We consider estimation of a linear regression model using data where some covariate values are missing but imputations are available to fill in the missing values. This situation generates a tradeoff between bias and precision when estimating the regression parameters of interest. Using only the subsample of complete observations does not cause bias but may imply a substantial loss of precision because the complete cases may be too few. On the other hand, filling in the missing values with imputations may cause bias. We provide the new Stata command gmi, which handles such tradeoff by using either model reduction or Bayesian model averaging techniques in the context of the generalized missing indicator approach recently proposed by Dardanoni, Modica, and Peracchi (2011, Journal of Econometrics 162: 362–368). If multiple imputations are available, gmi can also be combined with the built-in Stata prefix mi estimate to account for extra variability due to imputation. We illustrate the use of gmi with an empirical application in the health domain, where item nonresponse is substantial.
|Numero di pagine||30|
|Rivista||THE STATA JOURNAL|
|Stato di pubblicazione||Published - 2012|
All Science Journal Classification (ASJC) codes
- Mathematics (miscellaneous)