An ontological test of the IAT: Self-activation can increase predictive validity