permutation tests
jcrofts
Posted on 08/18/08 04:15:39
Number of posts: 29
j crofts
Hi, my statistics is not very good and I am struggling to see exactly what the permutation test is doing. I realise that by reordering the rows we are testing whether the particular singular values are likely to be the result of chance. But what I don't understand is why we conclude that the number of times a particular singular value is exceeded is a good measure of this, and why being much smaller does not. It seems to me that either way the singular value could be markedly different. I'm sure this is down to my complete lack of understanding but some sort of explanation would really help me. Thanks, Jonathan
Untitled Post
rmcintosh
Posted on 08/18/08 08:03:04
Number of posts: 394
A fair question. The singular value is a squared quantity (actually square root of the eigenvalue) so lower values get you closer to zero but not below. The singular value is also the covariance between the two matrices that you are analysing, so a low value indicates a weak (trivial) association.
Now this does take us into the sticky issue of whether null hypothesis testing is really what we need to be doing - an argument that is bound to get a number of very heated responses. This is why we also have the bootstrap which is a means of reliability assessment.