On Wed, 26 Feb 2003 16:01:25 -0600, "DCD" <[EMAIL PROTECTED]> wrote:
> I have collected some data using older adults have found a significant > difference (p < .05) between the three groups in terms of their activity > level. Subsequent to this analysis I am contrasting the three groups on a > balance measure and deciding whether or not to use activity level as a > covariate in the ANOVA calculations. There is a DIFFERENCE measured between groups in Activity; and you want subsequently wonder whether to use activity level as a covariate. Well, if it has a logical role as a covariate, then you might *have* to try it as a covariate, or else concede whatever arguments someone makes about it. Those could be strong arguments if it is strongly correlated. However, having mean-differences on a covariate is always problematic -- you really do, basically, *have* to plot the means and scores to see who is being affected. Are there ceiling/ basement effects? Do regression lines cross? Is regression-to-the-mean sufficient to explain everything that is otherwise interesting? On the other hand, if you stick in the covariate and it makes no difference that is discernible on any test, then you can say *that* about it. > > The question I have is whether or not to use activity level as a covariate > or not, based on the fact that the main effect for activity level violated > the assumptions of the homogeneity of regression slopes when compared with > one of the DV used in the study. - There are mean difference, *plus*, somewhere else, you see violations of homogeneity? - well, the heterogeneity (if there is any on this variable) is potentially less important than the difference in means. I should say, when I see the logic misused, I do wonder whether the underlying facts have been grasped. You can't write an adequate argument, in a non-randomized study (which this apparently is) if you leave out *any* factor that other people will find interesting or important. So, I think you *have* to look at this potential covariate to see what it is doing. It might be that the most important thing to announce might be, that it is badly distributed, or badly *confounding* the other variables; and you need help to figure it out further. > I do have a small sample size and > consequently did not use the covariate analysis but wanted to see if I was > justified based on the violation listed above. > > Any help would be appreciated. You can try to *discount* some analysis based on troubles with the scores. But for an observational study, that is the practically equivalent to admitting defeat, if someone has other ideas about that variable. -- Rich Ulrich, [EMAIL PROTECTED] http://www.pitt.edu/~wpilib/index.html . . ================================================================= Instructions for joining and leaving this list, remarks about the problem of INAPPROPRIATE MESSAGES, and archives are available at: . http://jse.stat.ncsu.edu/ . =================================================================
