> In group analysis should I apply 3dttest with –base1 0 and –set2
> the list of sub-briks of all subjects corresponding to the beta coef of
> the second regressor (but it is not PSC!)?

Seems reasonable.

> If the result is not statistically significant does it mean that BOLD
> response is not modulated by emotional valence?

At lease you can say the modulating effect is not statistically significant.

> Is there a way to compute PSC and use it to run group analysis?
> Does it make any difference compared to the approach described
> above?

I don't see any point converting the valence modulation coefficient to percent signal change. In other words, this is not an issue about percent signal change, instead I would be more concerned whether the valence across subjects are comparable. If not, maybe some sort of valence normalization is desirable as I mentioned last time.

Gang