In what context are you asking these questions? As R.MacG. Dawson pointed out on another thread, you can't get useful advice if you don't describe the problem adequately. ALL of the suggestions you cite (and, I rather imagine, all the suggestions you know about but don't cite) arose out of a particular context and in particular circumstances; do you have the same context, in the same circumstances? One may doubt it.
ALL of the suggested criteria are nonsense under SOME circumstances, and in SOME situations. (Just for one instance, a sample size of 100 would not be considered "large" in quite a number of contexts. Would it be considered so in YOUR context? Damfino. Again for instance: How many variables are you contemplating? Are you modelling interactions among these predictors? (And if not, why not?) Your point (3) can be circumvented, in any circumstances, by orthogonalizing the candidate variable with respect to all the predictors already in the model. Then the correlation of this variable with each of its precursors is zero, and the only point of interest is the correlation of the orthogonalized variable with the residual from the response variable at this point in the model. By and large, stepwise regression is something one does when one cannot think of something intelligent to do. (It may actually be helpful, in stimulating thought and thus helping one to arrive at something intelligent, but it is itself no substitute for ratiocination.) On Mon, 6 Oct 2003, Bastian wrote: > there are lots of suggestions for the minimum data for regression > models, i.e. > > 1) 1 var. for every 10 observations. > 2) Variables can be added until Adjusted R-square deviates > substantially Unadjusted R-square. > 3) With relatively large samples (n=100), a variable can be added > if is correlation with other variables is no larger than about > 0.80 or 0.85. > > My question is, whether the 1) suggestion should be considered with a > stepwise regression, too. That means when I do a stepwise regression > and get i.e. 20 variables are 200 obeserved values sufficient for > prediction or is it important to have as much observations as there > are "possible" variables before the use of the stepwise method. ----------------------------------------------------------------------- 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/ . =================================================================
