An Extended Item Response Tree Model for Wording Effects in Mixed-Format Scales

Likert scales are frequently used in social science research to measure an individual’s attitude, opinion, or perception. Recently, item response tree (IRTree) models have been proposed to analyze Likert-scale data because they could provide insights into an individual’s response process. A Likert-scale survey is often mixed with positively worded and negatively worded items, which might induce wording effects. Therefore, it is of interest to investigate how wording effects function in an IRTree model. In this study, we propose a new model—the bi-factor IRTree (BF-IRTree) model, in which combines an IRTree model and a bi-factor model in an IRT framework—to identify how wording effects influence response processes for negatively worded items. Twelve items of an extroversion construct from the Big Five personality inventory were used for demonstration. Results showed that the wording effects were varied on these negatively worded items.