Steve wrote: When I initially skimmed your Q. I thought you were referring to a common situation in repeated measures ANOVA where factors are non-orthogonal which means that standard packages won't run the SPSS model.
e.g., participants read several texts and you try to predict reading times based on participant or text characteristics (e.g., number of words, type of words and so forth). Lorch and Myers describe how to derive the correct error term for this kind of analysis (ISTR it involves running separate analyses by subject in packages such as SPSS) and computing the correct tests from these. A separate issue that you touched upon is categorizing a continuous variable for use in ANOVA - you correctly surmise that this is not a good thing to do (you can search the newsgroup for an animated discussion). e.g., or see MacCallum, R. C., Zhang, S., Preacher, K. J., & Rucker, D. D. (2002). On the practice of dichotomization of quantitative variables. Psychological Methods, 7, 19-40. A final point (error in the measurement of the predictor) is addressed by another poster. This strikes me as a _much_ broader issue (not confined to ANOVA). 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/ . =================================================================
