To address this potential bias, we used the Taylor series selleck chemical Trichostatin A linearization to express the estimates, which has been shown to be equivalent to the replication method (Kish & Frankel, 1974). Logistic regression analyses were conducted to examine the association between cigarette smoking status and smoking policy for daily versus never (referent), experimenter versus never (referent), and daily versus experimenter (referent). Because we anticipated a strong association between cigarette price and cigarette smoking status, we first adjusted the models for sociodemographic characteristics and then ran a second model adding cigarette price. In the analyses, ORs greater than 1 signified a higher likelihood of cigarette smoking in students living in states where the policies were less strict.

Given the number of logistic regression analyses that were conducted, an alpha level of .01 was selected to control for Type I error inflation in analyses. Thus, statistically significant OR in logistic regression will have a nonoverlapping 99% CI. The following sociodemographic categories were used as referents in regression analysis: female, White, college-level parent education, moderate FAS category, and high cigarette price. The FAS and parent education variables were retained in the models, given their small correlation coefficient (rho = .30, p < .0001; Cohen, 1988). In addition, because of the magnitude of the majority of the correlations among the youth access and clean indoor air laws variables (rho > .40), separate logistic regressions were performed.

Results Bivariate analyses: Sociodemographic characteristics and cigarette smoking status Respondents were excluded from the present study if the cigarette smoking behavior question was not answered (n = 1,479, 10%). The deleted sample was significantly different (p < .05) from the analytical sample in all sociodemographic variables. The excluded sample contained more boys than girls (57% vs. 43%); in addition, compared with the analytical sample, proportionately more Blacks (28%) and fewer Whites (49%) were excluded. However, the difference in family affluence was marginally significant (p = .0485; data not shown). The analytical sample (n = 13,339) was 47% male and 64% White, 15% Black, 13% Hispanic, and 8% ��other�� (Table 2). The majority of the sample (85%) never smoked, 10% had experimented, and 5% reported GSK-3 daily smoking. The proportion of experimenters and daily smokers increased with grade (��2 = 153.61, p < .0001) and was higher among boys than girls for both daily smoking (6% vs. 4%) and experimental smoking (11% vs. 9%; ��2 = 32.40, p < .0001). As shown in Table 2, Whites reported the highest proportion of daily smoking, followed by the ��other�� category.