American populations.
Turning to the medicalization dependent variables, schools and districts with relatively larger African-American populations are less likely to use services for behavior problems as mandated by either IDEA or Section 504. First, the association between racial composition and the use of IDEA is insignificant at the school-level. On the other hand, a one standard deviation (15.5 percent) difference in district-level percent African-American is associated with 20 percent [100*(e.018*15.6)-1] lower rate of IDEA services at the district-level. On the other hand, schools with one standard deviation greater percent African-American (13.3 percent) than other schools in their district are 5 percent less likely [100*(e.-011*13.3)-1] to provide services under Section 504. In addition, districts whose African-American population is one standard deviation higher (15.5 percent) have 15 percent [100*(e-.039 *15.6)-1] lower rates of Section 504 usage than districts with relatively smaller African-American populations.
There are a number of control variables are significantly associated with the use of different types of school disciplinary measures. Given that significant associations between school and district-level racial composition and differences in criminalized and medicalized school discipline remain after controlling for important socioeconomic and funding variables, the discussion will center on disadvantage and federal, state, and local funding variables.