3
$\begingroup$

Recently I got a global longitudinal data from several countries, and each county has one outcome variable and two predictors from 1995 to 2008. I found one of the predictors is always missing in each country in 1995, 1997, 1999, and 2001 because that variable was collected every two years before 2001. This situation seriously affects another complete predictor when using a complete case analysis because too many useful information is lost in that four years. Also, it obviously affects the fitting of a time smoothing function.

I am pretty wondering whether this situation is appropriate to use the multiple imputation method to generate data. As I know, this case is not a missing at random mechanism, so the multiple imputation may not be a perfect way to deal with my case. I am looking for any advice here to find out a solution to deal with those missing data. Any suggestion is appreciated.

$\endgroup$
4
  • 3
    $\begingroup$ If the data is completely absent at four time points, I don't think imputation is going to help. Multiple imputation helps when some observations are missing. But if there's nothing special about those four years, other than that data was not recorded in them, leaving them out of the analysis would be okay. $\endgroup$ – guest Mar 10 '12 at 21:12
  • $\begingroup$ Is it completely absent? I read it as saying that the dependent variable is still there. In which case there may be enough to impute with. $\endgroup$ – conjugateprior Mar 11 '12 at 21:10
  • $\begingroup$ @cchien: what makes you think it is not MAR? $\endgroup$ – conjugateprior Mar 11 '12 at 21:11
  • 2
    $\begingroup$ Just for the record: Complete case analysis (which you don't want), interpolation (linear or spline; best if the data is monotonic), or creating a more sophisticated model for the missing years are the options that cross my mind. You might also want to consider leaving out the years 2003, 2005 and 2007 from your regression process and use them for validation. On that note, I don't see a point in modeling a predictor after all. $\endgroup$ – krlmlr Apr 12 '12 at 20:18

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.