A New Empirical Framework for the Study of Positive and Negative Affect

Four assumptions are commonly made in affect research. Two pertain to the measurement of positive and negative affect (PA and NA): 1) PA and NA have similar, desirable measurement properties, and 2) PA and NA are adequately captured by the same response scale. Two pertain to individual differences in affect: 1) The factor structure of PA and NA is the same across individuals as it is within individuals, and 2) The correlation between PA and NA is the same across as within individuals. This dissertation project demonstrates the fallacy of these assumptions in two longitudinal data sets with different sample characteristics, item sets, periods of measurement, and response scales. Longitudinal item response models (IRMs) anchored across occasions revealed NA measures in both data sets poorly targeted respondents, produced less accurate person scores, and demonstrated more response scale parameter reversals compared to PA measures. IRMs of recoded data contrasting all possible ways of collapsing original response scales indicated a binary collapsed response scale was optimal for both NA measures, and a 5-category scale performed best for PA measures. Data were recoded according to these collapsed response scales and used in person-specific and occasion-specific exploratory factor analyses (EFAs) to challenge individual differences assumptions. EFAs exhibited substantial variation between individuals in
factor loadings and factor correlations, much larger than any variation found between occasions. Results refute the four assumptions examined in two longitudinal data sets. A new framework for affect research is proposed, in which measurement and individual differences are major foci, and recommendations are made for researchers.