help
489 Subscribers 
help > How to correct CONN inflation of subject number/T-statistic
Showing 1-1 of 1 posts
Apr 9, 2021 06:04 PM | Mikey Malina - University of Chicago
How to correct CONN inflation of subject number/T-statistic
Hello,
This post is in reference to a previous post (Possibly incorrect model, https://www.nitrc.org/forum/message.php?...) on the same project. We have been trying to sort out warnings of non-estimable contrasts, and like you said in your previous post, we believe it has something to do with redundant covariates (SZP and onlySZP, as an example) being included in our contrasts, covering the same exact space for one-sample t-tests.
The solution would be to only include one of these covariates, but this does not work for us as it causes t-statistic inflation. By including just SZP and removing onlySZP, the warning goes away, but our subject number shoots up to 785 when the actual number is 269. To be clear, the expected analysis sample is not supposed 269, it should be 204 after intersection with GoodQA2. Either way, we are getting subject number inflation. Are there are any ways to avoid this problem?
I have attached a document showing a marked difference in results (SZP vs onlySZP vs both) in an example of LSecAud connectivity for our patient sample.
Thank you,
Mikey
This post is in reference to a previous post (Possibly incorrect model, https://www.nitrc.org/forum/message.php?...) on the same project. We have been trying to sort out warnings of non-estimable contrasts, and like you said in your previous post, we believe it has something to do with redundant covariates (SZP and onlySZP, as an example) being included in our contrasts, covering the same exact space for one-sample t-tests.
The solution would be to only include one of these covariates, but this does not work for us as it causes t-statistic inflation. By including just SZP and removing onlySZP, the warning goes away, but our subject number shoots up to 785 when the actual number is 269. To be clear, the expected analysis sample is not supposed 269, it should be 204 after intersection with GoodQA2. Either way, we are getting subject number inflation. Are there are any ways to avoid this problem?
I have attached a document showing a marked difference in results (SZP vs onlySZP vs both) in an example of LSecAud connectivity for our patient sample.
Thank you,
Mikey