Date:         Tue, 9 Sep 2008 08:35:41 -0400
Reply-To:     Kevin Viel <citam.sasl@GMAIL.COM>
Sender:       "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
From:         Kevin Viel <citam.sasl@GMAIL.COM>
Subject:      FREQ: Yate's Corrections versus Fisher's Exact

Greetings,

  I am now on our fourth revision of an article.  Yes, I said fourth.  In
my opinion, it is a topic that will force a revolution in medical
treatment.  Unfortunately, our sample size is small.  The disease is rare,
affecting approximately 11 males per 10,000 males in the country (X-
linked).

  We received the following review:

"Unfortunately, the statistical significance of the results are very
marginal.  The key comparison of 9/19 [exposed] with [outcome] vs 12/57 is
on the edge of significance at the 0.05 level (0.026 on an uncorrected Chi-
square, 0.054 with the Yates correction, the lower bound of the 95% Exact
confidence interval is 0.96)."

  I was somewhat puzzled.  Wikipedia suggest that the Yates' correction be
applied with some cell has an expected count of less than 5.  Our smallest
cell has an expected count of 5.25 = 21 * ( 19 / 76 ).


  Further, I would forego the use of the Yate's correction and go straight
to the Exact statistic:


data one ;
  input f e d ;
datalines ;
9  0 0
12 1 0
10 0 1
45 1 1
;
run ;

proc freq data = one ;
  tables e * d / exact ;
  weight f ;
run ;

  Does my reasoning seem sound?  I appreciate any comments, criticism, or
advice.

Thanks in advance,

Kevin