12-10-2011 03:36 PM
What is the best method of conducting a bivariate analysis of two categorical variables? I used proc freq but I wasn't sure what to look for as a result. proc ttest doesn't seem right because I have no results for the Satterthwaite method. Any advice would be great.
12-10-2011 04:19 PM
yes its in a Yes or No format. Proc freq works out fine. Thanks. Now what if I have a continuous variable vs a categorical variable. I'm thinking proc reg should work?
12-10-2011 05:09 PM
I am not a statistician. However, if your dichotomies are all 1 and 0, and your continuous variables are at least interval or interval appearing (e.g., a Likert-type scale), I think you could use either for both.
12-11-2011 09:16 PM
You can't do it with proc reg because residual is required to conform Normal Distribution.
But for your situation, Logistic Model or proc mixed is a good choice.
12-12-2011 12:19 AM
KSharp, Just to make the discussion and possibilities more interesting, here is the start of a chapter from one of Gene Glass's books: http://epm.sagepub.com/content/26/3/623.extract
As I learned, again while not a statistician, one can assume that a dichotomy was forced from a factor that was actually normalliy distributed.
12-12-2011 01:05 AM
Yes. dichotomy will conform Normal Distribution, if your sample size is quite large.
Actually Normal Distribution is from bivariate distribution.
Normal Distribution can be deduced by formula of bivariate distribution when N is large enough.