In goo.gl/PSn3NG
Minitab´s statician explains why is there no R-squared for nonlinear regression.
1) Assuming it is right, why does Origin, SAS and others provide R2 for nonlinear reg.?
3) Is generalized R2 an alternative?
2) How to compare two different nonlinear regressions? RMSE? AIC or SE?
3) Is feasible calculate and use RMSE and AIC for nonlinear reg.? Is SE the best estimator for comparing two regressions?
Thanks in advance!
R square captures explanatory power of independent variables only when there is a linear association between the dependent and independent variables, as a result if the association is nonlinear R square fails to explain such association. The best approach to compare two nonlinear regression models is their predictive power.
What I usually do is to use the developed equation from the nonlinear regression to generate predicted values then plot them against measured or otherwise obtained actual values an y and x axes respectively then get the R square for comparison purposes between two or more nonlinear models.
.
the article below gives a nice review of "pseudo R-square" for GLM
http://www.stat.ufl.edu/~aa/articles/zheng_agresti.pdf
.
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics
- Regression