topic Re: Interaction visualization in PROC LOGISTIC in Statistical Procedures

Interaction visualization in PROC LOGISTIC

LucyB — Fri, 28 Apr 2017 19:24:33 GMT

I am running a proc logistic and am looking specifically at a 3 way interaction. Here is some descriptives on my data:

Outcome: mortality (y/n)

Var 1: gender (M/F)

Var 2: dichotomous lab value (above/ below certain critical cutoff- y/n)

Var 3: second dichotomous lab value (above/ below certain critical cutoff- y/n)

I am looking to see who is more at risk for the mortality outcome and have included an interaction for these terms in the model. My interaction p-value is not significant, but I want to report directions. Since I have 3 dichotomous variables, what would be the best way to present this (preferably graphically) in regards to who is more likely to experience mortality?

Re: Interaction visualization in PROC LOGISTIC

PeterClemmensen — Fri, 28 Apr 2017 19:59:04 GMT

You can probably use

effectplot interaction;

as documented here

https://support.sas.com/documentation/cdl/en/statug/63347/HTML/default/viewer.htm#statug_logistic_sect009.htm

Re: Interaction visualization in PROC LOGISTIC

LucyB — Fri, 28 Apr 2017 21:51:31 GMT

Thank you, so this plots the probabilities for 2 of my predictors but fixes the value for my 3rd. I want to compare across all 3 somehow..

Re: Interaction visualization in PROC LOGISTIC

Rick_SAS — Sat, 29 Apr 2017 01:02:53 GMT

Look at the article "Visualize an ANOVA with two-way interactions" and see if you can figure out how to extend it to your three binary variables. You will end up with 6 box plots.

Re: Interaction visualization in PROC LOGISTIC

Ksharp — Sat, 29 Apr 2017 03:28:03 GMT

Also check

effectplot  fitslice /  sliceby=  plotby= ;