Early adolescent substance use in Mexico: Identify individual and contextual risk factors through random forest analysis.
Class
Article
College
Emma Eccles Jones College of Education and Human Services
Faculty Mentor
Melanie Domenech Rodríguez
Presentation Type
Poster Presentation
Abstract
The current study sought to identify individual and contextual factors associated with substance use and intentions to use among elementary aged youths. Data for the present study include 52,171 5th and 6th graders who participated in the National Survey of Drug Use Among Students (ENCODE) in Mexico. Self-report of demographic, individual (e.g., school performance, self-esteem), and contextual factors (e.g., neighborhood, deviant peers) were examined as predictors of lifetime substance use and intentions to use. A Random Forest analysis (Liaw et al., 2002) was used to identify variables most relevant to substance use. An exploratory and confirmatory factor analysis were used to collapse identified variables. Multiple imputations (Donders et al., 2006) were then used to replace missing values differently in 5 copies of the dataset. Logistic regressions were then run on all 5 databases and results were pooled into a single output. Preliminary findings yielded 21 variables of highest relative importance to substance use from the original 67. Variables related to demographic and individual characteristics were of low importance to substance use relative to contextual factors (e.g., neighborhood, deviant peers). Neighborhood factors were highly important to alcohol and tobacco use. Items related to deviant peer relations were important to marijuana and inhalant use. Factor analysis on these items yielded four variables pertaining to neighborhood insecurity, positive neighborhood qualities, exposure to deviant peers, and prosocial behaviors that will be included in logistic regression models with demographic characteristics to predict substance use intentions and use. These findings suggest that early substance use may be primarily driven by contextual factors rather than individual and demographic characteristics. Additional analysis, clinical, and research implications will be discussed. Useful strategies for handling issues such as low internal consistency of measures,
Location
The North Atrium
Start Date
4-12-2018 9:00 AM
End Date
4-12-2018 10:15 AM
Early adolescent substance use in Mexico: Identify individual and contextual risk factors through random forest analysis.
The North Atrium
The current study sought to identify individual and contextual factors associated with substance use and intentions to use among elementary aged youths. Data for the present study include 52,171 5th and 6th graders who participated in the National Survey of Drug Use Among Students (ENCODE) in Mexico. Self-report of demographic, individual (e.g., school performance, self-esteem), and contextual factors (e.g., neighborhood, deviant peers) were examined as predictors of lifetime substance use and intentions to use. A Random Forest analysis (Liaw et al., 2002) was used to identify variables most relevant to substance use. An exploratory and confirmatory factor analysis were used to collapse identified variables. Multiple imputations (Donders et al., 2006) were then used to replace missing values differently in 5 copies of the dataset. Logistic regressions were then run on all 5 databases and results were pooled into a single output. Preliminary findings yielded 21 variables of highest relative importance to substance use from the original 67. Variables related to demographic and individual characteristics were of low importance to substance use relative to contextual factors (e.g., neighborhood, deviant peers). Neighborhood factors were highly important to alcohol and tobacco use. Items related to deviant peer relations were important to marijuana and inhalant use. Factor analysis on these items yielded four variables pertaining to neighborhood insecurity, positive neighborhood qualities, exposure to deviant peers, and prosocial behaviors that will be included in logistic regression models with demographic characteristics to predict substance use intentions and use. These findings suggest that early substance use may be primarily driven by contextual factors rather than individual and demographic characteristics. Additional analysis, clinical, and research implications will be discussed. Useful strategies for handling issues such as low internal consistency of measures,