An Examination of the Factor, Convergent, and Discriminant Validity of the Behavioral Inhibition System and Behavioral Activation System Scales

It was also possible to improve model fit by retaining all of the items in the five factor model and allowing some cross-loadings obtained through modification indices. We do not emphasize this model because: (a) It is more likely to be based on idiosyncrasies of our sample; (b) it results in cross-loadings among BAS scales, rendering scoring of the measure unclear; and (c) the fit indices for this model were no stronger than that of the reduced-item model we present.

It was possible to obtain excellent fit with the reduced item set by including method factors (orthogonal to the substantive factors) such that each BAS item loaded on an overall BAS factor as well as a method factor for its BAS scale. We do not emphasize this solution because no simple scoring system can be derived from it and the fit indices were only slightly superior to our final model (in the full dataset: CFI = .96, TLI = .96, RMSEA = .06, SRMR = .04). Importantly, the model did not have excellent fit using the entire item set. It should also be noted that we did not test whether the resulting method factors measured method variance alone or additional substantive variance that required decomposition. We wish to note the solution for the sake of future scale development, because it suggests that all the BAS items share underlying variance that cannot be accounted for by simple totals of items.

To test if the original BIS/BAS scales outperformed the BIS/BAS Revised scales, we conducted further analyses. Both the original and revised scales were entered into a regression for each outcome variable. Thus, the original and revised subscales were tested against each other directly. Instances in which one version of the scales might be preferred were then totaled. A version of the scale was counted as preferred if: (a) it predicted as hypothesized in the regression depicted in Table 2 or Table 3, and the other scale did not or (b) both scales predicted in the initial regressions, but in a regression using the two versions of the same scale as a predictor, only one scale maintained statistical significance. The original BIS/BAS scales showed stronger relationships in the hypothesized direction than the BIS/BAS Revised scales in 6 of 31 analyses (19%), whereas the revised scales showed stronger fit with hypotheses in 2 of 31 analyses (6%). In the other 23 cases, there was no indication that the original or revised scales should be preferred. There was thus little evidence that the original scales had superior convergent or discriminant validity.