From within-subject to Group-level analyses in terms of continuous relationship with power

Hi Mike,
Thanks for your thoughtful explanation about the statistical analyses on the single-trial correlations and single-trial multiple regression. However, they are performed at the individual level. I wonder how to do it at the group level? Say, I have 30 subjects and perform the correlation and/or multiple regression for each subject. Then, should the next step combine the individual results in order to do the group analyses, as the ANOVA does?

Hi Ya. The regression from each individual subject would yield a beta coefficient (the estimated regression parameter). That beta value would be zero if there is no effect, and can be positive or negative depending on the sign of the relationship. Thus, at the group level, you could do a t-test on the beta coefficients to test whether they are significantly different from zero.

If you are doing a multiple regression within-subject, then the group-level analysis might need to be adjusted, e.g., an ANOVA or correcting for multiple comparisons of multiple betas, etc. But the general principle is the same: Fit the regression model within-subject, then test the regression parameters against zero at the group level.

Thanks, Mike. It’s clear to me.