Hi everyone!
I calculated the mean activation response for some groups in my analyses. Now I would like to calculate the correlation between the mean activation response and the behavioural data I have.
My behavioural data are on a scale between 1 and 4 and I calculated the mean of each of them for the different group I want to calculate the corrrelation for. I did the correlation with MATLAb, but I only get 1 or -1 as result because I am correlating two values.
Do you know how I could do it differently so that I have more meaningful results?
The first attempt should be to calculate the correlations between individual activation responses and the respective behavioral scores for every group in your design. So you will get a correlation coefficient for each experimental group.
Depending on your experimenatl hypothesis there should be a difference between groups. Continuing statistics can be performed with SPSS for correlation testing.
But I'm afraid you should be careful with the behavioral data. To me it seems that they are distributed on an ordinal or even nominal scale, which makes statistics a little more complicated.
You need to add in your behavioral measures as covariates in the regression model / GLM you are using. Usually, you need to prepare a text file specifying the ID of each participant and the behavioral variables in columns. Then you insert the file into your group-level analysis. You also need to standardize your behavioral measures. Try converting them to Z or factor scores.
A finer model is to actually use Context-Dependent Psychophysiological Interactions (gPPI). Here is a useful reference:
Neuroimage. 2012 Jul 16;61(4):1277-86. doi: 10.1016/j.neuroimage.2012.03.068. Epub 2012 Mar 30.
A generalized form of context-dependent psychophysiological interactions (gPPI): a comparison to standard approaches.
McLaren DG1, Ries ML, Xu G, Johnson SC.
You can do this in SPM. Add a covariate like age as a vector. You go to "Specify 2nd-level" like you normally would, add the contrasts from the first level in the "Scans" entry. Then you add a "New Covariate" in the "Covariates" Section. You enter a vector that has the same amount of entries as the number of scans (one for each participant) in the same order and you also give it a name. Run it and then estimate the model. When you look at the results you have to calculate a contrast for the "Covariate", i.e., the contrast vector [0, 1], because the first column are the scans and the second column is the covariate. Keep in mind that the result is not a correlation from -1 to 1 but a paramter estimate from a linear model. But this is good enough, since a correlation is just a linear model with variables having a scaled variance [1].
[1] https://lindeloev.github.io/tests-as-linear/#3_pearson_and_spearman_correlation
- Badges
- Science topic