Imputation - non- and partial-responders, or partial only?

Moderators: statman, Analyst Techy, andris, Fierce, GerineL, Smash

NatalieAsks
Posts: 2
Joined: Tue Apr 19, 2016 11:24 am

Imputation - non- and partial-responders, or partial only?

Postby NatalieAsks » Tue Apr 19, 2016 11:39 am

Quick question re. multiple imputation. I have a database with 1800 participants for one year of a longitudinal study. 1400 of these responded to a self-admin questionnaire. So I have 400 non-responders about whom I have basic demographic info (because they responded to previous years of the longitudinal study) and also some partial respoonders who answered the questionnaire but missed some items.

I've been advised to do multiple imputation. Should I do it for a) the whole sample 1800, or b) only those who answered the questionnaire?

My main outcome variable has about 30% missing values if I take the sample as a whole and 8-10% missing values is I take only responders to the questionnaire.

Any opinions?

Thank you!
GerineL
Moderator
Posts: 1477
Joined: Tue Jun 10, 2008 4:50 pm

Re: Imputation - non- and partial-responders, or partial only?

Postby GerineL » Tue Apr 19, 2016 3:53 pm

do you have anything to impute on / base the imputation on for those 400 cases? if not, I would say only for the others, but opinions may differ.
NatalieAsks
Posts: 2
Joined: Tue Apr 19, 2016 11:24 am

Re: Imputation - non- and partial-responders, or partial only?

Postby NatalieAsks » Fri Apr 22, 2016 8:29 am

Thanks for the advice - I have some basic socio-demographic info but nothing else. I went ahead and only imputed for partial responders (but not non-responders). This meant imputing for missings of around 10% which I believe is less subject to bias than imputing for larger proportions. It seems to have worked pretty well!

Thanks again!

Who is online

Users browsing this forum: No registered users and 2 guests

cron