04-26-2012 12:17 PM
I have a dataset that looks like the following,
variable x :1 = very worried, 2 = slightly worried ,3 = Not worried
variable y : 1 = more likely, 2 = no change ,3 = less likely
I would like to comparsion the factor variable x and outcome variable y.
Which 's test inside sas will give the right p-value?
04-26-2012 12:43 PM
A good general purpose test for the relationship between two categorical variables is Fisher's exact test. If your number of observations is not too large ( < 100 say), use the following :
proc freq data=myData;
exact Fisher / maxtime=20;
If the computation doesn't terminate within 20 seconds, add the MC option to the EXACT statement.
05-03-2012 10:33 AM
"Right" is a very slippery term in this sort of analysis. There is probably no single answer. There is a huge literature on testing association among ordered variables.
The Fisher's Exact Test that PG recommended is a general purpose test that could be used, but it is known to be conservative. If your sample size is larger, then you could use the Chi-Squared test that is part of PROC FREQ to address general association.
Neither of these tests take advantage of the inherent ordering of the two variables. I prefer Kendall's Tau-b (also part of FREQ).