Validating models of opinion polarization on the micro-level