JMP > Statistical Methods > How to Compare Multiple Proportions?

[#1]

Hi JMP Community,

I would like to know which statistical method(s) in JMP is the most appropriate to test multiple proportions? I am familiar with the Fit Y by X Contingency analysis for comparing two proportions. Still, I cannot figure out if there is a generalized platform to compare multiple proportions across conditions or time points.

Typical Scenario (see mock data attached):

I monitor a binary outcome (e.g., response / non-response) through time and for multiple conditions (e.g., treatment, baseline status, and gender). I want to compare the effect of the conditions globally and at specific time points. 

System: Windows 10

JMP Version: 16.1

Any suggestions would be welcome.

Best,

TS

[#2 · Accepted solution]

Hi @Thierry_S ,

just by looking at the column names and imagining the scenario behind them i think i have an idea. it may be exactly what you need in some circumstances. If i understand correctly, ID is repeated over time, time is the "within" effect while gender, original status and treatment are "between" effects.

Since there is time and baseline status i would think a mixed model for repated measures is the way to go.

i would get the linear mixed model add in: https://community.jmp.com/t5/JMP-Add-Ins/Generalized-Linear-Mixed-Model-Add-in/ta-p/284627

Once you get it you can use it to manually enter the following roles:

Fit Model(
	Y( :OUTCOME ),
	Effects(
		:BASELINE STATUS, :TREATMENT, :GENDER, :TIME, :BASELINE STATUS * :TREATMENT,
		:BASELINE STATUS * :GENDER, :BASELINE STATUS * :TIME, :TREATMENT * :GENDER,
		:TREATMENT * :TIME, :GENDER * :TIME, :BASELINE STATUS * :TREATMENT * :GENDER,
		:BASELINE STATUS * :TREATMENT * :TIME, :BASELINE STATUS * :GENDER * :TIME,
		:TREATMENT * :GENDER * :TIME,
		:BASELINE STATUS * :TREATMENT * :GENDER * :TIME,
		:ID[:BASELINE STATUS, :TREATMENT, :GENDER] & Random
	),
	Personality( "Generalized Linear Model" ),
	GLM Distribution( "Binomial" ),
	Link Function( "Logit" )
);

problem with the current fit model platform (jmp pro 16) is that it doesn't sport this so you need to use the add in.

When running with the addin this full factorial repeated measures model i get singularity issues so perhaps some interactions that do not have any variance should be removed. perhaps even worth running first without interactions.

as a side note, this nifty tool can help sort the out the "between" / "within" effects very quickly: https://community.jmp.com/t5/JMP-Add-Ins/Full-Factorial-Repeated-Measures-ANOVA-Add-In/ta-p/23904

let us know if any of this is useful,

ron

[#3]

Hi @ron_horne,

Thank you for the thorough answer. The critical step I had missed was using the Binomial distribution in the GLM model.

I'll look into the Add-in.

Best,

TS