topic Re: Test significant differences between two group proportions using a non-binary categorical variab in Statistical Procedures

Test significant differences between two group proportions using a non-binary categorical variable

DFoster — Tue, 10 Apr 2018 13:24:46 GMT

Hi,

I have a data set with a pass/fail variable and would like to test for significant differences between these proportions by gender (M/F). The reason I am unsure about how to proceed with this analysis is because the pass/fail variable has three levels (fail below/ pass/ fail above). I have two questions:

1) Is there a test appropriate for determining statistically significant differences between two groups when the proportion being tested is not binary?

2) Taking it a step further, is there a way to test differences between each level (e.g. comparing the proportion of males who failed below with the proportion of females who failed below)?

Thanks!

Dominique

Re: Test significant differences between two group proportions using a non-binary categorical variab

Criptic — Tue, 10 Apr 2018 13:34:27 GMT

Hey,

I would suggest you read this article about statistically significant and determin your own level as needed.
Well you could easily divide failed males by failed famles and great a measure that way

Re: Test significant differences between two group proportions using a non-binary categorical variab

PaigeMiller — Tue, 10 Apr 2018 14:01:22 GMT

1. Use the Chi-Squared test in PROC FREQ

2. Use the Chi-Squared test in PROC FREQ

Re: Test significant differences between two group proportions using a non-binary categorical variab

StatDave — Tue, 10 Apr 2018 14:01:32 GMT

See this note - particularly the last section and the note it links to regarding multiple comparisons.

Re: Test significant differences between two group proportions using a non-binary categorical variab

sld — Tue, 10 Apr 2018 19:42:23 GMT

I can't quite make sense of your 3-level response ("three levels (fail below/ pass/ fail above") but perhaps it might be considered to be ordinal. If so, use your favorite internet search tool for "ordinal logistic regression sas" for approaches that specifically account for an ordered response, as opposed to a nominal (unordered) multinomial response.