Events at CERGE-EI

Presenter: Mauricio Olivares (LMU)

Title: "Specification Testing for Latent Variable Models."

Abstract: Latent variable models are widely used in economics to study processes such as skill formation, where key inputs are unobserved and proxied by multiple noisy measurements. Identification in these models typically relies on strong independence assumptions between latent variables and measurement errors, which are often difficult to justify in practice. This paper develops a simple, nonparametric specification test for these independence assumptions that avoids estimation of the full structural model. The approach exploits observable implications of the latent structure, translating them into testable independence restrictions among linear combinations of observed measurements. Building on these insights, we propose a testing procedure that remains valid in high-dimensional settings and allows the researcher to screen out the reliable measurements, while remaining computationally tractable. To illustrate its usefulness, we revisit an empirical application to health production in India. Importantly, the proposed test can detect misspecified measurements and improve model credibility by identifying a subset of measurements consistent with indentifying independence assumptions.