Modeling faking good on personality items: An item-level analysis

Research has consistently demonstrated that personality inventories can be faked; however, there is disagreement about what effect faking has on the measurement properties of tests. This research examines the effects of experimentally induced faking on item-level measurement using polytomous item response theory. Military recruits were instructed to complete a personality inventory under 1 of 3 conditions: answer honestly, fake good, or fake good with coaching. A graded response model (F. Samejima, 1969) was fit to items from 3 personality scales. Although there was a large difference in latent personality trait scores because of faking, there were few differences in the functioning of items across conditions. Results of confirmatory factor analyses suggest that faking leads to an increase in common variance that was unrelated to substantive construct variance. Implications for modeling and detection of faking are discussed.