Download the program G*Power and try the one way ANOVA, if you want to see the difference in 1 feature. However, you will be highly vulnerable to outliers and skewness in the data, so power is likely to be overestimated. I would definatley consider more cases then the estimate. Also, if you plan to conduct multiple tests, you can check power under Bonferroni (just change alpha = 0.05 to alpha = 0.05/number of tests). Really, at the sample size for most experimental studies, one should use permutation-ANOVA, bootstrap ANOVA or perhaps Kruskal-Wallis: they are more robust small-sample statistics, but I doubt reviewers will ask you for that.

This link should provide you with what you need:

https://stats.idre.ucla.edu/other/gpower/one-way-anova-power-analysis/

Download the program G*Power and try the one way ANOVA, if you want to see the difference in 1 feature. However, you will be highly vulnerable to outliers and skewness in the data, so power is likely to be overestimated. I would definatley consider more cases then the estimate. Also, if you plan to conduct multiple tests, you can check power under Bonferroni (just change alpha = 0.05 to alpha = 0.05/number of tests). Really, at the sample size for most experimental studies, one should use permutation-ANOVA, bootstrap ANOVA or perhaps Kruskal-Wallis: they are more robust small-sample statistics, but I doubt reviewers will ask you for that.

This link should provide you with what you need:

https://stats.idre.ucla.edu/other/gpower/one-way-anova-power-analysis/

​ Dear Dianne

Ther are many software can used to calculate sample size ( SPSS , Minitab , G*power )

My advice to use G*Power. G*Power software is an effective tool to calculate sample size for many ranges of experiments. Also, you can determine effect size and power of the test, G*Power is free to download and easy to use after reading the manual, the download link:

http://download.cnet.com/G-Power/3000-2054_4-10647044.html

The link to download G*Power manual:

http://www.gpower.hhu.de/fileadmin/redaktion/Fakultaeten/Mathematisch-Naturwissenschaftliche_Fakultaet/Psychologie/AAP/gpower/GPowerManual.pdf

I hope that be useful for you.

Good Luck

For animal models studies you do not need to calculate the sample size by formula because it needs either the disease prevalence or the area population both can not be decided for animals . Majority of the animal studies conducted contain 10 animals per group so you can take a total of 40 animal divided in 4 groups having 10 in each if further help required you can ask in details by providing your study details

How come i often see experiments using 6 mice or rats per group? Is there a justification on the use of this standard sample size?

I think Ashique Ali's answer is totally misleading!! Why should general statistical rules, like power analysis etc, not apply to animal studies??? It is not right to do something, because a majority of people are doing it, that is nonsense. Please have a look at the current discussion about non-reproduceable finding in science. It is most often because of such studies, which are greatly underpowered. There may be cases in which 10 individuals per group may suffice, but instead of just assuming it, you should show it!! Khalid Hassan's and Lasse Giil's proposals are perfectly fine. If you are neglecting basic concepts of (frequentistic) statistics, you are doing it wrong and in turn, your results and conclusions will also possibly be wrong.

Thank you. I used gpower and its indeed very easy and simple to use. I guess other studies used this software too, coz after I put in the means from similar studies, i am also getting a sample size of 6 per arm.

The sample size depends on the size tolerance of errors.. For confidence interval 95% and error margin 40% You need 6 animals independent to population size. Is this degree of error appropriate for Your further more precise investigation?

I explain the minimal number of 6 for sampple size on an false positive error of 20%. 5 tested object lead to 1 error event whole in number. The 6th object of observation gies overweigh for decision making.

Animal studies are not exempt from the clear association between samle size and the risk of making a type II error on any statistical hypothesis test with a given alpha. The reason animal studies are small is mainly due to ethical concerns and financial constraints. One of the main reasons they are often not reproducible, is that small samples gives higher risk of type I errors, often due to sensitivity to one or two outlying cases, or multiple tests of hypotheses (including repeated measurements of the same protein or gene). The suggestion that power does not apply to small samples or controlled experiments is simply plain wrong. Anyway, seems like the problem has been solved. Great!