Suppose I have some survey data on which I'd like to conduct an analysis similar to this:
Yang, Age-period-cohort analysis, Ch 7.

Suppose further there are missing data I wish to handle via multiple imputation. Good practice requires that I include in the imputation model all the terms that I intend to include in the model of interest.

However, $\text{age} = \text{period}-\text{cohort}$.

I'm afraid this will cause problems for imputation software.

Has anyone else dealt with this issue? Any suggestions? I have access to software that can treat period and cohort as random effects, but I'm not sure that's enough to overcome collinearity with age.

Any guidance appreciated.

  • $\begingroup$ One thing you could do is just use 2 of the three in your imputation, but the three of them in your age-period-cohort analysis. Also, if the missing data is on either age, period, or cohort, and you have the other two, then obviously you can just calculate the missing one directly. $\endgroup$ – Patrick Coulombe May 16 '14 at 2:42
  • $\begingroup$ Precisely what I've done for now. None of the age, period or cohort data are missing. Since much of my analysis is age by period tables, where cohort is implied, this makes much sense. I'm just not sure if I can get away with it when age, period nd cohort are all three explicitly included. $\endgroup$ – William Shakespeare May 17 '14 at 14:55

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.