Group Analysis of PPI Betas

I ran a first level PPI on a cohort of subjects, and generated beta-weights for the PPI terms (one PPI term for each task condition). I then threw the PPI beta weights into a group analysis using 3dLME and got some interesting results. However, when I dig into it, I find that many of the significant effects from 3dLME seem to be driven by a few outlier subjects with wildly high or low PPI beta weights.

  1. Any advice on how to interpret beta weights of an interaction term?
  2. Should I be taking the square-root of PPI beta weights on individuals before I throw them into a group analysis?
    (My thought is that the PPI beta weights are sort of "squared" because they are fit with covariates that are formed by essentially multiplying two other covariates together. Would square-rooting the PPI beta weight help "level the playing field" across subjects? How would I handle negative values?)

Technically you can assign the square-rooted R^2 with the same sign as the regression coefficient.

It is widely acknowledged that the hemodynamic response exhibits significant variability across different regions of the brain. In addition, BOLD is only an approximation for neural activities. As a result, relying on a standard canonical impulse response curve to accurately capture subtle fluctuations at the neural level across experimental trials may be overly optimistic. In other words, it is unsurprising that the underlying rationale for the PPI analysis remains somewhat elusive.

Gang Chen