There is something slightly off about sentences like this one, from the New York Times:
At a minimum, it is clear that Republican voters, after delivering three different winners in the first three stops in the nominating contest, are in no rush to settle on their nominee.
You see this kind of coverage a lot in primary season: commentators make sweeping statements about the electorate’s desires based on polling or primary numbers. The problem is that it hopelessly conflates individual and collective preferences. In this sentence alone, there are two such slippages. First, clear votes for different candidates are smushed together into a collective indecision. And then, that indecision is attributed back to the individual voters. But the fact that Republican voters in three states disagree about which candidate they would like to nominate does not tell us that they are happy to make the choice slowly. It could just be that their primary process is set up in a way that doesn’t lead to an unambiguous early winner, given this year’s candidates and political climate.
The mild version of this mistake comes up in two-party elections or in polling on an issue that has only two choices. To win an election by 10 votes out of 1,000,000 cast does not mean that the electorate collectively has spoken and pointed to you. (Indeed, with many voting technologies, this difference would be well within any reasonable margin of error, so we couldn’t even be sure that the “winning” candidate got more votes.) All it means is that the election process resulted in selecting one candidate over the other, which, by the rules of our system, means giving the winner the job. A “close” election could reflect a bitter partisan divide, or vast collective indifference as between two decent choices. The number alone says little. The same goes for margins of victory: even a 65-35 victory in the popular vote — an unprecedented margin for a presidential election in U.S. history — need not mean uniform national agreement on anything. The many millions of voters on the short side of the count need not be acquiescent, just outnumbered.
In multi-way contests — like early primaries — the gap between individual and collective preferences is even more severe. Indeed, thanks to Arrow’s impossibility theorem, there may be no coherent way to combine individuals’ choices at all. Every voting (or polling) system will fail in one way or another.
This year’s Republican primaries, in particular, seem to be suffering from a severe independence of irrelevant alternatives problem. The media has settled on a narrative in which the essential choice for Republican voters is between Romney and Not-Romney. But that choice is never on the ballot, and it seems never to be presented in polls. Instead, they’re asked to cast a single vote for one candidate, with a wide array of would-be Not-Romneys to choose among.
Under these circumstances, we simply do not know what the statistical preference of Republican voters looks like. It could be that as the various Not-Romneys drop out, their support will largely break for Romney, or for the other Not-Romneys. If the former, then Romney is in fact broadly preferred by the Republican electorate; if the latter, then he is in fact broadly rejected it. By the first-past-the-post standards of political reporting, these are two very different outcomes. But by those same first-past-the-post standards, Santorum “won” Iowa, Romney “won” New Hampshire, and Gingrich “won” South Carolina and we’re no closer to having a meaningful picture of overall individual preferences.
The secret, of course, is that reflecting individual preferences isn’t necessarily the point of a voting process. Simple coordination — telling the party stalwarts whom to believe in and coalesce around — is frequently valued in a primary. So is awarding delegates in a way that permits a relatively small group of party insiders to push a favored candidate (remember the superdelegates?). And at the general election, the media narrative of a decisive “choice” almost certainly helps promote social cohesion and effective governance, even if the narrative itself is a lie.
In any event, here’s my suggestion for how to perform more illuminating primary polling. Don’t just ask people who they’ll vote for. Instead, ask them to rank-order candidates from most preferred to least. Or ask them whether they consider each individual candidate “acceptable” or “not acceptable” as a possible nominee. No poll is perfect, but I think this sort would be substantially less misleading, because it would focus attention on what actual voters’ desires, not the fictional desires of a mythical collective hypothesized from raw vote totals.