> "Smith, David W." wrote: > > Is it possible that the subjects are getting better over the then > trials? That is, that they are learning to make the judgements > better? If there responses are random, due to no effect of the > treatment, then this won't happen, but if there is a treatment > effect, then it may be more pronounced in the later decisions than > in the earlier ones. Moreover, the effect over trials are likely > to vary among subjects. > > While other tests may be suitable, you might compare the first > trials and last trials against the null hypothesis. These are > relatively weak tests, however, even if exact tests are used.
The most powerful test of this might be a trend test for dichotomous data. (Although I can't immediately remember whether this citation is between or within subjects ...) Marascuilo, L. A., & McSweeney, M. (1967). Nonparametric post hoc comparisons for trend. Psychological Bulletin, 67, 401-412. There is a slightly easier version (to compute/follow) in: Meddis, R. (1984). Statistics using ranks: a unified approach. Oxford: Blackwell. Thom . . ================================================================= Instructions for joining and leaving this list, remarks about the problem of INAPPROPRIATE MESSAGES, and archives are available at: . http://jse.stat.ncsu.edu/ . =================================================================
