Nonlinear Enhancements

[MShroff]

Hello,

Can Nonlinear be enhanced to provide confidence intervals around the parameter values and a measure such as sigma (as is provided for Fit Parametric Survival) and R^2 (I know R^2 may have limited / restricted applicability so sigma might be better). This is available in competitor tools.

Regards

Mehul Shroff

[SamGardner]

@MShroff confidence intervals for parameters can be determined by clicking the "Confidence Limits" button in the Control Panel in Nonlinear, after the model fit solution has been found.  

[MShroff]

Thanks, @SamGardner, for this information.

Can you also comment on the availability of sigma or whether it can be suggested for a future enhancement?

Regards

Mehul Shroff

[SamGardner]

We will look into possibly adding more measures of model fit in Non-linear.  For "sigma", do you mean the root mean square error for the model fit?  The RMSE is reported in the Solution section of the report.  

[MShroff]

Hello,

No - I was referring to the measure of spread (e.g., a lognormal sigma). This is conceptually (though not mathematically) the inverse of R^2.

Regards

Mehul Shroff

[SamGardner]

@MShroff the Sigma parameter is intrinsic to the model that is chosen.  In the case of a Lognormal distrubution for survival analysis, if you want to see an example of this, use the Locomotive.JMP example data table and

• Run Parametric Survival with Time as the response and Censor as the Censor role, and use the Lognormal distribution and run the report.  You see the parameter Sigma is reported in the parameter estimates.
  
   
• Now fit the same model using Nonlinear.  Use Time as the response and the LogNormal Column as the Loss.  If you view the column information for the LogNormal column you can see that the Sigma parameter is specified as part of the loss function.  
  
  
  Run the report and you see the same value of Sigma is reported as the estimate, same as the Parametric Survival example.