(Mis)estimating Affective Polarization