1
$\begingroup$

I am participating in a Project with data from a complex survey. We are going to analyze data from a national fertility survey. Some of the questions in the questionnaire were only asked by a subsample. For example, a question about the job was only asked to employed people. This variable and others like that are important in our study and we would like to include it in our logistic model. What approach do you recommend? I have seen in some papers of this area to set up another category “else”, that is, they combine other people (to which the question was not raised) creating an additional category. If this approach is acceptable, should i attributed random values to the income variable in order to reduce bias?

I will be very grateful if you could help me.

$\endgroup$
1
$\begingroup$

You need to figure out the reason why some people didn't answer or why they were not asked. That may matter a lot, e.g. not clear why people without a job would input data on jobs.

This may not completely answer your question, but, when appropriate, you can use multiple imputations. See the function aregImpute in package rms

$\endgroup$
  • $\begingroup$ Thank you. About you question why some people didn't answer or why they were not asked. For example consider the response variable 1 - have children, 0 - haven't children. The predictor "My job is secure" is only answered to employed people. $\endgroup$ – user40529 Mar 6 '14 at 22:24
  • $\begingroup$ Yours is a special case in which creating a special category of unemployed is 0.9 likely to work, if everyone who is employed answers the job question. In general, creating new categories causes problems, and we use predictive mean matching (the default in aregImpute) to make sure imputations are one of the real categories. $\endgroup$ – Frank Harrell Apr 6 '14 at 12:53

Your Answer

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

Not the answer you're looking for? Browse other questions tagged or ask your own question.