statistics / samples sizes /significane [for experement]

Post Reply New Topic

I'm going to ask around a bit more, but was hoping someone can point me in the direction of or give me rough guidelines for the kind of sample sizes needed so reasonable establishment of a hypothesis (or levels their of). I think I should be able to get a sample size of 30 or so quite easily, and currently standing around 20 maybe more. (Which I know is a bit low, but I've seen sizes of 10 used before).. I know their are some standard stats, above the T test, so it would be nice if John or someone could point me in the direction of info on doing the stats.
I have a few different groups, (about 4-6 groups) with at least 6 that fit into 4 of the groups and a few (2-4) that fit into the other 2 groups. But those in the last two groups [their may be more] are mostly less accessible or reliant on 'older' data, so a kind of extrapolation.

There's already quite a bit of data out there, or results, that I can use as citations and to expand things a bit.

But really I want to be able to get something which is 'significant' enough to be able to extend the study and so get increasingly large sample sizes which may refine both the 'experiments' and the granularity of modality.

I'm getting some very strong and distinct correlations, so the current plan is to really start base lining things, and then it should be relatively easy to start filling in the gaps and get something very good and accurate very quickly (given enough samples)


I'm not going to go into too much detail about the exact nature or experimentation (as I will most prob need a good few people of wrong-planet.net to make up the numbers a bit more once things are more refined, esp language), but I'm getting some very very high correlation. All I'll say is it's not a 'theory' of mind, and if things pan out, it's looking like it's going to be incredibly significant.

There are plenty of online power and sample-size calculators. Choose one from http://statpages.org/#power or try either http://www.stat.ubc.ca/~rollin/stats/ssize/n2.html or http://www.danielsoper.com/statcalc/calc47.aspx for comparing pairs of means using Student's t-test.