Strategic voting is (usually) socially beneficial

The existence of strategic voting within a voting method is often viewed purely as a downside for that voting method. “Strategic” voters are often compared to “honest” voters, wrongly implying that strategic voting is necessarily dishonest — and also implying that strategy is bad since we all know that honesty is good. Various papers talk about some sort of “resistance to strategy” metric, taking it for granted that strategy is an insidious threat that must be resisted. Overlooked is the fact that strategic voting often yields better outcomes for the electorate as a whole, and not just from the perspective of the strategic voters.

Fewer wasted votes

A simple example: In Plurality, strategic voting limits the spoiler effect. Suppose a Democrat and a Republican are each the first choice of 45% of voters, and a Green and a Libertarian are each the first choice of 5%. When the Green and Libertarian candidates’ supporters decide to “dishonestly” vote for one of the majority party candidates, it yields an outcome that better reflects the preferences of the electorate as a whole. If everyone voted honestly, the preferences of 10% of the electorate would be essentially ignored. Strategic voting means that more people’s voices are heard, and this yields a more representative outcome.

In general, the main reason strategic voting is socially beneficial is that strategic voters don’t throw their votes away. As in Plurality, ignoring the preferences of voters tends to yield worse outcomes. Wasted votes are bad for society, not just for individual voters.

In Approval Voting, there are two ways of throwing your vote away: voting for none of the viable candidates and voting for all of them. Strategic voters will vote for some but not all of the viable candidates.

IRV also offers two main ways of throwing your vote away. If the ballot only gives you room to rank three candidates, but there are ten in the race, you can throw away your vote if you rank longshot candidates you love instead of viable candidates you tolerate. This is like falling victim to the spoiler effect in Plurality. An alternative scenario occurs when there’s a candidate with a good chance of advancing to the final round but essentially no chance of winning the final round. (Imagine, for example, a Democrat in a district that is 40% Democratic and 60% Republican.) Voting for this candidate will waste your ballot, since doing so denies you a say in which candidate will defeat the no-chance candidate in the final round. By not wasting valuable ranking on candidates who can’t win, strategic voters can avoid these pitfalls.

Other benefits of strategic voting

Strategic voting also mitigates the weaknesses of voting methods in other ways. One weakness of Approval Voting is mediocre expressiveness; when you follow the crude guideline of “vote for all the candidates you like,” your ballot will never distinguish between the candidates you love and the candidates you think are just pretty good. Strategic voting changes this. Consider two scenarios with three viable candidates:

1. One candidate is bad, another is okay, and a third is fairly good.
2. One candidate is bad, another is okay, and a third is outstanding.

Strategic voters will be less inclined to vote for the okay candidate in the second scenario than in the first one, thereby improving the chances of the preferred third candidate. Thus, strategic voting makes Approval Voting responsive to the intensities of voter preference which extend beyond whether or not a given candidate is “approved of.”

Similarly, one weakness of IRV is that it provides poor third party visibility — voters who prefer a third-party candidate to anyone else will have this support show up in the election results, but the support of voters who slightly prefer a majority party candidate to the third-party candidate will remain invisible. (Lower rankings are only looked at when the higher-ranked candidates are eliminated, which won’t happen for major party candidates.) A voter who slightly prefers a Democrat to a Libertarian but wants to see the Libertarian party grow could simply vote for both the Democrat and the Libertarian under Approval Voting. Under IRV, this voter could achieve a similar result by ranking the Libertarian above the Democrat; the Libertarian will see the additional support in the first round, get eliminated, and then the vote will be transferred to the preferred Democrat.

When strategic voting is socially harmful

There are several circumstances in which we might say that the harms of strategic voting outweigh the social benefits.

Strategic voting involves risking an outcome which most voters consider to be terrible. This is most pronounced in Borda Count. Suppose there are three candidates. Two are comparably popular, and you slightly prefer one to the other. There’s also a dark horse candidate who 90% of voters (yourself included) absolutely despise. If you dishonestly rank the dark horse between the two popular candidates you will double the support you give to your favorite as compared to your less-preferred of the viable candidates — but if enough voters do this, the dark horse might actually win. There can be a similar, but rarer and milder, issue in approval voting.

You think a voting method is virtually perfect when everyone votes honestly. There are strong arguments in this vein for Score Voting and Condorcet methods, and plausible ones for STAR voting and Borda Count. It’s impossible to improve upon perfection and strategic voting can change outcomes under any voting method, so obviously strategic voting must make things worse. That said, this does not mean that strategic voting makes things significantly worse. The effects of strategic voting must be assessed quantitatively, not in the simplistic binary of “Is this bad or good?”

You think maximizing group utility is more important than electing Condorcet winners. Suppose there’s a Right-leaning district and three candidates: Right, Center, and Left.

• 55% of voters assign 10 utility to Right winning, 8 utility to Center, and 0 utility to Left.
• 25% of voters assign 10 utility to Center winning and 0 utility to Right and Left.
• 20% of voters assign 10 utility to Left, 8 to Center, and 0 to Right.

Since Right is the first choice of a majority of voters, Right is the Condorcet winner. However, electing Center will maximize group utility (8.5 average utility for Center vs. 5.5 for Right). Under Approval Voting, strategic voters will never vote for both Right and Center since it’s obvious that Left has no chance, but non-strategic voters who prefer Right could vote for both. Right will always win under Plurality, IRV, STAR, and Condorcet methods, but center wins when there are sufficiently few strategic voters under Approval and Score. This is, however, a very contrived example; the Condorcet winner is the utility-maximizing winner the vast majority of the time, and when voting methods fail to elect a Condorcet winner it’s usually not because they’ve found someone better.

Strategic voting makes some parties appear less viable than they actually are. This occurs with third parties under Plurality. It can also occur with IRV in the case where one party is large enough to advance a candidate to the final round but not large enough to win in the final round. Either way, it’s strategically preferable to vote for a candidate who can actually win, so a widespread belief that a certain party is non-viable can become a self-fulfilling prophecy.

Comparing harms and benefits

A quantitative assessment of the social harms and benefits of individuals voting strategically is faced with numerous difficulties:

• Examining a single election or a small hand-crafted set of elections won’t suffice. Results must be averaged across a large set of realistic elections, and this requires computer simulations.
• Strategy must be assessed in a manner that quantifies its effects on society. Merely determining whether a certain pattern will yield a different winner or whether some subset of voters could change the outcome by voting strategically isn’t enough.
• Strategy cannot be assessed in a purely one-sided manner; the effect of strategic voting across the electorate must be analyzed.
• You have to ensure that the “strategic” voters are actually voting in a manner that does a good job of advancing their interests in the circumstances under consideration.
• The model must take into account realistic levels of uncertainty; it can’t be based on voters having perfect information.

I know of no model which does a good job across all of these criteria. Voter Satisfaction Efficiency comes the closest, but the current version uses an unrealistic and ineffective strategy for IRV which renders some of the results meaningless and does not incorporate uncertainty into its strategy models. (An upcoming version will address these issues, but correctly factoring in the realistic strategic voting by multiple factions remains a hard problem.) It does show the benefits of strategic voting for Approval and Plurality, however.

We can make a few broad points in the absence of a reliable model:

• We should expect relatively little harm stemming directly from applications of one-sided strategy; while one faction may have more strategic voters in it than another party, thought experiments in which everyone in one faction behaves strategically but other factions do not are unrealistic.
• Instances in which strategic voting yields a bad outcome in the eyes of most of the strategic voters must be infrequent; if your “strategic” voting repeatedly yields a bad outcome, it is not strategically optimal to keep voting like that. It can be strategically optimal to risk disaster, but not to accept disaster every time.
• The social benefits of strategic voting need not be similarly limited. The idea of strategic voters always shooting themselves in the foot is self-contradictory, but the idea of non-strategic voters always getting bad outcomes due to vote splitting is entirely consistent.

How voting methods compare

Plurality, Approval, and IRV all benefit from strategic voting, Borda Count is massively harmed by it, and Condorcet methods are (probably) slightly harmed by it. STAR and Score are more ambiguous. For the most part, voting strategically means you won’t have less influence than other voters — but if we’re trying to maximize group utility then we want voters with weak opinions to have less influence than voters with strong opinions.

A general rule of thumb: The more issues a voting method that are unrelated to strategy, the more that voting method stands to benefit from strategic voting.