Is all fit that glitters gold? Comparisons of two, three and bi-factor models for Watson, Clark & Tellegen's 20-item state and trait PANAS

Despite widespread use, few studies have evaluated the factor structure of the Positive and Negative Affectivity Scales (PANAS) in adults. Following renewed interest in bi-factor formulations, such a model has recently been suggested as fitting PANAS data better than earlier two- or three-factor models. Study 1 compared four models using responses from Australian adults in two timeframes (state, n = 1059; trait, n = 1333). Results showed an uncorrelated two-factor structure. Three factors improved the model fit somewhat. However, Gaudreau et al.'s model yielded two highly correlated negative factors, while the bi-factor model revealed a weak and non-replicated general factor. Study 2 (n = 334) investigated the concurrent validity of the four PANAS models in the context of Eysenck's and Grey's fundamental personality models. Gaudreau et al.'s third factor and the general factor in the bi-factor model were shown to reflect neuroticism/behavioural inhibition. Further investigation in the context of the tripartite model found no unique role for the afraid factor and a switch from NA to the general factor in the prediction of anhedonia, depression and anxiety. Despite the allure of marginally better fitting three-factor models, Ockham's razor suggests that the two-factor solution remains the preferred model for the PANAS.