24211_en-US

liliconcom

Show r-square in Nonlinear

On the nonlinear platform the R-square of fitted data does not comes out and It’s needed to recalculate it separately leveraging on other available statistics like RMSE. (Not a problem but it could be helpful to have r-square when it’s possible).

 

Tony Cafarelli  cafarelli.an@pg.com   Schwalbach Technical Center Germany

 

Re: Show r-square in Nonlinear - Status changed to: In Development

While RSsquare is not appropriate for all non-linear models (for example, RSsquare is not appropriate for non-linear models in which there is no intercept
term), JMP 16 will apply the following rules to determine whether to compute and display RSsquare or not:

 

• If the absolute value of the mean residual is less than 1e-8*max(1e-5,abs(yMean)) Nonlinear will now report an RSquare in the Solution report. Testing that the sum of residuals is zero generally guarantees that fitting the mean is a nested model, thus making RSquare relevant. 
• RSquares are not reported for Loss formula models.
• Exact fits will also produce an RSquare (of 1) since the sum of residuals is zero if the sum of squares of residualsis zero.

Re: Show r-square in Nonlinear - Status changed to: Open for Voting