Hello there,

I am doing a study on vigilance research by comparing different EEG methods (ERP, FFT and ERD) regarding their predictive power of reaction time, subjective state and task performance as outcome variables. I do this By using GEE with gamma log link.

Based on the Information Criteria, I have selected the best model for every EEG method and every outcome variable, that means; three one-predictor models.

The QICC values between the three models are small, so I want to do a model validation by comparing them on the residuals. What would be the best way to do this?

Based on literature:

1) I just make a scatterplot of the observed outcome values by expected values (XBpredicted)
2) make QQ plots, but than you have to specify the distribution of each outcome variable.
3) Use the Pearson residuals (but how do I compare them?)