See below. On Sun, Apr 27, 2014 at 9:48 PM, Andy Yeung <[log in to unmask]> wrote: > Dear all, > > I saw Buchel et al (Neuroimage 1998) explaining the use of parametric > modulation (PM) to show regionally specific word rate-dependent > timecourses. Is PM by nature a regression analysis showing regions of BOLD > signal having a linear relationship with some covariates/regressors? > Yes. > > If so, will the meaning and result be different if I employ different > models for 2nd level, eg 1 sample t-test with covariate or multiple > regression with covariate? > If the design matrix is the same, then you will get the same result. You can get the same design with a one-sample t-test with a covariate and multiple regression (which by definition has a covariate). SPM uses the GLM, so if the designs and variance settings are the same, both modules will lead to the same result. > Someone says multiple regression uses f test, does it mean we need further > 1 sample t-test to test for direction? If so, why not start with 1 sample > t-test? > There are two things here: (1) Your model - both the one sample t-test and multiple regression can create the same GLM (2) Contrasts - these are either F-tests or T-tests. Both contrast approaches can be used with either model (one sample or multiple regression. F-tests are non-directional. If you don't have a predicted direction of the effect, then you should use the F-test. If you are testing the slope is greater than/less than 0 or another slope, then you can use T-test. If the F-test contrast is a vector, then the F-test with be the T-test squared. (F=T^2). The significance of the F-test will be half the T-test (e.g. if T=0.01, then F=0.02). Hope this helps. > > Thanks. > > Best, > Andy >