| Date: | Wed, 29 Jan 1997 18:48:21 GMT |
|---|
| Reply-To: | Jay Weedon <j_weedon@ESCAPE.COM> |
|---|
| Sender: | "SAS(r) Discussion" <SAS-L@UGA.CC.UGA.EDU> |
|---|
| From: | Jay Weedon <j_weedon@ESCAPE.COM> |
|---|
| Organization: | None |
|---|
| Subject: | Re: one-way chi square |
|---|
| Content-Type: | text/plain; charset=us-ascii |
|---|
On Tue, 28 Jan 1997 10:31:30 -0600, Jamil Ibrahim
<ibrahim@FIONA.UMSMED.EDU> wrote:
>Does sas perform one way chi square? which proc
>Dr. Jamil M. Ibrahim, Ph.D.
>Institutional Research Associate, LRC
>Assistant Professor, SHRP
>The University of Mississppi Medical Center
>2500 North State Street
>Jackson, Mississippi 39216-4505 USA
>Telephone: 1-(601) 984-1197 Fax:1-(601) 984-1205
>INternet: ibrahim@fiona.umsmed.edu
Don't know of any existing routines, but here's a method:
* Assume X is the variable to be tested;
* First generate observed frequencies using proc freq;
proc freq noprint data=whatever; tables x /out=freqs;
* Get total number of observations;
data total; set freqs end=last;
total+count; if last; keep total;
* Compute expected frequencies, chi-square & p-value;
* The values 0.38, 0.22 etc. below refer to expected proportions
of X belonging to category 1, 2 etc.
according to some theoretical model I'm testing - plug
in your own values (they should sum to 1). In this
example there are seven relevant categories;
data _null_;
set freqs (rename=(count=observed)) end=last;
if _n_=1 then set total;
select (x);
when (1) expected=total*0.38;
when (2) expected=total*0.22;
when (3) expected=total*0.03;
when (4) expected=total*0.03;
when (5) expected=total*0.15;
when (6) expected=total*0.06;
when (7) expected=total*0.13;
end;
chisq+((observed-expected)**2)/expected;
if last then do;
df=_n_-1; p=1-probchi(chisq,df);
put 'Chi-square = ' chisq df= p= 10.4;
end;
| 
