Power calculation; - Is it possible to calculate statistical power after all experiments and tests ha (Nov/08/2006 )
Hallo people,
I am in an urgent need of your help! I submitted my paper (case-control study). I discussed the problem of small samples, which one reviewer appreciated but required to calculate statistical power. (His words: "While the lack of power to detect an association is openly discussed by the authors, a power calculation has not been carried out."
Just to bring you into the picture of what I did with regards to statistical evaluation, here is a paragraph from my paper: "Allele and genotype frequencies were determined by direct counting. Statistical differences in allele or genotype distributions were analyzed by either a chi2 or Fisher’s exact probability test. Significance was defined using a Bonferroni corrected P value lower than 0.05. The strength of the associations observed was estimated by calculating the odds ratios (OR) according to Wolf’s method with a Haldane’s correction."
Does anyone know what exactly am I asked to do? And how to do it?
Please help me, this is the only one reviewers' requirement and I need to have the paper submitted so much!!!
Thanks a lot to all of you.
Paja
Statistic power is the probability of rejecting a false statistical null hypothesis.
There are stat methods for it, google it (have myself never done).
Well i don´t know but sometimes I´ve been told that the statistic power means if the size of your sample is enough. I mean, if you have used enough cases and controls to conclude that your results are reliable. Obviously, the more samples you have, more statistically powerful your results will be.
Hi Paja,
Power is certainly one of the statistic fetaures that is most commonly misunderstood. There are different good introdcutions to the concept and how to use it. I would recommend that you try using the G-Power tool, as it comes along with quite a good introduction. Or take a look at StatMate (http://graphpad.com/StatMate/statmate.htm). You can download the trial version. It is really easy bit can help a lot.
One of the major problems might still be, that the reviewer himself also only has some vague ideas about power...
Good luck with your paper.
If you need more direct help, give us more details of your analysis!
PS: Are you reporting a negative finding? Than you need to fix a "meaningful" or clinically significant OR and perform a sample size calculation. If your sample size was bigger than the calculated sample size, the power of your study is high enough. If not, you might miss an association. Have a look at this paper, then you should understand what I mean: Pubmed ID: 16713503