Regarding Stephen's query about a p of 1 for Fisher's exact test:

Since p is typically the probability of obtaining results as or more
discrepant with the null, the typical interpretation of p = 1 would be that
the results you obtained were exactly what would be expected under the null,
which is clearly not the case here.  With your data, SAS produces a
left-sided p of .73 and a right sided p of .52.  Note that these sum to
greater than one.

I would not put too much effort into making sense of this, since it is
highly unlikely that Fisher's exact test is appropriate for your data.
Fisher's exact test is only appropriate when both of your pairs of marginals
are fixed -- that is, when you could specify, prior to obtaining the data,
exactly what both pairs of marginal frequencies would be.  This is very
rarely the case with real data.  One example of when it could be the case is
when you have defined both categorical variables by a median split of a
continuous variable.  In that case, both sets of marginals will be fixed,
with each marginal frequency being .5N.

A traditional Pearson chi-square, NOT corrected for continuity, would be
appropriate.  The correction for continuity is appropriate only with fixed
marginals (as with Fisher's exact).  For your data, the Pearson chi-square
produces a p of .78.  With the correction for continuity (which approximates
the Fisher's exact test), the p is 1.

For more details, consult Camilli & Hopkins, Psychological Bulletin, 86:
1011-1014.

 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~  
Karl L. Wuensch, Department of Psychology,
East Carolina University, Greenville NC  27858-4353
Voice:  252-328-4102     Fax:  252-328-6283
mailto:[EMAIL PROTECTED]
http://core.ecu.edu/psyc/wuenschk/klw.htm


---
You are currently subscribed to tips as: [EMAIL PROTECTED]
To unsubscribe send a blank email to [EMAIL PROTECTED]

Reply via email to