(Dr. David Curtis) writes:
>Please could you let us know the authors of these messages. We're interested in looking at this
>problem at the moment and think we may have come up with something useful. I'd very much like to hear
>who the other people who have been working on this are, especially if they have programs they're
>willing to distribute.
I was one of the people who responded to Don Bowden's original
query about comparing data sets with very few observations in some cells. I
wrote a DOS program that implements an algorithm of Roff & Bentzen (1989)
for calculating the probability of the Chi-Square and G-statistic
(ie liklihood ratio test) for any r x c contingency table with less than
10,000 observations.
I will send anyone a copy of the program if you send me a
formatted disk (either 3.5" or 5.25"). The program takes your data from
an ASCII file, generates many random tables with the same marginal totals,
and reports what fraction of the tables had a Chi-square (or G-statistic)
greater than your table. It will run on any DOS machine. It's reasonably
fast (eg a 2x25 cell table with 250 total observations, takes 15 seconds
to run 1000 simulations on a 386/40mhz machine).
While I wrote the program to analyze data comparing allele
frequencies of RFLP VNTR's I have used it on a wide variety of biological
problems.
George Carmody gcarmody at carleton.ca
Department of Biology (613) 788-3890
Carleton University (613) 788-4497 (fax)
1125 Colonel By Drive
Ottawa, Ontario K1S 5B6
Canada
