Causal Machine Learning with Latent Factors

• In Event: Addressing Challenges in Causality: Free Will, Mediation and Latent Factors, Heterogeneity, and Variable Selection

Abstract

A critical limitation of conventional methods (e.g., SEM) is that they typically provide consistent estimates and valid inferences only under assumptions of linearity and correct model specification (e.g., all non-linear terms/n-way interactions are known and included). Alternatively, machine learning (ML) methods can relax model specification requirements by empirically tracking these complex relationships. However, a significant weakness of extant ML methods is that they do not accommodate reflective latent factors. In this study, we develop a superlearner method for support causal inference when the outcome and/or covariates are latent factors. The results suggest that the proposed methods return nearly unbiased treatment effect estimates in finite samples even when the forms of relationships are nonlinear and/or nonadditive and those forms are unknown/misspecified.

Authors

• Benjamin Kelcey, University of Cincinnati

• Amota Ataneka, University of Cincinnati

• Jean Baptiste Habarurema, University of Cincinnati