"Conditional Dependence Models Under Covariate Measurement Error"
Speaker: Dr. Elif Acar, Department of Statistics
Thursday, November 28, 2019 in 111 Armes Lecture Building at 3:45pm
Refreshments will be served between 3:15pm-3:45pm in 318D Machray Hall before the seminar.
In many applications, covariates are subject to measurement error. While there is a vast literature
on measurement error problems in regression settings, very little is known about the impact of
covariate measurement error on the dependence parameter estimation in multivariate models. In
this work, we address the latter problem using a conditional copula model, and show that the
dependence parameter estimates can be significantly biased if the covariate measurement error is
ignored in the analysis. We identify the underlying bias pattern from the direction and magnitude
of marginal effect sizes and introduce an analytical bias correction method for the special case of
the Gaussian copula. For general conditional copula models, a likelihood-based correction
method is proposed, in which the likelihood function is computed via Monte-Carlo integration.
Numerical studies confirm that the proposed bias-correction methods achieve accurate estimation
of the dependence parameter.