2
$\begingroup$

I have a repeated measures data set with 5 time points and one binary dependent variable (yes/no response to a question). There are around 20 subjects.

If my understanding is correct, both "generalized estimating equations" and "generalized linear mixed models" are possible approaches to test if there is an effect of time point on this dependent variable. This is, in part, based on this CV answer: https://stats.stackexchange.com/a/32421/6389

Some of my subjects have missing data at some time points.

My understanding of linear mixed models is that they cope with missing data points. Does this extend to generalized linear mixed models?

What is the effect of missing data on generalized estimating equations?

Improve this question
$\endgroup$
4
  • 1
    $\begingroup$ In this setting, GEE isn't used to "test if there is an effect of time point on the dependent variable" unless you're including time point as a covariate. GEE is used to give inference about mean parameters that is correct even when the nature of the temporal associations is misspecified. Re your main queries: Yes, generalized linear mixed models can still use the available data for units where some time points are missing. In all software I'm familiar with, GEEs require the entire "cluster" to be present to pass to the "working covariance", so units with missing timepoints would be discarded. $\endgroup$
    – Macro
    Commented Aug 16, 2013 at 15:21
  • 1
    $\begingroup$ +1. Much depends what you mean by "cope with". I take it to mean: omit missings and use the actual numbers of observations in formulas. But depending on your software, missings might cause a function to fail. $\endgroup$
    – Nick Cox
    Commented Aug 16, 2013 at 15:25
  • $\begingroup$ Nick, yes by "cope with" I mean all data from a single subject is NOT discarded when only 1 trial for that subject is missing, or as you say, use the actual number of observations. I tend to use SPSS due to familiarity, but could use R when necessary. $\endgroup$
    – omian
    Commented Aug 28, 2013 at 13:56
  • $\begingroup$ Macro, thanks. I will look into using GLMM instead of GEE $\endgroup$
    – omian
    Commented Aug 28, 2013 at 13:57

0

Reset to default

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.