Comment 2: The inter-relationships among soil properties (e.g., amorphous manganese oxides and soil clay, soil total Hg and available Hg) made the stepwise regression results artificial. In other others, the authors could not assume they were independent variables by ignoring such inter-relationships, although some published papers have done this. The reviewer would recommend the authors to choose a better statistical method, which can exclude the inter-relationships among dependent variables, to reanalyze their data. These statistical methods could be found widely in the literature.
Let us not get worried about excluding the interdependent variables while executing step down regression analysis. Develop correlation matrix to reveal such interdependence amongst variables...
Peiman -
I don't think stepwise regression is what you want to use. Based on theory you could have some alternative models you could compare using "graphical residual analyses," and you could put more than one model's results on the same scatterplot for comparison. This is to study model fit. You could use "cross-validation" to study the possibility that you have "overfit" to your particular sample. People use LASSO also, but I think subject matter knowledge and cross-validation may help you to avoid spurious results.
Cheers - Jim
PS - The following is with regard to an important issue often ignored for regressions of form y = y* + e. This is most useful for predictions for finite populations:
https://www.researchgate.net/project/OLS-Regression-Should-Not-Be-a-Default-for-WLS-Regression
- Badges
- Science topic
- Similar topics
- Papers