Untitled Post
jcrofts
Posted on 09/25/08 05:12:42
Number of posts: 29
j crofts
Hi, I was at a conference recently, and a colleague told me that he believed that I should not be using PLS on my data to make inferences about functional connectivity since it has no temporal component. I am not a neuroscientist and was hoping you might be able to tell me if this is correct. Also, he claimed that one should only use PLS if they have at least 10 samples per latent variable (I know this is the case in factor analysis), is this true?
Cheers,
Jonathan
Untitled Post
rmcintosh
Posted on 09/25/08 09:09:42
Number of posts: 394
In its original definition, functional connectivity referred to temporal correlations since it was introduced for single cell recordings. As it was adopted into neuroimaging, the temporal part has become less relevant and the term has been used to refer to any correlation involving brain data. There is a rich literature, including a book "Handbook on Brain Connectivity" that Viktor Jirsa and I edited, that may help to sort this out. There is a very nice paper by Barry Horwitz that is worth a look: Horwitz, B. 2003 The elusive concept of brain connectivity. NeuroImage 19, 466–470.
As for sample size and PLS, the statement of 10 samples per latent variable (I think they mean 10 samples per dependent measure), this is really only a requirement for parametric statistics, which are not used in PLS. It is true that a large sample size will always improve generalizability and stability of the results, but if you try get 10 samples for dependent measure, especially in brain imaging data, you would never get to analyze your data.