DoE - difference between stepwise and least square regression
Apr 20, 2020 11:06 AM(593 views)
I am having difficulties understanding the difference between stepwise and least square regression. I normally chose standard least square as the personality when I fitted a model, then I removed the factors with high p-value. In this case, isn't the way I did the same as the backward stepwise method? I also tried to use both methods to analyze my data, and the results (number of factors left in the model) were the same. So I don't really know how to choose between these two methods.
Yes, your method of removing one term at a time is the same as the backward direction used by the Stepwise platform. The platform is a productivity tool. It can quickly perform the deletions when the model has many terms. You can set parameters about the deletions and the stopping to suit your preferences. The final model can then be transferred to Fit Least Squares for the actual regression analysis. Stepwise is dedicated to model selection, not analysis.
Mark, of course, is completely correct. I prefer starting with the saturated (or full) model and removing terms myself (using R-square/R-square adjusted deltas, RMSE, etc.) as I can combine not just the statistics, but practical implications as terms are removed.