EXTSIG -- Exact Significance Tests Comparing Two
Binomial Proportions
EXTSIG performs tests for the difference in two binomial proportions.
After a two by two table is entered, the program reports one- and
two-sided p-values from Fisher's test and the chi-square test.
EXTSIG also computes exact significance levels by the unconditional
method. A sort criterion is chosen (e.g., difference in proportions,
chi-square, Fisher's exact p-value) that determines the support for
the alternative hypothesis for each outcome. The set of outcomes is
identified which favors the alternative hypothesis at least as much as
the observed table. The significance level or p-value of the observed
outcome is the maximum over p of the probability of this set assumming
p1=p2=p where p varies from 0 to 1.
Advantages: (1) the significance level is exact even with tables with
small entries; (2) usually the power of the exact test is somewhat
greater (10% in small tables) than that of the chi-square or Fisher's
test. Disadvantage: there are many exact tests -- one for every
reasonable sort order -- and they give different answers. In tests,
Fisher's p-value works well compared to the four other sort criteria
incorporated in EXTSIG and it is the default.
WHERE TO GET IT
Software written by members of the section is freely available to
anyone. Reposting on other archives is encouraged. The code is
furnished in source form and as PC and Macintosh executables.
Available Software:
http://odin.mdacc.tmc.edu/anonftp/
Barry W. Brown
Department of Biomathematics,
Box 237
University of Texas M. D.
Anderson Cancer Center
1515 Holcombe Blvd
Houston, TX 77030
[log in to unmask]
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|