Given that linear discriminant analysis (LDA) for two groups and multiple regression essentially the same results, could they be used as confirmatory techniques? For example, could a DA be used to classify students in high vs. low performing groups, followed by linear regression to confirm DA predictions or vice versa? Would they always yield the same results or could the predictors differ?
Asked
Active
Viewed 622 times
0
-
http://stats.stackexchange.com/q/31459/3277. As said in my answer there, `In case the DV consisting just of 2 groups the two analyses are actually identical.` And they give same p-value. And because there is just 2 groups, the process of extracting the discriminants and the classification are basically one process (`Some people tend to fail to recognize...` [here](http://stats.stackexchange.com/a/169483/3277)). – ttnphns Feb 15 '17 at 04:14
1 Answers
-1
No, it does not make sense to use one to confirm the other. That they give you the same answer is because they are doing basically the same thing. If one is bad, then the other will also be bad.
Sometimes they will have different predictors, but this is only due to different conventions (e.g., most people select variables for regression using significance tests, whereas with LDA cross-validation is more popular).
You can improve predictions by using multiple models and taking their average prediction (provided all the models are sensible), however, combing the results of LDA and linear regression would be pretty pointless as they are too similar (unless you used different predictor variables).
Tim
- 3,255
- 14
- 24