We are running a 4 x 3 x 3 model. All within subjects factors.
3 Factors:
Visit: 1, 2, 3 4
Temp: 1 2 3
Trial: 1 2 3
We do have missing data for 1 subject. That subject has missing data from 1 visit. They only have data from 3 out of 4 visits.
In addition to the random effects question, there is something else:
We’ve run this model without specifying the correlation structure, so it is using compound symmetry, no? Something weird is going on with the output, that doesn’t look right compared to when we run 3dMVM with a complete data set. We aren’t expecting identical results, but not such a wide disparity… I don’t think it is something with 3dLME because I get the same results using SPSS on extracted data from a voxel. Just wondering if having data missing for an entire visit has substantial effect on the results.
How much difference is between the two results? Dramatic difference? For comparison, try 3dLME without the subject with missing data, and see if the two results have a better match.
Since you have played the data at the voxel level, you may try to plot out the data at those voxels/regions where you see substantial differences, and see if that subject has significant different values from other subjects.
The
National Institute of Mental Health (NIMH) is part of the National Institutes of
Health (NIH), a component of the U.S. Department of Health and Human
Services.