When CURVEFIT fits a nonlinear model, what it actually does is to fit a linear model to transformed data. Although the saved predicted values and residuals are in the original metric, the output results such as the ANOVA table and R^2 statistics are in the metric of the linear model on transformed data rather than in the original metric.
I think that IBM SPSS Statistics should be showing information for model comparisons in the original metric. A better design would be to actually fit the true nonlinear models. Can this feature be implemented in a Statistics release?
This has been submitted as an Enhancement request to IBM SPSS Development for consideration to implement in a future release. However, IBM cannot commit if a feature will be implemented in a future release of IBM SPSS Statistics.
(Reference ID: ECM00193474)