Hello everyone,
I am running 3 different regression models on determinants of water access (i.e. 3 different levels of access: basic, intermediate and optimal).
I am using the same socioeconomic and demographic predictors (independent variables) across the three models, only the dependent variables changes (although they are naturally quite similar since they all look at water access). I have found notable differences in the pseudo R2 (using the adjusted Cox & Snell, i.e. Nagelkerke R2) across these three models The resulting values are 0.33, 0.54 and 0.60. From my review of existing research and other statistical tests (such as Chi2) it makes sense that the same variables are better at predicting optimal access as opposed to basic access. I have found literature which indicates that pseudo R2s can be compared when the independent variables change (i.e. improvement within the model due to a new predictor variable). However, I could not find any information on whether this is also reasonable to assume when the dependent variables change but the independent variables stay the same. Could I say that my chosen predictor variables explain 33% of the variance in model one and 60% in model three, thus indicating that demographic and socioeconomic factors are more important for predicting optimal as opposed to basic water access?
Thanks!
The following paper maybe useful.
https://www.glmj.org/archives/articles/Smith_v39n2.pdf
Hello Katharina,
I'd be wary of comparing any pseudo R-squared value across models having different DVs. I would suggest instead that comparing sensitivity, specificity, or AUC for the resultant models would be more informative.
Good luck with your work.
- Badges
- Science topic
- Similar topics
- Mathematical Sciences
- Statistics
- Statistical Software
- SPSS