If the experiment is opt-in, there will surely be a bias in the sample population. In this case, the strategy will select risk-seeking individuals, because risk-averse ones will not want to participate.

Perhaps it should be required that the net income from this lottery be 0.

That would be nice, but then you're not doing an experiment.

Why not? You can still observe how people behave differently if they have higher or lower taxes.

I guess that's true. But what I meant to say was that in order to arrange for the results to net a zero difference from what the government would collect normally, you're not really randomly assigning tax rates or running a lottery, you're just fixing people's rates at something other than what they would normally be.

Something about manipulating the conditions of the experiment (even a stupid thought experiment) to force a desired outcome doesn't sit right with me.

He means the higher and lower taxes would cancel themself out.

Not that the people in it earn nothing.