 |
 |
Hi Rafael, Performing group statistics on correlations rather than beta weights has been done quite often, most commonly in studies of connectivity, but also when using an amplitude-modulated regressor which is based on some external variable such as reaction time, physiology etc. In both these cases, techniques such as PPI might be more appropriate within SPM. There's nothing specifically wrong to do group analyses on correlation coefficients, though they should usually be z-transformed to make the statistics valid. Having said that, you need to be clear about what it is that you're testing. While betas give an indication of magnitude of signal change, correlations depend both on magnitude of signal change as well as unexplained variance. The latter could be what drives any effects you find, and the source of variance might not be easily identifiable or might have nothing directly to do with brain activity (e.g. it could be due to all sorts of artefacts). -Tom Johnstone University of Reading 2010/4/19 Rafael Lüchinger <[log in to unmask]>: > Dear statisticians > > In classical SPM fMRI analysis beta wights are used for interpreting signal strength in the data. Regressors recieve a wigth so that the model as a whole describe the date best. > > Now we are interested in the correlation between a regressor and the data. This poses the question for statistical "tightness" between regressors and the data and not the relative weigth of a regressor in a model. We further want to compare the regressor correlation between groups, asking if they differ in terms of regressor-BOLD coupling strength. > > To recieve the multiple correlation from a multiple regression (GLM) is mathematically very easy. To my knowledge to automatically recieve correlation values for specific regressors is not implemented in SPM. > > Why are beta wheights but not regreessor correlation of interest in general? Has someone applied fMRI analysis on correlation corefficients? Is there a reason why such an analysis would not make sense? > > Thank you for any comments > > Rafael >