Interval estimation for treatment effects using propensity score matching