Publication type
Conference Paper
Series
Riunione Scientifica della SIS Conference
Author
Publication date
June 18, 1998
Abstract:
A selection mechanism is ignorable for a regression model of interest if we can make inference on this model ignoring the process that causes missing data. The conditions that allow neglecting the selection process had been given in Rubin (1976) for the cross-section case. The extension to the panel data case is immediate when the data are independently and identically distributed across units and over time. Unfortunately, while the data on different units can be generally assumed independent, the repeated observations on the same unit are probably dependent. This is the reason for the large use of dynamic regression models. In this paper we give the conditions that a dynamic regression model of interest and the selection mechanism must satisfied to guarantee ignorability, using some suggestions coming from the theory of statistical models reduction.
#517565