Finally, the outcome of interest was gleaned from a question asking about cigarette use in the past 30 days, to which respondents could answer: never use, have used but not in the last 30 days; 1�C2 days; 3�C5 days; 6�C9 days; 10�C19 days; 20�C29 days; used daily. sellckchem These responses were combined into smoking statuses of never-smokers (i.e., never used cigarettes), ever-smokers (have used but not in the last 30 days), and current smokers (those who reported any smoking in the past 30 days); this definition of current smoking has been used with other national studies of youth and young adult smoking (Ling, Neilands, & Glantz, 2009; Wechsler, Rigotti, Gledhill-Hoyt, & Lee, 1998). The smoking status recoding process was based on an iterative category reduction process through assessment of the parallel regression assumption for ordered logistic regression analyses (see below).

Bivariate differences were assessed using chi-square tests of independence by sexual orientation (e.g., bisexual vs. heterosexual) and by gender and sexual orientation (lesbians vs. heterosexual women). Based both on significant bivariate differences within the sample and a robust literature that documents gender and sexual orientation differences in violence victimization (Faulkner & Cranston, 1998; FBI, n.d.; Herek, 2009; Russell, Franz, & Driscoll, 2001) and discrimination (Hatzenbuehler et al., 2010; Mays & Cochran, 2001), a series of stratified multivariate models were conducted based on sexual orientation and gender by sexual orientation.

Moreover, stratified models allowed to better answer the driving research question of examining the heterogeneity within sexual orientation groups. Multivariate ordered logistic regression models were used to test the association between key independent variables and the proportional odds of smoking status (never-smoker, ever-smoker, and current smoker). Ordered logistic regression procedures assert that an outcome with more than two categories has a hypothesized, but unquantifiable, hierarchical order among categories (Long & Freese, 2006). The outcome of smoking status in this analysis maintains that never-smokers would represent an absolute zero use, that ever-smokers represent at least some use that is greater than never-smokers but less than current smokers, and that current smokers represent the most use.

The proportional odds ratios are interpreted as the likelihood of being in one of the three smoking status categories that indicates more smoking according to the aforementioned three-category ordinal structure (e.g., whether respondents who indicate victimization have increased odds of being in a category representative of more smoking when compared with Anacetrapib a reference group of respondents who do not indicate victimization).