Direct versus indirect measures of mixed emotions in predictive models: a comparison of predictive validity, multicollinearity, and the influence of confounding variables.

Abstract

Mixed emotions have been assessed using both direct measures that utilize self-report questionnaires as well as indirect measures that are computed from scores of positive and negative emotions. This study provides a pre-registered methodological examination on the use of direct and indirect measures of mixed emotions in predictive models. Two samples (N = 749) were collected, and path analyses were performed to compare direct measures and indirect measures in predicting psychological conflict, receptivity, and well-being, controlling for demographics, positive emotions, and negative emotions. We also tested whether trait dialecticism, need for cognition, social desirability, or acquiescence could account for these associations. In both samples, results suggest that indirect measures may be more susceptible to multicollinearity when controlling for positive and negative emotions. Specifically, variance inflation factors (VIF) were consistently higher for indirect measures calculated using the minimum index (MIN; VIFSample-1 = 3.53; VIFSample-2 = 9.46) than direct measures (VIFSample-1 = 2.52; VIFSample-2 = 1.68). Direct measures remained consistently associated with increased conflict and reduced coherence upon controlling for positive and negative emotions, while indirect measures remained consistently associated only with increased conflict. We found little evidence that response biases explained associations between direct measures or indirect measures with each of the outcomes. Specifically, associations between mixed emotions with psychological conflict, receptivity, and well-being largely remained unchanged in models that controlled for trait dialecticism, need for cognition, social desirability, or acquiescence. Implications and recommendations based on our findings are discussed.