Table 1. Participant Characteristics Factor Structure A confirmatory sellckchem factor analysis was conducted to replicate the structure of the WSWS as reported in Welsch et al. (1999). The initial 28-item, 7-factor model produced an adequate fit (��2 [329] = 802.86, p < .0001; RMSEA =0.06, CFI = 0.92, TLI = 0.91, SRMR = 0.07). Examination of factor loadings and modification indices revealed two potential offending items. The first was item 1, ��food is not particularly appealing to me�� which loaded on the hunger factor in Welsch et al. but did not load on the hunger scale in the current sample, and was significantly associated with all other subscales. Removal of this item produced a model that adequately fits the data (��2 [303] = 709.38, p < .0001; RMSEA =0.06, CFI = 0.93, TLI = 0.92, SRMR = 0.

05), and this 27-item model was a better fit than the initial model as indicated by a smaller AIC (26,205 vs. 26,364) and BIC (26,606 vs. 27,777). The second offending item was item 7, ��I have felt upbeat and optimistic�� which loaded on the sadness factor in Welsch et al. This item loaded adequately on the sadness subscale but was also strongly associated with three other scales (anger, anxiety, and craving). Removal of this item produced a model that adequately fits the data (��2 [278] = 611.08, p < .0001; RMSEA =0.06, CFI = 0.94, TLI = 0.93, SRMR = 0.04), and the model's smaller AIC (25,151) and BIC (25,541) indices indicated that this 26-item model was a better model than the previous two models. Table 2 lists standardized and nonstandardized factor loadings for the 26-item measure.

This 26-item measure was used in the remaining analyses. Table 2. Standardized and Unstandardized Factor Loadings for the 26-Item Wisconsin Smoking Withdrawal Scale Measurement Invariance Consistent with the recommendations of Vandenberg and Lance (2000), the structure of the WSWS was first examined for overall equivalence of measurement (total invariance) across the three racial/ethnic groups. A multiple-group CFA was conducted in which all parameter estimates (i.e., factor loadings, item intercepts, item residual variances and covariances, factor means and variances, and factor covariances) were constrained to equality across the White, African American, and Latino subsamples. This resulted in a model with adequate fit (��2 [1032] = 1561.51, p < .0001; RMSEA =0.064, CFI = 0.

91, TLI = 0.91, SRMR = 0.09), suggestive of overall equivalence of measurement across AV-951 groups. Predictive Equivalence Because of the importance of the implications of null findings in these particular analyses (i.e., null findings are evidence against predictive bias), an analysis of the smallest detectible effect size was conducted. The analysis indicated that the smallest detectable effect (in terms of an odds ratio) was 0.67 and 1.