Help with power analysis!

[redskins]

So, I've never taken a statistics course before. Just completed my honors thesis and graduated. Now I'm trying to do power analysis (post hoc) on my data.

Using G power, please help me with what to do/how to interpret the results.

First of all, I looked at two independent groups (group 1, N = 9; group 2, N = 25). I found that group 1 had significantly higher testosterone levels (hormone levels ln transformed) than group 2 (independent samples t-test, p = 0.015).

So, how do I do power analysis on this variable in G power?

Go to:

Test family: t-tests
Statistical test: Means: Difference between two independent means (two groups)
Type of power analysis: Post hoc: Compute achieved power - given alpha, sample size, and effect size.

Right?

Input:

tails: 2
Effect size d: ??? <----what do i put here?
alpha error prob = .05
sample size group 1 = 9
sample size group 2 = 25


What do I put for effect size? .8? Also, how would I interpret the results, and report them?

Say power (1-beta err prob) comes out to 0.5144291, what does that actually mean, and how do I report that in my manuscript?

Thank you for the help ahead of time. Bear with me, I've never taken a stats course and am pretty bad at this math stuff.

 

[redskins]

Oh yea, please let me know if this is the right place to post this question. It's my first post, and I'd like to have it in the right place for the opportunity to get answered.

Thanks!

 

[noetsi]

What do I put for effect size? .8? Also, how would I interpret the results, and report them?

For a two sample t test the effect size should be the difference in the two means. Unless you have a specific journal requirement, APA is the gold standard how to report the results. I strongly suggest getting the sixth edition and looking there (I can send you a link that a university created how to make APA tables if that helps).

Say power (1-beta err prob) comes out to 0.5144291, what does that actually mean, and how do I report that in my manuscript?

It means that you have approximately a 51 percent chance to detect a signficant effect, if one in fact exists. Alternately it means that you have a 51 percent chance of not making a type II error. I would look, again, at either APA or a journal you are submitting this to in order to determine how to cite it (but it is normal to cite power these days).

 

[redskins]

Thank you!

Let me try this again, so:

From SPSS: group 1 mean Testosterone levels (ln transformed)= 4.5491, group 2 mean: 4.213, difference = .3361 (does this = effect size?)

So in G power: determining effect size d,

mean group 1 = 4.5491, mean group 2 = 4.2130
SD o group 1 = .2621
SD o group 2 = .47772

effect size d = 0.8723064

Then I plug that into the input parameters, i get critical t = 2.036933
Df = 32
power (1-Beta err prob) = .5856663

So I can say, there is approximately a 59% chance to detect a significant effect, if one in fact exists? Hmmm, doesn't seem very good does it, hahah. 80% is what I'm looking for generally right?

Yes please post the link to the APA stuff.

THanks again.

 

[noetsi]

difference = .3361 (does this = effect size?)

This should be the effect size.

You are right that a power that low is not very good. You usually want at least a .8 (meaning 80 percent). I don't see your N (which usually goes into Gpower calculation of power) but I would guess it's small which is why your power is low. Well your df is 32 so you have 34 cases right? If so then that is your problem. Its simply way too small for good power.

As the note says this does not follow strict APA standards in all cases. But I found it very useful.

http://people.oregonstate.edu/~acock/tables/