Hi On Mon, 26 Jan 2004, Thom wrote:
> jim clark wrote: > > You don't say what kind of contrasts you specified (polynomial? > > some other?), or what pattern of differences among the means you > > were expecting. In general, contrasts are the best way to make > > sense of differences among multiple means, especially when there > > are clear theoretical predictions. As for reporting the results > > While I agree, I would remove the "especially". Contrasts, especially > the polynomial contrasts that SPSS prints out are not going to be useful > in most cases unless the contrasts map onto clear (a priori) theoretical > predictions. For example, it is pretty rare that a quartic or cubic > contrast is readily interpretable in most contexts that I'm familiar with. I think Thom and I pretty much agree, although our wording and emphasis might differ. But sometimes contrasts can be useful even when the "theory" is weak. If one has ordered levels of a factor (and little else to go on), for example, it is true that the cubic or quartic or other higher-order components might rarely be useful. Nonetheless, a substantial loading of SS treatment on the single df linear component can enhance the statistical significance of the effect, even turning a non-significant omnibus effect into a significant linear effect. I would still call this useful. > In this case I think the op wants to report all significant effects that > SPSS prints out and without knowing the context ... Exactly right! Best wishes Jim ============================================================================ James M. Clark (204) 786-9757 Department of Psychology (204) 774-4134 Fax University of Winnipeg 4L05D Winnipeg, Manitoba R3B 2E9 [EMAIL PROTECTED] CANADA http://www.uwinnipeg.ca/~clark ============================================================================ . . ================================================================= Instructions for joining and leaving this list, remarks about the problem of INAPPROPRIATE MESSAGES, and archives are available at: . http://jse.stat.ncsu.edu/ . =================================================================
