That assumes you interpret not happening as being a separate third category, but for these purposes it seems more reasonable to consider it as always happening after (i.e. happening at time infinity), since we want lower probability of it happening soon to cause the median date to increase.