Hi,
So I have been working hard on a project. However, I am a little stuck. For the question, " How well did students’ estimates of ‘how many smiles they correctly identified’ relate to their actual performance?", I wanted to use a correlation, but a group I was working with wanted to use a t test. However, the t test compares two means between two groups? This is one group of students that had to guess whether or not smiles were correct, and then estimate how well they did. This seems purely correlational to me, ie: the higher the estimate of success, the higher the performance
Am I crazy?
Did you ask them how they plan to run the t-test and what they think it will tell them? I don't really see how it would relate to the question at hand. Unless they are planning on doing a regression and running a t-test on the slope of the regression to see if it's different from 0. But it doesn't sound like that's the route they were planning.
I ran it through SPSS but then it just give me two t's. I am no expert but I feel like this is not right, this did just this as well. I know that sometimes more than one test can be used, but a correlation is suffice here?
Originally Posted by Dason
A test for a significant correlation and a test for a significant slope in a regression will always come out with the same p-value so it doesn't matter how you test that.
I actually have two different p values.
When I ran this info through a t test my p value was 0, for the correlation it was .268
What are you doing a t-test on though? I was talking about the t-test associated with the slope of a regression line. A normal t-test in this situation doesn't make any sense. That's why I suggested asking your group members why they thought a t-test would work. They probably can't explain it other than 'it's what we always do'.
man stats is hard, I am just happy I figured out this one error
Ha. It's not so bad. It's a little interesting because there are usually a lot of reasonable ways to go about doing an analysis. But everybody has their own opinion on what the 'best' way to analyze certain data is.
The important thing is to try to justify any assumptions that you make. If you do this then you should be fine. Don't let the details bog you down too much though. I know I hated my first stats course because it just seemed like a big mess of unrelated tests. There was a z-test, t-test, regression, sign test, normal distributions, chisquare distributions, correlation, means, variances (and the professor wasn't that great either...). And I didn't really get much motivation in that class. But then I took another class and realized that all we're trying to do it find a reasonable way to learn something from data I actually started to like it. The different types of tests just correspond to different assumptions and different types of data we could be working with. That's not the important part.
Posting Permissions
|
|
Reply With Quote
