Realism, ontology, and latent variables

Drawing on scientific realism, recent psychometric literature presents a claim for ontology for latent variable analysis. The purpose of such presentation is to underlabourapurported causal relationship between psychological attributes and manifest outcomes, which is said to drive the use of these analyses in psychological research. This thesis examines the principles of this claim in three steps. Firstly, a meta-analysis of three approaches to philosophical realism is conducted with specific focus on causality and relations, clarifying the logical ground of the claim for a realist ontology for latent variables. The outcomes are used in conceptual analysis of the terms „ontology‟, „causality‟ and „latent variable‟, and the assumptions of scientific realism are tested against the principles of philosophical realism. The implications of factor indeterminacy and realist measurement theory for latent variable modelling as measurement theory are set clear. Finally, the conceptual analysis and meta-analysis are brought together to examine a suggestion of interchangeability between observed and latent variables. These three forms of analysis, logical, conceptual and empirical together form a critical inquiry that indicates minimal support for the claim for a realist ontology for the use of latent variable modelling in psychological research.