«The Social Control of Childhood Behavior via Criminalization or Medicalization: Why Race Matters DISSERTATION Presented in Partial Fulfillment of the ...»
For this project, I reduce my sample to the male children of African-American and White mothers who were between the ages 6 to 14 during the years 1988 to 2010. I focus on young boys because they are overwhelmingly more likely to be suspended or expelled and diagnosed with behavior disorders than their female peers (Bertrand and Pan 2013; CDC 2012). Furthermore, many of the mechanisms behind the social construction of misbehavior vary across gender lines. For example, teachers and administrators are more likely to view boys as defiant and disruptive than girls (Newcomb et al. 2002; Skiba et al. 2013) and girls’ diagnosed behavior problems are typically associated with attention The age range for the entire NLSY-CYA is 0-31 years old. However, since the age range for the current chapter is 4 to 14, all children’s variables are taken from the Child Survey.
deficit or anxiety and not hyperactivity or disruptive behavior (Cuffe, Moore, and McKeown 2005). After removing the boys who were missing information on the independent and dependent indicators of interest, my final sample includes 3,631 boys who contributed 11,802 person-years to the analyses described below.
Dependent Variable To construct a dependent variable that indicates whether the NLSY respondent’s behavior was medicalized or criminalized, I create the following categorical measure of social construction: (1) neither punished nor therapy/medication; (2) Therapy/Medication only; (3) Punishment only; (4) both punishment and therapy/medication.
Therapy/Medication includes those boys who received therapy or medication for behavior problems in the past year, taken from the child’s response to one of two questions: (1) has child seen a psychiatrist or psychologist for troubles in school or for tantrums, hyperactivity, or disruptive behavior in the previous year? and/or; (2) is child currently taking drugs to control his/her behavior? 2. School punishment is measured using the Mother’s response to the question “Has your child ever been suspended or expelled from school?” The question does not refer to any specific drug or behavior. However, behavior problems like hyperactivity are the most commonly diagnosed disorders in childhood, particularly for young boys.
Moreover, the NLSY-CYA asks about medication for other common ailments but not hyperactivity or other behavior problems. Finally, this measure has been used in prior analysis on childhood behavior problems (Currie and Stabile 2006; Currie, Stabile, and Jones 2014).
Respondent’s race is based on based on the report of the mother, taken from the NLSY Survey, and is coded as 0 if the NLSY Child identifies as nonHispanic White and 1 if he identifies as nonHispanic Black. I use the survey year to capture the increases in the frequency of both school punishment and medically diagnosed behavior disorders over the past 25 years. Furthermore, while increases in both school punishment and the use of therapy or medication for behavior problems were relatively sharp during the 1990s, this growth has slowed somewhat since the turn of the century. To capture the nature of this increase, I include measures of year and year-squared 3.
Childhood misbehavior also figures prominently in the analyses. To capture misbehavior, I use the externalizing behavior scale adopted from the Child Behavior Checklist (CBCL) and derived from the Behavior Problems Index (BPI) (Guttmannova, Szanyi, and Cali 2007). Externalizing behaviors are those behavior characterized by a lack of emotional control or an inability to suppress impulses, leading to rule breaking (Guttmannova, Szanyi, and Cali 2007; Parcel, Campbell, and Zhong 2012). Importantly, the CBCL externalizing behavior score is consistent across and within racial and ethnic groups, and has been recommended for cross-group comparisons (Guttmannova, Szanyi, and Cali 2007) 4. A full list of variables in the externalizing behavior scale is available in To provide sensible estimates for race * time interactions, my year variable begins at zero and is measured every two years until 22. As a result, a value of zero for year is equal to the calendar year 1988.
Guttmannova, Szanyi, and Cali (2007) compare their CBCL measure of externalizing behavior scores with the Behavior Problems Index score created by Parcel and Menaghan (1988) and argue that the BPI score is biased and lacks construct validity across the three predominate racial and ethnic groups in the Appendix G. As the list shows, the externalizing behavior scale includes a number of behaviors that could possibly lead to school discipline, including getting into trouble with teachers, being disobedient at school, and bullying or being cruel to others. To note, these behaviors are also listed by mental health professionals as “symptoms” of childhood behavior disorders. For example, cheating/lying and bullying are included in some conduct disorder symptom checklists, disobedience is often a sign of Oppositional Defiant Disorder, and confusion, restlessness, and inattention are considered to be classic ADHD symptoms.
I include a number of additional control variables in regression analyses to statistically control for potential confounders. First, to control for changes in disciplinary and educational expectations between elementary and middle school, I include a dummy variable equal to one if the NLSY child is still in elementary school. To account for prior difficulties in school, I include a dummy variable equal to one if the boy ever repeated a grade due to academic issues. I measure child’s academic achievement using the child’s standardized score on the PIAT Reading Recognition and Mathematics tests, designed to capture the boy’s mastery of basic skills taught in public school. Additionally, I include a dummy variable equal to one if the child has been enrolled in Head Start. To control for the effects that living in poverty has on behavior, education, and health care, I include a dummy variable equal to one if the total household income was less than the poverty NLSY-CYA. In a series of sensitivity tests, I ran models using the BPI score, as well as the following other measures of childhood misbehavior: oppositional action (Cooksey, Menaghan, and Jekielek 1997), total child delinquency (a scale based on the child’s self-reported answers to seven different questions about deviant and/or illegal behavior between the ages of 10 and 14), and school delinquency (a scale based on the four behaviors most related to in-school activities). These models yielded results very similar to those reported in this chapter and are available from the author by request.
level. To capture other aspects of family socioeconomic status, I use measures of mother’s education (in years) and a series of dummy variables indicating if the mother was unemployed (reference), employed part-time, or employed full-time. Because access to health care may influence decisions regarding therapy and medication, I include a series of dummy variables indicating whether the child is covered under a private insurance plan (reference), public insurance plan, such as Medicaid, or has no insurance coverage. To account for variation in family composition, I include measures of mother’s marital status (currently married, cohabiting, single, or never married), whether or not the child lives in a single-mother household, and the number of siblings living in the home. To capture the disciplinary environment, cognitive stimulation and emotional support provided by the boys’ primary caregiver(s), I include the Home Environment Score taken from the interviewer’s assessment during the Home Inventory Scale.
Additional time-varying variables include, respondent’s age (in years), whether the respondent lived in a suburban, rural, or urban residence, the region of the country in which the respondent resides (Northeast, Midwest, South, or West). I also include timeinvariant measures for mother’s age at birth (dummy variable equal to “1” if the mother was under 18 years old), birth order, and whether or not the mother smoked during pregnancy.
Analytic Strategy To capture the social construction of misbehavior in young African-American and White males in the United States, I employ multinomial logistic models. To account for heteroskedasticity and nonindependence of error terms, I estimate robust standard errors clustered at the level of the individual. The multinomial logistic model can essentially be thought of as estimating simultaneous binary logit models for all possible comparisons among the outcome category (Long 1997). As demonstrated in the equation presented above, the outcome represents the log-odds of each individual boy falling into category m relative to the base category b.
In a series of regression models, I test the likelihood of punishment, therapy/medication, or both punishment and therapy/medication versus experiencing no labeling event. However, because I am examining these possible outcomes in the same model, I am able to make comparisons across all possible choices (e.g. likelihood of punishment versus therapy/medication). Thus, unlike prior work that considers only one form of social construction (e.g. school punishment versus no school punishment), I am able to make comparisons across a broader, more exhaustive range of potential outcomes (Box-Steffensmeier and Jones 2004; Long 1997).
To test my hypothesis, I run a series of four models. First, to test for changes in the social construction of childhood behavior over time, I run models including only measures of time (year and year-squared) and control variables. Second, to test for racial differences in how the behavior of children is socially constructed, I add a race dummy variable that identifies whether the respondent was nonHispanic Black or nonHispanic White. Third, to test whether racial disparities in the frequency of behavior problems can explain differences in how behavior is social constructed, I add a variable measuring the frequency of externalizing behavior symptoms. Finally, the fourth model uses race by year (and year-squared) interactions to capture whether there are racial disparities in how the behavior of children has socially constructed over time.
To handle issues of missing data, I use multiple imputation techniques to generate values for all covariates using the “ICE” command in Stata (Royston 2005). ICE relies on a chained equation approach in which a conditional distribution for missing data using the appropriate specification for each variable (e.g., logistic regression for dichotomous variables) and multiple datasets are created using Gibbs sampling techniques (Royston 2005; van Buuren 2012). I created five distinct data sets 5, which were all used in conjunction with the mi command in Stata to complete both descriptive and multivariate analyses. Following von Hippel (2007), I impute values for all variables, including the interaction terms, in a given model and then delete observations with missing data on either behavior or dependent variables before running our regression analyses.
Typically, the number of imputed datasets is dependent on the amount of total missing information, with 3 to 5 datasets being a common recommendation for models containing up to 20% missing information, far greater than the 5% missing information in the current analysis (Royston 2005;
Rubin 1987; van Buuren 2012).
Results Table A.1 presents weighted means and proportions for the variables included in the study for African-American and White boys. These descriptive statistics reveal clear racial disparities in how misbehavior is socially constructed. Almost nine out of ten White boys receive no label for their behavior, compared to a little over three-quarters of African-American boys. Approximately 16 percent of the African-Americans report having been suspended without receiving therapy or medication in a given year, compared to only 3.4 percent of White boys. Meanwhile, 7.2 percent of White boys reported receiving therapy or medication for behavioral disorders in a given year, compared to just 4 percent of African-Americans. In addition to racial disparities in school punishment and therapy or medication, African-American boys display, on average, significantly higher levels of externalizing symptoms than White boys. The average score on the CBCL externalizing behavior scale is 6.12 for African-American boys and 5.5 for White boys.
In addition to differences in my dependent and key independent variables, there are a few other noticeable racial disparities on important covariates. African-American boys are significantly more likely to have repeated a grade, score significantly lower on the PIAT Reading Recognition and Math achievement tests, and are more likely to attend Head Start. Furthermore, African-American boys are almost four times as likely to be raised in a poor household and are more likely to be raised by a single mother. These disparities suggest that African-Americans in this sample are placed at significant structural disadvantage when it comes to their education, potentially placing them at a greater risk of school punishment than their White peers.
Table A.2 presents the results from the first multinomial logit model examining the likelihood of punishment or therapy/medication between 1988 and 2010. In Table A.2, coefficients represent the log-odds and exponentiated coefficients represent odds ratios of therapy/medication, punishment, and receiving both therapy/medication and punishment versus receiving no label. Results from Table A.2 provide support for the first hypothesis. Specifically, there are significant and sizable increases in the likelihood of both school punishment (suspension or expulsion) and the use of therapy or medication for behavior disorders between 1988 and 2010. On the other hand, looking at the odds of school punishment versus no labeling, the significant and negative coefficient for year-squared suggests that the increase in suspensions and expulsions has slowed recently. These trends can be observed in Figure B.1, which displays predicted probabilities of punishment or expulsion and therapy or medication between 1988 and
2010. As Figure B.1 demonstrates, the likelihood of receiving therapy and/or medication for a behavior disorder without being punished increased steadily over the 22 year period.
Meanwhile, the likelihood of school punishment increased rapidly in the early 1990s with the rate of increase slowly declining over time.