Abstracts Psychology

Factor-Analytic Investigations of Personality Structure: Do Data Take the Shape of Your Measure?

by Lara D Mercurio

Since the early 1960s, the primary tool for identifying individual differences in personality structure has included the use of factor analysis to identify a small but interpretable number of dimensions that summarize the inter-individual differences in the participants’ qualities. Many researchers have interpreted these factor-analytic dimensions as being psychological structures that exist in the mind of individual persons and are causally responsible for the observed variations in psychological characteristics. An alternative interpretation is that the analysis of language-based data primarily yields information about the structure of language. Semantic overlap between items contributes to the obtained correlations among test items and thus influences the resulting factor structure. In principle, the semantic overlap may be sufficient to account for the resulting factor structure. This latter possibility is tested in a novel manner in the present thesis using a computer simulation. At the outset of the simulation, each member of a population of respondents has no personality characteristics (Study 1) or they have a personality structure that is distinct from the most commonly accepted dimensional model of personality structure, the Big Five model (Study 2). Test item responses are then updated as a function of semantic overlap among test items. The empirical question addressed is whether this semantic updating is sufficient, subsequent to factor analysis, to generate traditional factor-analytic personality structures. The computer simulations do indeed show that in both populations, semantic connections between the items are sufficient to reliably produce a factor-analytic structure that largely coincides with the Big Five. These studies suggest that lexical redundancy or semantic overlap within a measure can indeed shape or re-shape the data collected to reflect the structure entailed within the measure, regardless of whether that structure was present in the original population. Advisors/Committee Members: Cervone, Daniel P (advisor).