sample size for medical statistics?

#1
Hello! I hope I'm posting in the right place..

I'm doing a research in biomedicine for my bachelor's degree (testing how many embolus a machine can find compared to how many I can find = testing how good the machine is) and i'm trying to calculate which sample size I need for statistical power. I got a note from my supervisor saying "The power calculation is based on Binominal testing. We want to be sure that the machine finds >95% of all the emboli, then we need 59 emboli to be found for p<0.05."

He's telling me, in other words, that I need to have a sample size of 59 emboli for statistical power.. i've tried to trace it down a bit, and found this formula:

n0 = (z^2 *p*q) /(e^2)

from what I understand, z= the z value (1.96 for 95%), p= the percentage I picked (0.5), q is (1-0.5) and e... might be the confidence interval?

My problem is.. I don't know what "e" is, for starters, or how to get a hold of it.. and i can't find any solid proof that "59" really is the number of emboli I need.. and "because he said so" looks very bad in a report =/


Sorry about the textwall :eek: hope you can understand what i mean, i'm not a native speaker of English..
Sincerely
Hanna
 

Link

Ninja say what!?!
#2
1) I'm not familiar with that formula. Where did you find that?
2) e is not a variable. It's a number, approximately equal to 2.718 (sort of like pi)
3) You'll have to do a lot of reading to understand what statistical power is and how to calculate sample sizes. Here are some links. The third one will calculate the sample size required for you. I have not verified that it is correct at calculating. Maybe you can verify it while you're learning.

http://www.wadsworth.com/psychology...hops/stat_workshp/statpower/statpower_10.html
http://www.statsoft.com/textbook/power-analysis/
http://www.dssresearch.com/toolkit/spcalc/power.asp

Hope that helps.
 
#3
I think your formula is from that of the test statistic in the case of proportion. "e" should be the length of the maximum error you want to allow. If you want to be sure that the population value is within 1%, you'll have to take a sample of the size ~1/0.01^2 and so on.

d