With two candidates, strategic range voting is the same as single preference: everybody is going to vote 0 and 10 for the two candidates (assuming a 0-10 range). Anything else de-emphasizes their preference.
With multiple candidates, the "ideal way" to vote strategically depends, as I've said before, on your estimate of how everyone else will vote. Consider: if you know exactly what all the other votes are, you know if the vote is close enough for you to affect it, and if so, how. This might mean voting 10 for your second-favorite candidate and 0 for your third-favorite, even though your actual preference between those two is extremely weak (your actual preferences being, say, 10, 8.1, 8.09, 5, 3, 2, 1, 0.5, 0).
It's easy to show that voting 10 for your favorites, and 0 for the rest, is optimal for some value of . But you don't know which value.
So the simulations used a strategy that depended on voters knowing in advance the approximate results; what happens when those assumptions are erroneous?
The simulation results also didn't describe the distribution of preferences; if they were independent random variables, they don't resemble real preferences well enough to be useful.
As for bad examples, I can construct some (based on voters guessing wrong about other voters) where the winner is not the majority or Condorcet winner, nor produces the greatest social utility. In fact, he can be in the bottom half of the pack.
Re: Disaster?
With multiple candidates, the "ideal way" to vote strategically depends, as I've said before, on your estimate of how everyone else will vote. Consider: if you know exactly what all the other votes are, you know if the vote is close enough for you to affect it, and if so, how. This might mean voting 10 for your second-favorite candidate and 0 for your third-favorite, even though your actual preference between those two is extremely weak (your actual preferences being, say, 10, 8.1, 8.09, 5, 3, 2, 1, 0.5, 0).
It's easy to show that voting 10 for your favorites, and 0 for the rest, is optimal for some value of . But you don't know which value.
So the simulations used a strategy that depended on voters knowing in advance the approximate results; what happens when those assumptions are erroneous?
The simulation results also didn't describe the distribution of preferences; if they were independent random variables, they don't resemble real preferences well enough to be useful.
As for bad examples, I can construct some (based on voters guessing wrong about other voters) where the winner is not the majority or Condorcet winner, nor produces the greatest social utility. In fact, he can be in the bottom half of the pack.