Choose Language Hide Translation Bar
Community Trekker

Multi-level contingency analysis

I am doing a contingency analysis / Chi square test for independence on two categorical variables where and each variable has more than two levels. AFSC has 8 levels and Impact is a composite variable (from survey) that is a summartion of 5 different likert-scale questions (implication being it has 20 different levels). I am trying to determine iwhich AFSCs affect Impact. By default, Fit Y y X does a contingency analysis with a mosaic plot and Chi Square tests for LR and Pearson. JMP tells me that yes, AFSC does affect Impact with P values of 0.0012 and 0.0082. This is nice but I need to know WHICH AFSCs affect Impact and what the directions are. How do I do this in JMP? Is just AFSC = 1 that makes a difference, or is it AFSC = 1, 3 and 5? How do I determine the direction of the relationships? It could be AFSC = 1 leads to increased Impact and AFSC = 3 leads to decreased Impact? Is there a way to easily do this analysis in JMP?

0 Kudos
Super User

Re: Multi-level contingency analysis

I have seen in some cases where people use a One way anova esepcaily when working with a sum or average of Likert scales.  The overall test provides a level of significance, and then multiple comparison tests assess which categories are different.


However, a logistic model might be more appropriate. Instead of nominal logistic, since it is a likert scale, an ordinal logistic might be better. 


You did not mention which version of JMP you are using. After reviewing ordinal logistic in the Scripting Index, look up Categorical analysis.


JMP sample data has an attached script with the model using multiple x (independent) variables.



0 Kudos