An Index of Local Sensitivity to Nonignorability Under Pseudo-Likelihood

• Activity Number: 333
• Type: Contributed
• Date/Time: Tuesday, July 31, 2007 : 2:00 PM to 3:50 PM
• Sponsor: Biometrics Section
• Author(s): Fang Zhu*+ and Gong Tang
• Companies: University of Pittsburgh and University of Pittsburgh
• Keywords: missing data ; sensitivity analysis ; ISNI

Abstract

Sensitivity analysis is recommended for regression analysis of data with nonresponse. Index of local sensitivity to nonignorability (ISNI) (Troxel, et al., 2004), was devised to detect the sensitivity of a parameter estimate to the ignorability assumption. However, ISNI requires specifying a parametric model for the missing-data mechanism. Here a new sensitivity index is proposed under a pseudo-likelihood method (Tang et al., 2003) that does not require specification of the mechanism. As an example, for bivariate data (x,y) when the mechanism is assumed to be an arbitrary function of x+lambda*y, this new index is defined as the 1st derivative of a parameter estimate as a function of lambda at lambda=0. For bivariate normal data a closed form of this new index was derived. Simulation studies have shown that this new index has comparable performance with the ISNI.