Prevalence rates were estimated in SAS (SAS Institute, 1999) and compared across gender and age cohorts using Stata (Stata Corp, 2003). Latent class analysis, a categorical variant of factor analysis, was performed in MPlus (version 3; Muthen et al., 2004). The maximum likelihood sandwich estimator (MLR) was used to adjust standard errors for clustering of twin data. Therefore, latent class models were fit while simultaneously accounting for the correlations between twin pairs. Additionally, cigarette smokers who were missing responses to certain withdrawal symptoms (e.g., decreased heart rate) were retained in the latent class analyses. Therefore, instead of allowing a listwise deletion to exclude any individual with a missing value for a withdrawal symptom, we used the ANALYSIS=MISSING option for missing data estimation. The Bayesian Information Criterion (BIC) was used to assess model-fit. Logistic and multinomial logistic regression analyses across gender and cohorts examined which factors predicted nicotine withdrawal class assignment in smokers. Analyses were conducted in Stata (Stata Corp, 2003), using the Huber-White Robust Variance Estimator to correct for the non-independence of measures in twins. Variables considered included: difficulty quitting,
