outliers in GEE

[tan1721]

I am fitting GEE model for a repeated measure dependent variable. Attached figure shows boxplots. My question is, is it necessary to delete the outliers (showing as star signs) to fit a GEE model with gamma distribution?

[ballardw]

Here's a suggestion: Do the model both ways. See what the effect on the resulting model parameters might be.

It helps to describe your graph. For example are those numbers shown the observation number or the number of observations with that outlier value? Big difference on meaning. And how many total observations are involved. If you have 1,000,000 observations with, possibly, 4 total outliers (one interpretation of that graph), the outliers are likely to have negligible effect. If there are only 60 observations there might be more effect.

[tan1721]

These are observation numbers. I have 34 observations from a driving simulator experiment. While did the analyses without these 2 outliers (2 and 26) and with these outliers,  number of significant factors are higher when these 2 outliers are deleted. Considering this difference as substantial difference, should I include the results from both analyses?