Sample size calculator

p.pete

Does anyone have any recommendations for a site that explains how to do a sample size calculation? I've done a study with 16 participants, and my mixed anova's not significant, but if I simply double the data it does pop out significant.
Cheers!

opinion guy

I'm not really sure what you are trying to do here ? Sample size calculations are normally done before you do a study. Do you want to calculate the power of your study is that it ?

What do you mean by "simply double the data" ? Enter each result twice ? That would not be valid - you can't draw any conclusions from stats based on a double entry of the values - and a p value on such data would be meaningless because you have artificially removed some variability.

I haven't used any sample size calculators but this one looks ok:
http://www.nss.gov.au/nss/home.nsf/pages/Sample+Size+Calculator+Description?OpenDocument
That page explains how it works - there is a link at the bottom to the actual calculator

p.pete

I've already done a study, and it wasn't significant.
Yes, I just doubled the date, and damn sure I'm not using that to draw any conclusions! I am trying to calculate, if I ran a similar study, how many participants I would need (sample size) for it to be a significant study. I've had a look through a few stats pages but haven't spotted anything that can reduce things to my level of understanding!

Seems to be some nice explanations in your link so I'll work through that for starters, thanks!

opinion guy

p.pete wrote: »

Yes, I just doubled the date, and damn sure I'm not using that to draw any conclusions! I am trying to calculate, if I ran a similar study, how many participants I would need (sample size) for it to be a significant study.

Ah right well double entry of the data would not be useful for that either. Bascially the data is not random cause it repeats itself. I think - it would give be biased to give you a significant p value when there isn't one

Yeah sample size calculations are tricky. Hopefully that page helps! You have to enter in the confidence level, the population level and the proportion of people you think have the outcome first. Then you have to enter one of confidence interval, standard error, relative standard error or sample size and it will give you the rest

Ostrom

Non-significant results are important too!

http://www.badscience.net/2011/04/i-foresee-that-nobody-will-do-anything-about-this-problem/

2Scoops

Non-significant results are important, but can be highly misleading to the less statistically astute reader if the study is underpowered to detect a legitimate difference. OP, I can't resist a wagging my finger at you! :pac: A power analysis should have been done with preliminary data or estimated from other papers. Where did n=16 come from?

Anyway, for future reference, G*Power or PS from Vanderbilt can be used for quick power calculations.

Ostrom

Might also have violated some assumptions with low n. Any non-parametric alternative?

S_S116

I was given a really good spead sheet by my stats prof to calculate the power of a study with a given sample size. It's tpp big to attach though. I could leave it somewhere on the web for you though or e-mail it.