Barite | Level 11

## Testing the equality of 2 proportions / Chi-square / Z-test

I want to test the equality of 2 proportions, but I want to do it for each level of a nominal, polychotomous variable. For example, gender (dichotomous) and race (polychotomous). I want to compare the proportion of males that are caucasian, for example, to the proportion of females that are caucasian, and so on. I want the same comparison for each level of the polychotomous variable race. If there are 5 levels of race, I want 5 comparisons w/ p-values, difference in proportion and CI of the difference.

I understand that if the 2 variables are dichotomous I can use PROC FREQ w/ the CHISQ and RISKDIFF options for the statistical significance, the difference between proportions and the CI for that difference, but is there a way to perform this analysis without having to dichotomize a polychotomous variable?

Thanks!!

1 ACCEPTED SOLUTION

Accepted Solutions
Super User

## Re: Testing the equality of 2 proportions / Chi-square / Z-test

Your BY variables shouldn't be in the TABLES statement

4 REPLIES 4
Super User

## Re: Testing the equality of 2 proportions / Chi-square / Z-test

Include the polychotomous variable as a BY variable.

Barite | Level 11

## Re: Testing the equality of 2 proportions / Chi-square / Z-test

@Reeza

When I do that I get the following NOTES in the log:

NOTE: No statistics are computed for race_ethn * gender because all data are missing.
NOTE: The above message was for the following BY group:
race_ethn=.
NOTE: No statistics are computed for race_ethn * gender because race_ethn has less than 2 nonmissing levels.
NOTE: The above message was for the following BY group:
race_ethn=Caucasian
NOTE: No statistics are computed for race_ethn * gender because race_ethn has less than 2 nonmissing levels.
NOTE: The above message was for the following BY group:
race_ethn=Black/African American/Hispanic
NOTE: No statistics are computed for race_ethn * gender because race_ethn has less than 2 nonmissing levels.
NOTE: The above message was for the following BY group:
race_ethn=Native American/Asian/Pacific Islander
NOTE: No statistics are computed for race_ethn * gender because race_ethn has less than 2 nonmissing levels.
NOTE: The above message was for the following BY group:
race_ethn=Other
NOTE: There were 2774 observations read from the data set SURVEY.CSCSP_2016_POST_SORT.

Super User

## Re: Testing the equality of 2 proportions / Chi-square / Z-test

Your BY variables shouldn't be in the TABLES statement

Super User

## Re: Testing the equality of 2 proportions / Chi-square / Z-test

```Possion regression might do this.
proc genmod + offset= option

```
Discussion stats
• 4 replies
• 2362 views
• 1 like
• 3 in conversation