A Test of Missing Completely at Random for Regression Data with Nonresponse

• Activity Number: 396
• Type: Topic Contributed
• Date/Time: Tuesday, August 3, 2010 : 2:00 PM to 3:50 PM
• Sponsor: Section on Survey Research Methods
• Author(s): Gong Tang*+
• Companies: University of Pittsburgh
• Keywords: missing data ; nonresponse ; MCAR ; test

Abstract

We consider testing of MCAR in regression data with nonresponse. Likelihood ratio tests have been discussed in the context of multivariate data with missing values but these tests require the specification of the joint distribution of all variables (Little, 1988). Subsequently Chen & Little (1999), and Qu & Song (2002) proposed a Wald-type test and a score test for generalized estimating equations with using the same fact that all sub-patterns share the same model parameters under MCAR. For regression analysis of data with nonresponse, we propose a Wald-type test for missing completely at random by comparing two sets of consistent estimators of regression parameters under MCAR instead of dealing with pattern-specific parameter estimates. This method can be applied in testing MCAR for longitudinal data with dropouts.