Distribution-free estimation with interval-censored contingent valuation data: troubles with Turnbull?