Subscribe Bookmark RSS Feed

Validating a general regression cox model with an External Data Set

vince_faller

Super User

Joined:

Mar 17, 2015

Is this possible?  Even if I put in a validation column in fit model to split the data, all I get for validation options are AICc & BIC.

 

*EDIT* I found an example of what I mean.  

 

dt = Open("$SAMPLE_DATA\VA Lung Cancer.jmp");
dt << Fit Model(
	Censor( :censor ),
	Censor Code( "1" ),
	Y( :Time ),
	Effects( :Cell Type, :Treatment, :Prior, :Age, :Diag Time, :KPS ),
	No Intercept,
	Personality( "Generalized Regression" ),
	Generalized Distribution( "Cox Proportional Hazards" ),
	Run(
		Fit( Estimation Method( Maximum Likelihood ), Validation Method( None ) ),
		Fit(
			Estimation Method( Lasso( Adaptive ) ),
			Validation Method( AICc ),
			Set Solution ID( 143 ),
			Parameter Estimates for Centered and Scaled Predictors( 1 )
		)
	),
	SendToReport(
		Dispatch(
			{"Adaptive Lasso with AICc Validation", "Solution Path"},
			"Fit Generalized Solution Path",
			FrameBox,
			{DispatchSeg( Line Seg( 4 ), {Line Width( 3 )} )}
		)
	)
);

I actually did make a validation column here and here was the fit model dialog.  It's weird that's it's not transferring the validation column to the script (the script was made mostly by JMP). 

 Capture.png

 

4 REPLIES
markbailey

Staff

Joined:

Jun 23, 2011

The script continually opens the data table from the JMP Sample Data folder. Did you add the validation column and save the data table? If the column is missing then it will be deleted from the launch dialog / analysis.

vince_faller

Super User

Joined:

Mar 17, 2015

I have the validation column in there, I only included the script to show that  save script doesn't have a validation column in it.  But when I specifically do a cox model. Everything besides AICc and BIC are greyed out.  THe left is normal distribution, the right is cox.  

 

Capture.png

michael_jmp

Staff

Joined:

Jun 23, 2011

The Validation Column method is not available for the Cox Proportional Hazards model in Generalized Regression, as noted here:

http://www.jmp.com/support/help/13/Model_Launch_Control_Panel.shtml#456887

 

Validation Column
  • Uses the column specified in the Fit Model window as having the Validation role. The final solution is the one that minimizes the negative log-likelihood for the validation set. This option is not available when the specified Distribution is Quantile Regression or Cox Proportional Hazards.
Michael Crotty
Sr Statistical Writer
JMP Development
vince_faller

Super User

Joined:

Mar 17, 2015

Sad Face. :(   

So is there just no way to validate because the betas are actually calculating hazards and not responses?

 

I've seen stuff about calculating a concordance index for cox models, but I'm still trying parse the papers on it.  

Would that be a route to take?