help > RE: Possible to re-compute contrasts without re-estimating the entire model??
Feb 24, 2019  03:02 AM | Phil Kuhnke
RE: Possible to re-compute contrasts without re-estimating the entire model??
Hi Arianna,

your con-images for Faces > Shapes from the first-level PPI should "include" the negative correlations as they reflect the difference between the parameter estimates (betas) for Faces minus Shapes - a value in a con-image can be negative. Thus, at the group level, to get Faces < Shapes (or equivalently, Shapes > Faces) you can simply set '-1' as the contrast weight when estimating a one-sample t-test across your Faces > Shapes con-images (assuming this is what you want to do).

If, for some reason, you want the subject-level con-images for Shapes > Faces, you could multiply the Faces > Shapes con-images by -1 (e.g. using SPM's ImCalc).

Maybe someone else can confirm whether my reasoning here is correct.

Hope this helps!

All the best,
Phil

Threaded View

TitleAuthorDate
Arianna Gard Feb 14, 2019
RE: Possible to re-compute contrasts without re-estimating the entire model??
Phil Kuhnke Feb 24, 2019
Arianna Gard Feb 27, 2019
Donald McLaren Feb 24, 2019