We constructed LMMs to regress startle response on stimulus valence, session, and group. We found that the model with random slope effects for both stimulus valence and session significantly improved the model's fit over the model with a random slope effect for stimulus valence only (Δχ2 = 32.9, p < .0001) and the model in which all predictors were treated as fixed effects (Δχ2 = 63.9, p < .0001).
