Using a spike model to deal with zero response data from double bounded dichotomous choice contingent valuation surveys