Below are the t-test results for the variables c, e, and u themselves, for the a1 fold ans S63, S60 groups.
What's interesting is that u differs significantly, but not e, whereas we have already seen that it is ln(c/e) that differed significantly, not ln(c/u.
In particular, it makes sense that u should differ signficantly, and that mean u should be higher for S63 than for S60 rather than vice- versa, since u indicates over-representation, and S63 was originally selected because of its overrepresentation.
But what's puzzling is why this difference in u between S63 and S60 doesn't create a difference in ln(c/u) between S63 and S60.
And vice versa, although there's no difference in e between S63 and S60 (or at least much less than u), there IS a difference in ln(c/e) between S630 and S60.
Variable: c a1 Fold , x1 driver, (63) Study vs (60) Study
N Mean Var 2-Tailed ln(GM) S63 S60 S63 S60 S63 S60 P