Hi,
Is it possible that two strongly correlated independent variables ( r between them= 0.91) that correlate very similarly with the depended variable (r =0.29 and r= 0.28) appear differently on stepwise regression model that predict the depended variable?
Meaning, one of the variables was added to the model at the first step and remains in the second too (the model was reached in two steps) and explained the highest degree of variance in the depended variable scores (with small effect size though of ~9%) and the second was excluded from the beginning (overall there were 6 possible predictors that were added to the model, only two were significant).
Can it be valid result?
Hi,
First of all, what type of regression have you done ?
In any case, if two variables are strongly correlated (as yours), you can not put theses two variables in the same model. You have to make a choice. You choose the variable which is the most important depending on the topic, or you choose the one that is the most significant in the univarite model.
Yes it can be.
Another concern is that stepwise selection methods are not recommended. There are many threads on RG about this.
So basically I've to run two separated regressions (say Forward regression) with each contain all of the variables except for one of the inter-correlated independent variables in complementary way. Then to report on the one which explained more of the variance in the dependent variable scores? Or that I should report on these two regression and then in the discussion to deal only with the better one?
It is what I would do yes, running two separeted regressions. Your variables are so correlated that the results should be the same...
You should form a model not the method you use (ie, stepwise regression). You could try methods like ridge regression that accepts correlated variables if both variables need to be in the equation.
Two separate models do not come to me meaningful.
- Badges
- Science topic