Can I use a pearson correlation with repeated measures?

#1
I have measured 5 different behavioural responses of birds to birdsong playbacks from a speaker (e.g. how long they took to sing back to the playback song, how long they took to approach the speaker etc..). Each bird was measured twice for each behaviour.

Now I would like to know if behavioural responses are correlated (for example, do birds that sing back quickly also approach the speaker quickly). Normally I would use a simple Pearson correlation for this, but as I have two measures for each bird I don't know if I can do that here.

So, my questions are can I use Pearson correllations given that my data is balanced (every bird has two measures for every behaviour)?

If not, is there a repeated measures alternative to the Pearson correlation?

Thanks!
 

hlsmith

Less is more. Stay pure. Stay poor.
#2
I would think that you wouldn't want to use it, since the measures are not independent. I would also think there should be some work around that I am not familiar with that is available.
 
#3
If you have SAS, then this paper will give you guidance on using a mixed model framework for estimating the correlations between two variables with replicates:

http://www2.sas.com/proceedings/sugi29/198-29.pdf

In this case, though, you wouldn't be just estimating the correlation between response A and response B, but also A at time 1 vs. A at time 2, A at time 1 vs. B at time 1, A at time 1 vs. B at time 2, and so forth...

I'm sure you've also considered just averaging the repeated responses and running Pearson's correlation on those. Sometimes, simplification isn't such a bad thing.
 
#4
Hi ab-stats, I don't have SAS but I will have a look to see if I can use the same methods in R. I have also already calculated correlation between averaged responses and found some really interesting results. However, a post-doc at my University wasn't sure if that method was valid or not which is why I'm trying to find an alternative.
 

hlsmith

Less is more. Stay pure. Stay poor.
#5
I was also thinking that a multilevel model could work, but I did not have a reference to post like ab-stat.