Again, I ask that you look at this from the perspective of an actual participant in the survey. That person is imagining grabbing a random person and tossing them off a bridge on very short notice. Numerous factors come into play, and the participant is going to consider them whether or not you assure them that they don't matter.

Another reason the trolley problem is bogus is that if you were really in such a situation, you wouldn't be sure your attempt to push the guy onto the track would even succeed. What if he saw you coming and resisted? Pushing a lever with 100% of success is different from pushing a guy with 87% estimated success and consequences if you fail.