 |
 |
> From [log in to unmask] Mon Sep 24 17:06:44 2001 > Date: Mon, 24 Sep 2001 10:52:31 -0500 Dear Darren, > I'm wondering how to respond to a review on an article and would > appreciate any help. In the article I used a random effects analysis to > examine activations in 17 subjects. One reviewer said "... it's hard to > evaluate the extent to which the significance in the group data may be > driven by activation in only a few of the subjects." They then > suggested providing numbers of subjects showing each of the activations > in my main table. > > I don't see any problem in providing the subject numbers but I don't > think this will settle the question. I thought random effects analyses > specifically guarded against the possibility of one subject skewing the > results. Also while in most areas of activation a majority of subjects > demonstrated the activation as individuals, in one area that showed > activation in the RanFX analysis (Z > 4.2) only one subject showed an > individual activation at the p<.001 threshold. Fifteen other subjects > showed consistent positive parameter estimates at this point that were > just below the individual threshold. Hence in the RanFx analysis this > region had low inter-subject variation and activity. You are absolutely right. The second-level analysis compares the mean activation to the intersubject variation in that activation. Activations "in only a few of the subjects" will inflate the error variance and reduce the t value. The reviewer's concern pertains only to fixed-effects analyses. I hope this helps - Karl