Estimating Parameters for Binary Data with Time-Dependent Covariates Using the Generalized Method of Moments

• Activity Number: 483
• Type: Contributed
• Date/Time: Wednesday, August 7, 2013 : 8:30 AM to 10:20 AM
• Sponsor: Biometrics Section
• Author(s): Maryann Shane*+
• Companies: University of Northern Colorado
• Keywords: generalized method of moments ; binary covariates ; binary response ; time dependent covariates ; convergence failure ; quasi-complete separation

Abstract

Generalized Method of Moments (GMM) has been used with longitudinal data models with time-dependent covariates. Lai and Small (2007) applied the two-step GMM (2SGMM) to continuous longitudinal data with time-dependent covariates to improve efficiency over the independent Generalized Estimating Equations (GEE) approach (Zeger & Liang, 1986; Liang & Zeger, 1986). The continuously updating GMM (CUGMM) procedure, outlined by Hansen (1982), was discussed in the JSM 2012 topic contributed session, Correlated Logistic Regression with Time-Dependent Covariates (Lalonde et al, 2012), and has been applied to binary longitudinal data with time-dependent covariates. When a binary response and binary time-dependent covariate are present, convergence failure has been observed for 2SGMM.

This paper presents the use of 2SGMM for binary longitudinal data in the presence of at least one binary time-dependent covariate. The GMM algorithm fails to converge. Possible reasons and explanations, including two-way tables of the data, are offered.