The all smokers cohort smoked an average of 15.9 cigarettes per day in the CTS2 survey. Among these, those who had quit smoking at the time of the HCC2 survey had smoked significantly less at the time of the CTS2 survey (p < .01, 10.7 vs. 17.1 cigarettes per day) than those who had not quit smoking. The ADM smokers cohort smoked significantly more at the time of the CTS2 survey (p < . 01, 17.0 vs. 14.9 cigarettes per day) than the non-ADM smokers cohort. shows the ADM smokers cohort and non-ADM smokers cohort were equally likely to receive smoking cessation counseling (p not significant, 72.9% vs. 69.9%). However, the ADM smokers were less likely than non-ADM smokers to be successful quitters (p < .05, 17.1% vs. 22.0%). The two cohorts also significantly differed in several other characteristics, with ADM smokers more likely to receive exercise counseling (p < .01, 44.3% vs. 34.9%), to be younger, to live in the West, to have lower levels of education, to have lower levels of income, to be not married, to be unemployed, to be uninsured, and to be less physically active.
Successful Quit Status and Sociodemographic Characteristics for All Smokers, ADM Disorder Smokers, and Non-ADM Disorder Smokers
shows the results from probit regression analyses of successful quitting for the all smokers cohort and the separate ADM smokers and non-ADM smokers cohorts. In the analyses without using the instrumental variable, there was a negative significant association between receipt of smoking cessation counseling in the past year with successful quitting (coefficient = −1.04, p < .01 for all smokers; coefficient = −0.93, p < .01 for ADM smokers, coefficient = −1.16, p < .01 for non-ADM smokers). The Durbin–Wu–Hausman specification test could not reject hidden bias in the analysis for all smokers (χ2 = 76.68, p < .01), for ADM smokers (χ2 = 54.04, p < .01), or for non-ADM smokers (χ2 = 52.59, p < .01), which suggests that using an instrumental variable approach to address hidden bias is appropriate.
Estimated Probit Regression Models of Successful Quitting
When exercise counseling was included as an explanatory variable instead of smoking cessation counseling in the regression analyses of quitting, exercise counseling had a positive association with smoking cessation status for all smokers (coefficient = 0.19, p < .05), for ADM smokers (coefficient = 0.25, p < .10), and for non-ADM smokers (coefficient = 0.19, p < .10). In the first-stage regression model of the instrumental variable analyses in which smoking cessation counseling was as a function of past year PCP exercise counseling and other covariates, the χ2 test was 34.16 for all smokers, 13.35 for ADM smokers, and 24.26 for non-ADM smokers, suggesting that exercise counseling was a valid instrument. In the second-stage regression model, when exercise counseling was used as an instrumental variable for smoking cessation counseling, the predicted smoking cessation counseling by PCPs had a positive significant association with quitting for all smokers (coefficient = 0.88, p < .01), for ADM smokers (coefficient = 1.06, p < .01), and for non-ADM smokers (coefficient = 0.86, p < .01).
We used the probit regression results estimated by the instrumental variable approach to generate the predicted probabilities of quitting for two hypothetical situations: (a) if no study individuals received past year PCP smoking cessation counseling or (b) if study individuals all received past year PCP smoking cessation counseling. The predicted probabilities of quitting without smoking cessation counseling were 9.2% (95% CI: 6.1%–13.4%) for all smokers, 6.0% (95% CI: 2.9%–11.3%) for smokers with ADM disorders, and 10.5% (95% CI: 6.4%–16.3%) for smokers without ADM disorders. The predicted probabilities of quitting with smoking cessation counseling were 32.7% (95% CI: 22.6%–44.2%) for all smokers, 31.3% (95% CI: 16.1%–50.5%) for smokers with ADM disorders, and 34.9% (95% CI: 22.1%–49.6%) for smokers without ADM disorders.
Sensitivity analyses that included the changes in cigarette price between the time of the two surveys showed no change in the results for the association between smoking cessation counseling and successful quitting (all smokers coefficient = 0.87, p < .01; ADM smokers coefficient = 1.06, p < .01; non-ADM smokers coefficient = 0.85, p < .01). Sensitivity analyses that included specific ADM disorders also showed no change in the results for the association between smoking cessation counseling and successful quitting (all smokers coefficient = 0.98, p < .01; ADM smokers coefficient = 1.17, p < .01).