attrition: comparing *surviving* sample to original

#1
Hi,
I have a question about attrition analyses in longitudinal databases.

I know that I can compare the subjects who dropped out ("nonresponders") to those who responded, but what I'd like to do is to determine whether the surviving sample is representative of the original sample. If I do t-tests on demographic data comparing the surviving sample to the original sample, I violate the assumption of non-independence (they are the same subjects). Is there a way to do this correctly?

Thank you!

Lisa
 
#2
My strong feeling is that it is not possible to compare the surviving group with the original (since they are the same), and that it is much more informative to compare nonresponders to responders. This would be how I would address the problem of attrition. However, someone has asked me whether it is possible to do it the other way (surviving vs. original). Any help on this would be greatly appreciated.
 
#4
foerverpostdoc, I don't know if I'm the best person to answer this question, but I'll tell you what I think.

Now, the original population can be considered a mixture of individuals, those nonresponders, and those that remain in the study. So the mean for a given attribute of the original population is a weighted mean of the nonresponder mean and the mean for the group who remained in the study.

This indicates that the difference in the mean for the group who remained in the study and the original mean can only be due to the effect of the nonresponders.

Hence, the remaining subsample will be representative of the mixture population only if the remaining subsample and the nonresponder subsample are alike.

So, I think you could make your argument by testing the remaining subsample against the nonresponders as is consistent with your initial thinking. If you find that these two groups are different, you can be fairly confident that a each subgroup will be different from the mixture of the two.

let me know if this helps,

~Matt