PakMediNet Discussion Forum : Biostatistics : what test to apply?
what test to apply?
I am stuck up to find out the p-values for freq of signs in one of my article. I don't know how to apply Chi square or whatever test applicable. Table is as under
l help
SIGN GP A Gp B p-val
(n=16) (n=15)
conjunct 10 8 ?
keratitis 6 6 ?
uveitis 3 5 ?
glaucoma 1 3 ?
Posted by: eyespec Posts: 2 :: 14-07-2006 :: | Reply to this Message
Re: what test to apply?
Salaam,
It would be helpful if you tell us the difference between group A and group B. Another good practice is to describe null hypothesis.
From the available information, I think you should use a Fisher's exact test (better than Chi square test especially for small sample sizes).
For this we require 2x2 contigency tables, as given below:
Conjunctivitis
YES NO Total
Group A 10 6 16
Group B 8 7 15
p= 0.7224
Keratitis
YES NO Total
Group A 6 10 16
Group B 6 9 15
p= 1.000
Uveitis
YES NO Total
Group A 3 13 16
Group B 5 10 15
p= 0.4331
Glaucoma
YES NO Total
Group A 1 15 16
Group B 3 12 15
p= 0.3326
None of the p value is significant.
I suggest you try it yourself with a free online interactive calculator available at http://graphpad.com/quickcalcs/contingency1.cfm
This excercise will be useful for future.
If you would like me to go through your final munuscript before submission, please do not hesitate to drop a line.
Regards,
Rehman Siddiqui
Clinical Lecturer and Specialist Registrar,
Department of Ophthalmology
University of Aberdeen
UK
Posted by: asiddiqui Posts: 26 :: 16-07-2006 :: | Reply to this Message
Re: Re: what test to apply?Why have you used Fischer Exact test when the variables on both groups are Continous? Fischer exact is used to see the association between proportions.
Posted by: yasir Posts: 90 :: 17-07-2006 :: | Reply to this Message
Re: what test to apply?
Of course, it will be helpful if 'eyespec' can tell us whether his numbers describe a continous or discrete variable. To me these are number of patients in each group with the particular ailment, in which case he should use Chi-square or Fisher's Exact test.
One suggesstion, instead of performing a separate 2x2 analysis for each outcome, investigator may want to perform a full Chi-square test first including all the outcomes and see is there any difference in any of the groups. If Chi-square statistic is significant, then one can try to find which one (or all of the outcomes) are significant.
In my view, there are two advantages of this approach. One is that after one test one will know whether one has to do more tests or not, and the second is that one is less likely to run in the problem of testing multiple hypotheses and the need to correct p-values accordingly.
The overall Chi-square statistic for this data is 1.63 (degree of freedom = 3) which is not significant and, therefore, you don't need to do any further testing. Investigator may like to collect more data as the overall power of the study is very low except under very unrealistic assumptions.
Posted by: rqayyum Posts: 199 :: 17-07-2006 :: | Reply to this Message