Multivariate equivalent of a matched pairs t-test?

#1
Hi there,

I have a multivariate data set, and I have performed the main analyses on it and am happy with the results.

Now, I took duplicate measurements for half of my samples, and what I want to do is determine how repeatable/accurate my measurements were - i.e. how similar the two duplicate measurements are, but only comparing the two measurements from a single sample with each other, and then looking at how similar the duplicate measurements were to the original ones across all the samples, if that makes sense.

If it was univariate data, I'd perform a matched pairs t test. But as it is multivariate (with LOTS of variables), I'm not sure what to do, can anyone help?

I've been using Primer-E for the rest of the analyses so if there's a routine I can use in that it would be easiest!

Thanks for any help.
Alex
 

bugman

Super Moderator
#2
We do a similar thing for QC/QA purposes and simply calculate bray curtis distances between samples. The requirement is >95% for this purpose.

Do you have the Permanova add-on?
 
#3
Hi Bugman,

Thanks for the reply.

I too was beginning to think of simply calculating the bray curtis resemblance matrix and then looking at the similarities between the pairs, and maybe coming up with and average similarity %. What do you think?

Yes I have PERMANOVA add on - is there something more I could do using this?

Cheers
Alex
 

bugman

Super Moderator
#4
It was really just an after thought - thinking about using permdisp betweent the two samples.
I would be pretty confident just using the BC distance in this case though.