Hi everybody,
In Tabacknick and Fidell's "Using Multivariate Statistics", chapter 5, there are two types of regression "Sequential Multiple Regression" & and "Statistical (Stepwise) Regression" that I can not differ very well. Actually I have only one predictor variable (with two sub-scales) and a moderator variable. Shall I use Stepwise? He also puts it this way for Stepwise Regression:
"Cross-validation is highly recommended for statistical regression and is accomplished through several steps. First, the data set is divided into two random samples: a recommended split is 80% for the statistical regression analysis and the remaining 20% as the cross validation sample (SYSTAT Software Inc.. 2004, p. !I-16). After the statistical regression on the larger subsample is run. predicted scores are created for the smaller cross-validation sample using the regression coefficients produced by the analysis. Finally, predicted scores and actual scores are correlated to find R' for the smaller sample. A large discrepancy between R. for the smaller and larger samples indicates overfitting and lack of generalizability of the results of the analysis." (P. 139 & 140)
with my situation described above, should I need a cross-validation?
Thanks in advance for your time and energy, :)
Sara
Given your small # of IVs, I'd recommend stepwise (subsatsntive) regression.
Hope this helps
Matt
Sara -
A little testing (cross-validation) can help avoid overfitting. I also suggest graphical residual analysis to consider the fit. You can also compare two or more models by doing the graphical residual analysis on the same graph (perhaps color coded or otherwise delineated) for both/all models. Put predicted values on the x-axis and the estimated residuals on the y-axis. You could pick the models to compare based on your subject matter knowledge. This could help avoid spurious results.
This is general advice. I'm not sure of your application. I don't know what you mean by "sub-scales," but as I just noted, the above is general advice.
Best wishes - Jim
Having two models compared by graphical residual analysis was quite helpful, I will do so.
Thanks a lot James R Knaub
NO NO NO, NEVER use stepwise regression. It is highly unstable.See the attached paper and particularly some of the ref's for details, Please. Best wishes, D. Booth
James R Knaub You are of course correct, in what you say, but stepwise is a really bad start. See the attached paper. BTW I read the paper you recommended the other day by Brick (modern sampling). I found it quite interesting. Thanks. David
James R Knaub I understand, but the first comment did. You may find the papers of some interest anyway. Best, David
- Badges
- Science topic
- Similar topics
- Mathematical Sciences
- Statistics
- Statistical Software
- SPSS