The Big Five Personality Traits Are Composites Rather Than Common Causes

The Big Five personality traits are often modeled with reflective measurement models such as factor analysis models and item response theory models. In reflective models, traits are conceptualized as common causes of the items. However, two core assumptions of reflective measurement models, unidimensionality and local independence, are implausible because the Big Five do not correspond to five biological, environmental, or mental entities and their items have direct causal effects on each other and semantic overlap. Despite their implausibility, researchers continue to use methods and theories that implicitly or explicitly assume reflective measurement models. I propose a more plausible alternative: a composite-formative measurement model. In composite-formative models, indicators are summed to form a composite variable, without assuming they are caused by a latent trait. This model is better suited for the Big Five, as the true data-generating process is unclear and likely too complex to be modelled comprehensibly. Furthermore, it aligns with the goal of the lexical approach to summarize diverse personality characteristics with a few variables. A shift from a reflective to a composite-formative model implies that researchers should use composite-formative rather than reflective measurement models in structural equation models. Additionally, item retest reliabilities rather than Cronbach’s alpha or McDonald’s omega should be used to estimate reliability and correct for unreliability. For descriptive and predictive purposes, the composite-formative model for the Big Five is at least as useful as the reflective measurement model. However, regardless of the modeling approach, the Big Five are hardly useful for investigating causal effects.