Date of Award
Doctor of Philosophy
The purpose of this research was to examine the efficacy of four models for predicting person reliability, which was defined as across-session item consistency. Two prediction models were based on item characteristics alone, namely, p-values and social desirability scale values. Two prediction models attempted to take into account both item and person characteristics. These models were based on individuals' latencies for responding to particular items and on individuals' thresholds for answering items in terms of some item characteristics. The latter model was derived from Jackson's (1968) threshold model for responding which describes the response process with reference to a threshold, marking an individual's transition from the tendency to reject to the tendency to endorse items relative to some item characteristic.;Models were evaluated in three separate studies. In Study 1, models were compared using test-retest responses to personality items relatively neutral in desirability and measuring normal personality content. Study 2 focussed on the role of item characteristics. Three models, the two based on item characteristics and one on threshold theory, were compared using test-retest responses to personality items measuring psychopathological content and ranging widely in social desirability scale values and p-values. Study 3 emphasized decision difficulty. Decision difficulty was experimentally manipulated through a rating task designed to induce desirable responding. Decision difficulty was also examined by comparing prediction based on subjects' own ratings of item desirability to prediction based on independently derived desirability scale values.;Overall, the response latency model was found consistently to be the strongest predictor of exactly which items individuals changed on retest. Theoretically, this finding was interpreted as support for distinguishing between the influences of item and person characteristics on the process of responding to personality items. Practically, this finding has applications for constructing person reliability indices. Results failed to support the hypothesis that the second model taking into account person and item characteristics, the threshold model, was a better predictor than models based on item characteristics alone. The relative lack of success of this model was considered in light of the need to specify correctly the item property underlying responding.
Fekken, Geziena Cynthia, "Comparison Of Four Models For Predicting Person Reliability" (1983). Digitized Theses. 1297.