Date:         Mon, 21 Dec 1998 23:31:46 GMT
Reply-To:     Steve Gregorich <gregorich@PSG.UCSF.EDU>
Sender:       "SAS(r) Discussion" <SAS-L@UGA.CC.UGA.EDU>
From:         Steve Gregorich <gregorich@PSG.UCSF.EDU>
Organization: University of California, San Francisco
Subject:      Re: Still have problem with Chi Square...
Content-Type: Text/Plain; charset=US-ASCII

The SAS syntax to generate the probability
of a chi-square statistic should be:
 
   pvalue= 1 - probchi(chi, df);
 
where for this application, df=3.
 
NOT:
>                pvalue=probchi(N-1,chi);
 
The latter syntax, swaps the position of
the chi-sqaure varaite and the model df and
does not subtract the obtained quantity
from unity.
 
 
A chi-square of 6.35 judged against 3 df
has a p-value of 0.0955--not 0.1605.
Still, the null should be rejected at
conventional levels of statistical
significance.
 
 
 
Steve Gregorich
 
 
 
 
>data one;
>input obs expec;
>cards;
>439 9
>168 3
>133 3
>60  1
>;
>data two;set one nobs=N;
>retain chi 0;
>expec=expec*800/16;put obs= expec=;
>chi=chi+((obs-expec)**2)/expec;
>if _n_=N then do;
>                put chi=;
>                pvalue=probchi(N-1,chi);
>                put pvalue=;
>              end;
>run;
>
>In Log, you have:
>OBS=439 EXPEC=450
>OBS=168 EXPEC=150
>OBS=133 EXPEC=150
>OBS=60 EXPEC=50
>CHI=6.3555555556
>PVALUE=0.1605662382
>So you should accept the null hypotheses
>Javier