Looking for a JMP test that is comparable to SAS 9.4's "CATMOD" procedure

[#1]

I am looking to compare 3 treatments (control, 1, and 2) that yield categorical results (yes/no). I have a summary table that lists the treatment, outcome, and count for each treatment. Previous applicable work in my field has utilized SAS 9.4's CATMOD procedure with the generalized logits response function to determine if treatment effect was significant. When it was, they used the contrast statement of the CATMOD procedure to determine if certain treatments were significantly different from each other. I do not have SAS, but do have JMP Pro (16.2) and so am looking for a comparable test to run.

Based on what I can read on the Help Center guide (https://documentation.sas.com/doc/en/pgmsascdc/9.4_3.4/statug/statug_catmod_overview.htm), the CATMOD procedure using the generalized logits response function is a logistic regression that uses the maximum likelihood estimation method to obtain parameter estimates (https://documentation.sas.com/doc/en/pgmsascdc/9.4_3.4/statug/statug_catmod_gettingstarted02.htm) and the contrast statement uses Wald statistics to make any further comparisons (https://documentation.sas.com/doc/en/pgmsascdc/9.4_3.4/statug/statug_catmod_syntax03.htm). I'm not a statistician, though, so if that is wrong please let me know.

Based on the above assumptions, I'm thinking that a nominal logistic regression analysis in JMP would be comparable as it uses the ML method to estimate parameters and I can select to have the Wald statistics calculated to compare effects. Is this accurate? Or should I be using a different method?

Thank you! Lea

[#2]

Take a look at the Fitting Linear Models guide in the JMP Help. Pay close attention to the Nominal and Ordinal Logistic Regression section through the Fit Least Squares platform (the JMP platform is 'like' a SAS PROC). JMP uses the generalized logits when fitting the nominal logistic regression model, so your response should use the nominal modeling type.

[#3]

Hi Mark - JMP automagically choose nominal logistic regression if you only have 2 response types. As my treatment produces a yes/no response, it was actually chosen for me. So, yes, I would be using the nominal logistic regression model and my response is nominal.

Does anyone know if the contrast statement is indeed the same as the Wald statistics calculated when selecting Wald Test in the analysis output window after fitting the nominal logistic regression model?

Thanks, Lea

[#4]

JMP is simply reporting the square of the ratio of the estimate to its standard error. See this example in JMP:

Names Default to Here( 1 );

dt = Open( "$SAMPLE_DATA/Big Class.jmp" );
nomLogRegr = dt << Fit Model(
	Y( :sex ),
	Effects( :weight ),
	Target Level( "M" ),
	Personality( "Nominal Logistic" ),
	Run( Likelihood Ratio Tests( 1 ), Wald Tests( 1 ), Logistic Plot( 1 ) )
);
nomLogRegrW = nomLogRegr << Report;

est = nomLogRegrW["Parameter Estimates"][NumberColBox("Estimate")] << Get( 2 );
se  = nomLogRegrW["Parameter Estimates"][NumberColBox("Std Error")] << Get( 2 );

WaldChiSqr = (est / se)^2;

Show( WaldChiSqr );	// WaldChiSqr = 1.08416718977359;

 This CATMOD documentation shows that Chi square is calculated the same way.

[#5]

Great. Thank you!