, 2006). However, weight counseling has several characteristics that make it a less appealing instrumental variable for smoking cessation counseling. First, it is generally targeted to overweight or obese individuals; exercise counseling is a more applicable to a broader range of www.selleckchem.com/products/Paclitaxel(Taxol).html individuals since few meet the recommendations for exercise levels (U.S. Department of Health and Human Services, 2008). Second, smoking cessation has been associated with weight gain (Eisenberg & Quinn, 2006; Klesges, Meyers, Klesges, & La Vasque, 1989; Klesges et al., 1997; Williamson et al., 1991), which raises issues of additional hidden bias between weight counseling and smoking cessation status. As a result, we only considered exercise counseling as the potential instrumental variable for smoking cessation advice in this study.

Analysis We conducted probit regressions in 2009 and 2010 separately for all smokers, ADM smokers, and non-ADM smokers cohorts to estimate quitting as a function of past year PCP smoking cessation counseling and other covariates. These covariates included gender, age, race/ethnicity, region of residence, education, nativity, household income, marital status, employment, BMI, health insurance, and physical activity. We then checked for hidden bias between past year PCP smoking cessation and quitting by conducting the Durbin�CWu�CHausman specification test (Durbin, 1954; Hausman, 1978; Wu, 1973). To correct for the hidden bias, we used the instrumental variable approach in a two-stage model.

In the first-stage model of the instrumental variable approach, smoking cessation counseling was specified as a function of past year PCP exercise counseling and all the other covariates specified above. We tested the validity of using exercise counseling as the instrumental variables based on the partial R2, and the ��2 test from AV-951 the first-stage model, which is similar to the F test from the first stage regression (Staiger & Stock, 1997); the Sargan test was not performed since we only used one instrument (Greene, 2008). In the second-stage model, quitting behavior was estimated as a function of the predicted smoking cessation counseling from the first-stage model and all the other covariates specified above. The probit model without using the instrumental variable and the probit model with the instrumental variable were estimated by using the probit and ivprobit functions in Stata 10 (StataCorp, 2007). Statistical significance was defined by a p value < .05. We also conducted sensitivity analyses that controlled for logarithmically transformed changes in cigarette price between the time of the CTS2 survey and the time of the HCC2 survey and the addition of variables for presence of the specific ADM disorders.