Box Tidwell for logistic regression

[aande2]

Hi JMP community,

One of the assumptions of logistic regression is linearity of variables and log odds. However, I still have not figured out how to test this in JMP. The specific test I have seen elsewhere is the Box Tidwell test. Is there a way to get this test on JMP? Failing that, is there another way to test linearity of a logistic regression model?

Thanks!

[Sarah-Sylvestre]

Hi @aande2, thank you for your suggestion! We have captured your request and will take it under consideration.

[SamGardner]

Hello @aande2 after some research we have decided to not pursue this.  You can do this in JMP manually by adding in the interactions or each continuous term with its log transform. That is, if Logit(Y) = aX, then fit the model Logit(Y) = a X + b X*Ln(X).  If the parameter estimate for the X*Ln(X) effect is significant, then Logit(Y) is deemed not linearly dependent on X. 

Run this script to see how to do this:

dt=Open( "$SAMPLE_DATA/Ingots.jmp" );

// Logistic regression model
dt << Fit Model(
	Freq( :count ),
	Y( :ready ),
	Effects( :heat, :soak ),
	Personality( "Nominal Logistic" ),
	Run( Likelihood Ratio Tests( 1 ), Wald Tests( 0 ) )
);

// model with Box Tidwell effects
dt <<Fit Model(
	Y( :ready ),
	Effects( :heat, :soak, :heat * :"Log[heat]"n, :soak * :"Log[soak]"n ),
	Target Level( "Ready" ),
	Personality( "Nominal Logistic" ),
	Run( Likelihood Ratio Tests( 1 ), Wald Tests( 0 ) )
);