The purpose of this secondary analyses study was to examine trajectories of delinquent and criminal behavior from adolescence into young adulthood, test its association with childhood maltreatment, and assess whether hypothesized protective factors affect the link between maltreatment and delinquent and criminal behaviors. Variation by youth's sociodemographic characteristics was also examined.
Child Trends' researchers addressed the following research questions:
- What is the relationship between childhood maltreatment and delinquent and criminal behaviors from adolescence into young adulthood?
- How does the relationship between childhood maltreatment and delinquent and criminal behaviors vary by sex, race/ethnicity, and sexual orientation?
- Do any of the following protective factors decrease the risk that someone who experienced maltreatment would go on to engage in delinquent and criminal behaviors? The protective factors include school connectedness, relationship quality with a mother and/or father, time spent with friends, and neighborhood collective efficacy.
- Do the effects of any of these potential protective factors vary by youth's sex, race/ethnicity, and sexual orientation?
The data files used for this study included Waves I, III and IV in-home Add Health questionnaires along with the longitudinal weights. These files were merged together. Next, researchers created the control, independent, dependent, and moderator variables). When creating scales, researchers standardized the items before summing if the items varied (e.g., some were Likert scales, some were 0 to 3) and then standardized the overall scale score. Investigators also set the scale score to missing if a respondent was missing >25% of items in the scale. With the data and variables prepared, levels of missingness within and across variables were examined. The level of missing never reached 10% on any analytic variable and the sample level of missingness was well below the 20% threshold, so complete case analysis was used. Finally, the data were organized by age (13-30) rather than by wave (Wave I, III, IV) in wide to long data transformation. After the data was transformed, the age variable was squared, and a flag variable for maltreatment frequency at Wave I was created only for use in analyses.
In preparing the Add Health data for analyses, investigators noticed a surprising increase in violent offending frequency at Wave IV for three behaviors specifically. Researchers investigated this pattern in numerous ways and eventually contacted Add Health administrators for further explanation. They indicated that these are mostly implausible values and should be dropped. Code for dropping the implausible values, receiving an updated codebook, and/or receiving an already cleaned Wave IV data file can be received by contacting the data producers directly at email@example.com.
The study used linear mixed effects models to test the research questions. This resulted in 50 models for each of the two dependent variables (violent and non-violent offending frequency):
- Patterns of offending: With a given dependent variable (e.g., violent offending behavior) researchers modeled the base pattern (M1); added covariates (M2); tested for moderation of base pattern by sex, race/ethnicity, and sexual orientation (M3-5, respectively).
- Association between maltreatment and patterns of offending: Next, investigators added an adolescent measure of independent variable (maltreatment) and tested its association with the dependent variable's intercept (M6); and slope (M7); and tested if these associations (depending on what is statistically significant) vary by gender, race/ethnicity, and sexual orientation (M8-10, respectively).
- Moderation of the association between maltreatment and patterns of offending by protective factors: Next, researchers tested if potential protective factors (school connection, time with friends, maternal relationship quality, paternal relationship quality, and neighborhood collective efficacy) alter the connection between maltreatment and later offending in either the intercept (models 11, 13, 15, 17, and 19) or slope (models 12, 14, 16, 18, and 20).
- Sociodemographic variation among protective factors: Finally, investigators tested if the protective factors varied by gender, race, and sexual orientation (M21-50).
This study used data from the National Longitudinal Study of Adolescent to Adult Health (Add Health), a longitudinal study that includes a nationally representative sample of U.S. adolescents who were in grades 7-12 in the 1994-95 school year (Wave I, ages 11-19). Add Health has conducted four in-home interviews to date, with the fifth currently in the field. The present analysis sample is restricted to respondents interviewed at Waves I, III (ages 18 to 26), and IV (ages 24 to 32), who had valid sampling weights (N=12,288) and complete data on all variables of interest (N=10,613, 86 percent). The second wave of data collection did not include students who were seniors in high school in the first wave. Therefore, Wave-II data were excluded from secondary analysis.
The initial Wave I, Stage 1 in-school sample was a stratified, random sample of all high schools in the United States. The in-school questionnaire was administered to more than 90,000 students in grades 7 through 12. The Stage 2 in-home sample of 27,000 adolescents consisted of a core sample from each community, plus selected special over samples. For detailed information on Add Health sampling methods through multiple waves, including information on special oversamples, please visit the Add Health Study Design web page.
Adolescents in grades 7 through 12 during the 1994-1995 school year. Respondents were geographically located in the United States.
Variables used in secondary analyses included:
- violdelsc~en - Frequency of violent offending
- nonvioldel~e - Frequency of nonviolent offending
- schoolconn~z - Standardized scale of school connectedness
- timefriend~n - During the past week, how many times did you just hang out with friends?
- momrelscale - Scale of relationship quality with mother figure
- dadrelscale - Scale of relationship quality with father figure
- nbcolleff_~z - Standardized scale of neighborhood collective efficacy
- longwt - POSTSTRAT GRAND SAMPLE UNTRIMMED LONGIT WGT W134
- region - STRAT VARIABLE - REGION
- psuscid - CLUSTER VARIABLE - SCHOOL ID
- publicassist - Before you turned 18, did anyone in your household ever receive public assistance?
- repeatgrade - Have you ever repeated a grade or been held back a grade?
- suspended_~n - Have you ever been suspended, expelled, or dropped out?
- substanceuse - Have you ever used alcohol, cigarettes, or illicit substances before Wave I?
- foster_clean - Did you ever live in a foster home?
- race - Race/ethnicity
- female - Female
- sexualmino~y - Sexual minority
- age - Age
- maltreat_cat - Maltreatment freque
Response rates for waves I, III, and IV of the Add Health series were as follows:
- Wave I: 79 percent
- Wave III: 77.4 percent
- Wave IV: 80.3 percent
Researchers created scales for secondary analyses by standardizing items before summing if the items varied (e.g., some were Likert scales, some were 0 to 3) and then standardized the overall scale score. Researchers also set the scale score to missing if a respondent was missing >25% of items in the scale.