Creating Scale Measures of Latent Factors: A Genetic Algorithmic Approach

Abstract

Genetic algorithms (GAs) are quantitative optimization techniques that have exclusively been utilized for scale abbreviation despite their potential application to new scale development. Here, we modeled the trait constructs of the triarchic psychopathy framework (boldness, meanness, disinhibition) as latent factors, and then applied a modified GA to select items for assessing each using model-estimated factor scores as targets. Items for the new scales were selected from a separate construct-relevant inventory, the Elemental Psychopathy Assessment, based on their ability to efficiently index each triarchic factor, with consideration given to scale intercorrelations and item polarity. Structural and item response modeling methods were then used to refine the GA-selected item sets. The resultant EPA-Triarchic scales correlated highly with their target factor scores and exhibited stronger loadings than the pre-existing scale indicators when added into the model. This work, illustrating a GA approach to devising new scales for indexing latent factors, has broad potential applications in clinical assessment.