topic Re: Hi in Statistical Procedures

Hi

Stat_vijay — Fri, 11 Jul 2014 06:48:15 GMT

For data has 2 treatment and 3 category.

trt1 : 1 0 0
trt2 : 0 0 0

is it possible for do fisher exact test in proc freq?

Re: Hi

Stat_vijay — Fri, 11 Jul 2014 06:51:07 GMT

Hi any one please can reply above question

Re: Hi

Ksharp — Fri, 11 Jul 2014 14:15:50 GMT

Yeah. I think so. and change your data structure firstly. But your cell freq better all have value . what is your whole sample data.

proc freq;

table treatment*category /fisher;

weight wth;

run;

Xia Keshan

Re: Hi

JasonXin — Sat, 12 Jul 2014 13:50:58 GMT

Hi,

You can look into proc transpose to change the structure to fit the FREQ procedure input requirement.

Jason Xin

Re: Hi

stat_sas — Sat, 12 Jul 2014 16:52:44 GMT

There is no problem in performing fisher's exact test using proc freq but your data does not meet the requirement for the test.

Re: Hi

Reeza — Sat, 12 Jul 2014 19:13:45 GMT

Your sample size is too small to draw any statistically significant conclusion. Performing a test would be misleading.

Re: Hi

SteveDenham — Mon, 14 Jul 2014 14:22:13 GMT

Listen to . Note that for 2x2 tables, Fisher's Exact Test will require at least 5 observations per row (and complete separation) to get a significant result. If you do not have that, you may well be wasting your time trying to get a p value.

Steve Denham

Re: Hi

stat_sas — Mon, 14 Jul 2014 14:49:38 GMT

Hi Steve,

I think Fisher's Exact test does not depend on large-sample distribution assumptions, and appropriate even for small sample sizes as well. Please correct me if I am wrong, is it expected cell frequency or obervations per row that should be at least 5?

Regards,

Naeem

Re: Hi

SteveDenham — Mon, 14 Jul 2014 15:07:56 GMT

Sorry, what I wrote was easily misinterpretable, and wrong in the details anyway. For a balanced design, you need at least three observations per row, completely separated by columns to obtain significance using Fisher's Exact Test Nothing to do with asymptotic assumptions or expected cell frequencies.

data one;

input a b wt;

cards;

0 0 0

0 1 3

1 0 3

1 1 0

;

proc freq data=one;

tables a*b/exact;

weight wt;

run;

Note that if you redistribute the 6 obs or have fewer total observations, no significance can be found.

Steve Denham

Re: Hi

stat_sas — Mon, 14 Jul 2014 15:44:21 GMT

Thanks Steve - This is interesting to know.

Naeem