On Wed, 28 May 2003, Scheltema, Karen wrote ...: > I know about the perils of stepwise, and I agree with you that it is > a less than desirable procedure. This researcher, however, is not > as convinced as I am about not doing stepwise. Sigh. He has more > variables than would comfortably fit a 5-1 case to variable ratio > for a forced entry regression, which is why he was hoping stepwise > would help him narrow his model. Any suggestions I can give him, > short of telling him to scrap everything?
... in reply to Paul Swank, who wrote, inter alia: > ... If the interaction is significant then the main effects that > make up the interaction must stay in the model. This is the conventional advice, but one needs to consider it with some skepticism and attention to context. E.g., if two "main effects" are dichotomies coded (0,1), their product is also (0,1) and "1" represents those cases that are "1" BOTH variables. For an example where it is entirely reasonable to omit a main effect when the interaction is significant, see my paper on interactions in multiple regression at the Minitab web site (under "White Papers"): the effect on pulse rate of having run in place (rather than resting) was significant; the effect of sex, in the presence of (sex-by-run) interaction, was not; the effect of the (sex-by-run) interaction was significant; showing that for those who did not run, the regression of final pulse rate on initial pulse rate was unaffected by sex (which was to be expected), but for those who ran, females showed a greater increase in pulse rate (which might have been expected by a competent physiologist, but was interesting and informative for more naive observers). > By the way, there is backward selection, forward selection, and > stepwise selection (plus several other more esoteric procedures) but > I, for one, have never heard of backward stepwise. I would guess (am I right, Karen?) that the investigator started by entering all available predictors at the beginning, and then invoked the usual stepwise procedure as a means of discarding apparently redundant predictors. -- Don. ----------------------------------------------------------------------- Donald F. Burrill [EMAIL PROTECTED] 56 Sebbins Pond Drive, Bedford, NH 03110 (603) 626-0816 . . ================================================================= Instructions for joining and leaving this list, remarks about the problem of INAPPROPRIATE MESSAGES, and archives are available at: . http://jse.stat.ncsu.edu/ . =================================================================
