Hello
I'm analyzing data from a survey, using weighted least square regression in SPSS (because of Heteroscedasticity), and i'm concerned with the VIF value that is much more than 10 (in the OLS with no weight the VIF value for the same variables is close to 1). Is there a problem with the model or is it Ok to have high VIF in a WLS. Thank you in advance.
I am not sure what you did. Could you please give a small example? Best, D Booth
Hi David!
I'm studying the impact of neighborhood life quality satisfaction and social capital on depression, when I run an OLS in SPSS, the VIF value for all my independent and control variables is less than 2, but when I run the WLS using the same model, I get VIF values of more than 10, this is probably due to the WLS weight, my question is: can I still trust the results of the regression?
I followed the steps indicated in this video to create the weight
weighted regression spss
Thank you.
I had a similar situation some years ago that was successful. I have attached a copy of of a paper that resulted. Perhaps that will help good luck. D. Booth
BTW if I were doing that research today I would look for a transformation using an automatic Box-Cox procedure. A very good one can be found by Google using automatic Box-Cox procedure R. Best regards, David Booth
Afrad -
This is how you can do WLS regression, considering the heteroscedasticity in the error structure:
https://www.researchgate.net/publication/333642828_Estimating_the_Coefficient_of_Heteroscedasticity
This is part of https://www.researchgate.net/project/OLS-Regression-Should-Not-Be-a-Default-for-WLS-Regression
which is for various regressions of form y = y* + e.
With regard to how well your model is doing, you can use a "graphical residual analysis" to check the fit, and some form of "cross-validation" to consider "overfit" to your particular data set.
Cheers - Jim
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics
- Regression