Towards a better understanding of when to apply propensity scoring: a comparison with conventional regression in ethnic disparities research