A study of polytomous IRT methods and item wording directionality effects on perceived stress items

This study investigated the effects of polytomous item response theory models and item wording direction on estimated latent trait scores. The original version of the 10-item Perceived Stress Scale (Cohen, Kamarck, & Mermelstein, 1983) was administered to the sample, along with 10 matched, reverse-worded items. Model fit indices, test information, standard errors, and trait scores from three polytomous IRT models were compared for responses to the set of 20 items (10 positive and 10 negative items), and the subsets of 10 positive, 10 negative, 10 original items, and 10 reverse-worded items. The nominal response model may provide more maximal for those with a trait score in the middle of the distribution, but the graded response model provides more information (and smaller standard errors) of trait scores across a larger span of the scale. Scores were highly correlated, but those from subsets of positively worded and negatively worded items were only moderately correlated. The use of scales with all positively worded items was statistically more favorable, but further research is necessary to determine favorability in practice. The results have implications for researchers and practitioners using psychological scales or developing scales with polytomous responses for diagnostic purposes.