Items Through Imputation: An Application to the NSOPF
Fahimi, M. (2005, August). Items Through Imputation: An Application to the NSOPF. Presented at Joint Statistical Meetings, Minneapolis, MN.
For complex surveys, the task of imputing a large number of variables as a major undertaking, since the resulting data must satisfy various consistency checks that are often intertwined. For instance, values of certain variables might have to add up to those of others, while in other cases certain variables might restrict the values other variables can take on. When an imputed value is deemed inconsistent with other variables, it is sometimes logically imputed (edited) to resolve the observed inconsistency. Alternatively, such values are set to missing and then imputed to avoid the preceding laborious process. This paper discusses a method for measuring and reducing the number of inconsistent cases through imputation, which inconsistency is measured as a function of the number of cases that are set to missing during the editing process. Missing data are imputed using a weighted sequential hot-deck methodology, which results in values that are consistent with respect to all known skip patterns and logical constraints. The research is based on the 2004 National Study of Postsecondary Faculty (NSOPF:04) survey data.