Response choices ranged from (1) never to (4) often, with higher scores indicating higher symptoms INCB018424 of depression. The depression score was the mean of the four items with the standardized Cronbach��s alpha = .76, indicating satisfactory internal consistency. A pilot study conducted with Chengdu, Wuhan, and Qingdao 10th-grade adolescents (n = 1,388) showed a correlation of .74 between this 4-item scale with the 20-item Center for Epidemiological Studies Depression Scale (CES-D; Radloff, 1977, 1991). Depression was dichotomized, with those who scored at the top 20% of symptoms (score ��2.5) coded as being high risk for depression (1) versus low risk (0). Comorbidity Comorbidity of depression risk and smoking was assessed by taking the product of high depression score (high risk = 1) and Wave 1 thirty-day smoking (smoke = 1).

If an individual had high risk of depression and had smoked in the past thirty days, then they were coded as having a comorbidity (CoM = 1) versus all others (CoM = 0). Perceived Friend Prevalence Friend prevalence was assessed by first priming students by asking them how many male/female friends they have and then, of those friends, the number of friends they think have smoked in the past. The mean number of male and female friends who smoke was used for the perceived friend prevalence variable. Friend prevalence estimates were significantly higher among thirty-day smokers (t value = ?10.38, p < .0001) and among those at highest risk for depression symptoms (t value = ?3.60, p < .0004).

Demographic Measures General measures such as age, weekly allowance, academic performance, grade, and class/school attended were also collected. Data Analysis Linear and logistic models were used to test the study hypotheses. Statistical package and procedures, SAS 9.1 Proc GLM and Proc Logistic, were used for analyses. Attrition Analyses The propensity score analysis technique (Austin, Grootendorst, & Anderson, 2007; Grunkemeier, Payne, Jin, & Handy, 2002) was used in prior reports on this cohort (Sun et al., 2007) and duplicated in this study in order to statistically control for possible bias due to unbalanced attrition between the intervention conditions. Propensity for attrition scores was estimated for each participant from logistic regression on boys who participated in the Wave 1 survey.

Age, number of days smoked in the last thirty days, academic performance, weekly allowance, hostility, depression, and program condition were used to predict attrition status (whether participants were reassessed one year later). Only academic performance and weekly allowance were found to Dacomitinib be significant predictors of one-year attrition. The propensity for attrition score was calculated by regressing attrition on program, gender, Program �� Gender, age, depression, thirty-day smoking status, education, academic grade, and weekly allowance.