<div>Utility starts from the assumption that the purpose of elections is to make the right decision. If this is true, then an election system should be measured by its outcome: what is the quality of decisions that it helps to make? As WDS has convincingly argued, this kind of analysis leads inevitably to measuring voting systems by their utility, summed or averaged over the voters. Since the voters' own views - while perhaps mistaken - are the only unbiased measure of that utility; and yet it tends to be impractical to get a broad and honest measure of those views; the best way to evaluate voting systems for their utility is to use monte-carlo computer simulations.</div>
<div><br></div><div>However, that still leaves several problems. Simulations require a model for the electorate, both in terms of utility and in terms of strategy. While a good system will probably function well with a variety of utility models, the strategy model is always going to be a debatable factor that affects results. Thus, though I personally think that utility simulations are a crucial part of voting systems research, and I am grateful to the work of pioneers - particularly Warren Smith - I am not satisfied with the strategy models or related results of any simulations so far.</div>
<div><br></div><div>The other objection is more philosophical: what if you don't believe in commensurable utilities? For instance, I've heard that some people are "super tasters", 10 times as sensitive as average to fine distinctions in flavor. If that were true, would such people deserve all the best food, as it makes a bigger difference to their utility than it does to others? While I understand this objection, I can't really give it too much weight; normalizing utilities might be a philosophically- and even mathematically- hard-to-justify operation, it's simple and it works well enough for me.</div>
<div><br></div><div>(In this discussion, I use the term "utility" to refer exclusively to the utility of the outcome - the winner. I understand that, properly speaking, the other values could clearly be seen to have their own utility; but to use the term in that way would confuse the issues.)</div>