Help with Correlation

#1
Hello! I need help for an issue i have.

Let's assume that we have data of two variables (X and Y) for 14 different people.
We can find the correlation between those two variables, for each one. In my case i find moderate correlation in some of them (which is also stat. significant), and for some other i don't (and it appears to be statistical insignificant).

Is there a way to extract a result about the correlation for the general population using something as multiple correlation or something like that?
 

Dason

Ambassador to the humans
#2
Are you saying you have more than two variables but you were only looking at the pairwise correlations between those two variables?
 

hlsmith

Omega Contributor
#5
Yes, your posts are very vague.


I think they have two variables, 14 subjects, with 100 repeat measurements per subject, correct?
 
#6
Yes that's what i mean..
Ok i will give you an example..

I have 14 people and for each one i have two variables AREA and duration (this is about fixation points while reading). For each person i have from 70-150 points (pairwise).

I cannot extract a median or mean value for each one, because each point is not recorded in the same condition.

I can run a correlation test for each one separately, but the purpose is to find a result for the general population.
 

hlsmith

Omega Contributor
#7
"I cannot extract a median or mean value for each one, because each point is not recorded in the same condition."




I don't understand this sentence:


each one = ?


same condition = ?


In addition, each person has a different number of repeated measures recorded? And the timing between these repeated measurements is not the same between subjects?
 
#8
Yes, each person has different number of measures recorded. The timing between those measuremenets is different.

For the first part of your question, I mean that i could extract a measure of central tendency for each person, to represent those two variables, (median because they are non-normal distributed), and with 14 points run a correlation test. But 14 people are a small sample size to do it, and the results are not stat. significant.

That's the reason i am trying to find a way to use all the data and extract a result for all of them.
 

hlsmith

Omega Contributor
#11
You can probably put all data into a multilevel model for repeated measures. You may need to investigate the covariance structure to deal with the varying time intervals. This isn't a novice procedure so research will need to be conducted on your end. This approach will give you association estimates