Estimation of Latent Variable Models for Panel Data via Pairwise Likelihood Methods
Generalized linear latent variable models (GLLVMs) provide a flexible framework for analyzing multivariate longitudinal data, but likelihood-based inference is often computationally prohibitive due to high-dimensional integrals. Pairwise likelihoods based on bivariate densities offer a practical solution. We consider d-order pairwise likelihood estimation for longitudinal GLLVMs with binary responses, using separate maximizations. The d-order pairwise likelihood is decomposed into components for pairs of observations separated by at most d time units. Each component is maximized independently, and the resulting estimators are combined for inference on the full parameter vector. Under standard regularity conditions, the resulting estimators are consistent and asymptotically normal, with a block-structured Godambe covariance matrix. Identifiability issues are discussed, and simulations support the results.
Palabras clave: Pairwise likelihood methods high dimensional integrals binary panel data