- #1
mirumirai
- 3
- 0
Using t-tests to get "trials until significance"?
Hi all,
I am stumped on how to achieve this with the data I have. My PI thinks there is a way, but I can't seem to find it. We want to see whether we should go ahead with this experiment or if it will take too much time.
Basically, I have 2 sets of data. I am looking to see which one conforms to a logarithmic regression line better, i.e. which one has better fit. To do that, I checked the r^2 values, and then did a t-test for unequal variance on the residual sum of squares. The difference was miniscule (p=.73).
Now, what I have at the moment is the formula for the t-test for unequal variances:
t=(xa-xb)/sqrt((sa+sb)/n)
where xa and xb are the sample means, sa and sb are the sample variances, and n is the number of trials (the actual formula uses sa/na + sb/nb, but I have the same number of data points for each set of data)
I manipulated it to show how many trials I need until t=1.96 (basically just close to significance)
n=(sa+sb)/((xa-xb)/t)^2
The issue I have with this, though, is that the means and variances could be different. How can I use this to calculate the estimated number of trials when the data I have may not be accurate? Is there even a way to do this?
Thanks for any help.
Hi all,
I am stumped on how to achieve this with the data I have. My PI thinks there is a way, but I can't seem to find it. We want to see whether we should go ahead with this experiment or if it will take too much time.
Basically, I have 2 sets of data. I am looking to see which one conforms to a logarithmic regression line better, i.e. which one has better fit. To do that, I checked the r^2 values, and then did a t-test for unequal variance on the residual sum of squares. The difference was miniscule (p=.73).
Now, what I have at the moment is the formula for the t-test for unequal variances:
t=(xa-xb)/sqrt((sa+sb)/n)
where xa and xb are the sample means, sa and sb are the sample variances, and n is the number of trials (the actual formula uses sa/na + sb/nb, but I have the same number of data points for each set of data)
I manipulated it to show how many trials I need until t=1.96 (basically just close to significance)
n=(sa+sb)/((xa-xb)/t)^2
The issue I have with this, though, is that the means and variances could be different. How can I use this to calculate the estimated number of trials when the data I have may not be accurate? Is there even a way to do this?
Thanks for any help.