How accurate are dichotomous choice contingent valuation welfare measures when agents have heterogeneous preferences?