But you wrote that your outcome is a yes/no variableI've been trying to use PS (independent t-test)
(on the aggregate group level: a rate), not an interval
scaled measure.
With kind regards
K.
Sorry if this is in the wrong thread. I am trying to determine if a research project is feasible to conduct by calculating the sample size needed.
I am seeing if a new protocol is more effective than an established protocol via an randomized controlled trial. Outcomes will be measured via a definition we established for a "complication" which can be a variety of things. A previous study stated that the complication rate for the established protocol is about 92%. I am trying to see how many people I would need to follow in order to see a reduction by 20% in the complication rate, so 72% complication rate. My power is going to be 80%.
I've been trying to use PS (independent t-test) with an alpha=0.05, power=0.8, delta(difference in mean)=0.2, sigma(STD)=?, m(ratio of control to experimental)= 1
but it requires me to input a standard deviation of which I don't have since I haven't measured anything and the previous study did not give a STD. Everything else I think I have except the STD.
I'm really rusty with statistics so any help would be appreciated. Sorry if this post is confusing / not clear.
But you wrote that your outcome is a yes/no variableI've been trying to use PS (independent t-test)
(on the aggregate group level: a rate), not an interval
scaled measure.
With kind regards
K.
efficientfuel (07-16-2014)
Ah thanks for pointing that out. I got caught up on the %. So i'm going to be using an uncorrected chi square test. with an independent, prospective design, using relative risk? in PS. So to calculate an approximate relative risk I made a table as seen in the pic below.
the table is labeled new protocol (the thing i want to test) against an old established protocol (control) with complication as yes or no. The old protocol has a high complication rate of 90% so i just put 90 (out of a total of 100) with the complication no for old protocol as 10 and the same for the new protocol which I estimate is 20% fewer complications than the older protocol from previous literature.
I think I did this right, but am not sure. I'm very very rusty with stats.
can anyone please let me know if this is right?
PS: I'm not sure if relative risk or P0 is correct..
So i just found out what matching means and my experiment is not matched since it will be a randomized controlled trial..
and i'm pretty sure the picture with the PS calculation is wrong as P0 should be .9
Last edited by efficientfuel; 07-16-2014 at 09:15 PM.
Tweet |