Coordination failure

Here is a simple explanation of what I mean by “coordination failure.” If you are reading this, you may have heard the term come up in discussions of certain electoral reforms. Here is some of I what I’ve written about the problem. I will give a simpler explanation below.

Two repeal cases have been subject to ballot-level analysis, and results are consistent with a predictability story. In Pierce County, Alvarez et al. (2018) found a two-dimensional structure in rankings from the CFB [come-from-behind] race. One dimension was partisanship, and they term the other “preference for independence.” Because there was only one independent in the race, the orthogonal dimension suggests a candidate whose ballots did not flow reliably to either side. Similarly, in Burlington, AV critics have long argued that “the system failed” to elect a Condorcet winner. This suggests lack of coordination along the dimension that was salient to most voters (cf. Nagel 2006). Hence critics conclude that Burlington’s repeal was due to “the surprise outcome of the election.”

Let’s say we are asking voters to rank, approve of, or assign scores to candidates. Let’s assume that elites/activists are not urging bullet voting. And let’s say those voters populate a two-dimensional space that looks like this. It represents the 2016 electorate.

Santucci and Dyck (2022)

The first dimension captures liberal-conservative policy views. I prefer to think of it as a major-party dimension. Note how Trump (red R) and Clinton voters (blue D) appear on the right and left, respectively.

The second most strongly captures measures of political discontent. Charles Franklin recently called this dimension (in Congress) institutional/anti-institutional. I like that.

There is no problem as long as candidates differentiate themselves along one or the other dimension. Voters can (in theory) rank/score candidates according to how far away they perceive those candidates to be. (I am less confident about this under an approval ballot, as that ballot cannot capture preference intensity.)

The problem arises when both dimensions are active at once — as in the figure. What does an anti-institutional liberal do when the only anti-institutional candidate is on the right side of the major-party dimension? What does an institutional conservative do when the only institutional candidate is on the left side of the major-party dimension?

This is where we could get into scenarios, questions about distance, questions about not voting, questions about information quality, the idea of cross-pressure, and so on.

The simple answer to both questions above is “it depends.” That’s the point.


Further reading:

What a journalist might watch for in different kinds of RCV elections

I recently wrote a post for 3streams on ten types on RCV. What sorts of campaigns might they engender?

The working assumption here is that two sides will emerge in politics. They may not track party lines, but they should be identifiable if one looks at the right data in a context where the system is ‘settled.’

To summarize the 3streams post, these systems can be used to try to get any of three things: single majority winners, majority-slate sweeps, or both sides represented in a multi-seat district. It is a bit more complicated than that, but this serves to introduce the ideas that follow.

“Anybody but X” campaigning and/or electioneering. Look for this in STV and maybe AV/IRV when they are working ‘well.’ Laver (2000) pointed to this logic in a chapter on government formation.

Slate formation. Look for this in any of the multi-seat systems. Savvy candidates would want to benefit from their “vote pooling” properties. Maja Harris has been following this in Portland (OR).

Spread-the-preferences (STP) strategies. This term is from the comparative literature. It means optimizing two imperatives: run a number of candidates that can win, and ensure a more-or-less even distribution of their first-choice votes. Neighborhoods are therefore good for recruiting candidates, doing GOTV, and possibly targeting policy benefits. Watch for STP in STV and bottom-up. I need to think more about BPV. I don’t see it as a factor with numbered-post.

Fragmentation in primaries. I need to think more about this. One, running for a nomination often is not the same as running to win power. Two, that is what our best source on the history reports. Three, most papers on “exhaustion” (IIRC) point to non-majority winners emerging from fragmented fields. Four, the game theory I have seen suggests that candidates appealing to the same group of voters may not have an incentive to encourage ranking (among other issues).

Nonpartisan primaries (so-called). I have no clear prediction beyond “anybody but X.” And yet there is the game theory I just mentioned.

I may update this later. Thanks to HB for the suggestion.

What do we know about historic use of IRV in nominating primaries?

Alan Ware (2002, p. 231) writes the following. His book covered the politics of nominating primaries. Its key contribution was to argue that major-party leaders imposed direct primaries to help avoid party splits.

Maryland and Minnesota (both in 1912) had started to use the Alternative Vote electoral system for their primaries. The other possible solution for ensuring majority nominations – the run-off election – was deployed in six southern states, five of which had adopted it before 1917. Indeed, at various times other states had also used forms of so-called preferential voting to ensure that nominees were not the choice of merely a small minority.10 The solutions to the problem of vote fragmentation were well known. The far more intractable problems posed by direct primaries for the parties were, first, interest aggregation, because nomination decisions were now individualized, rather than being considered as part of a “package.” This increased the likelihood of intraparty tensions, and also the risk that unelectable tickets might result. The second problem was that the direct primary both reduced the ability of the party to control the “quality” of candidates selected, and in some cases made it more possible that wealthy individuals might triumph over poorer rivals having broader support among party activists.

Weeks (1937) gives the fullest account of these systems. Whether voters will use rankings sufficient to “guarantee” majority winners is an issue. There also is an open question in the literature: do single-seat reforms (like IRV) eventually induce coordination failure (so that voters do not use markings to get majority-supported outcomes)? Here is what Weeks says about these issues:

1. All states except Alabama and Oklahoma did not require the voter to register more than a first choice for any office. It seems to have been quite common in all the states indicated above that a great many voters failed to avail themselves of the privilege of registering second or more choices, which resulted in the practical restoration of the plurality system in many primary races. This failure was due to several causes: ignorance of the voter; his desire not to have his vote counted for any but his first choice; or his refusal to accept what was thought to be a complicated system, which, it was felt, could be easily corrupted or readily subject to mistakes in the count, or which seemed to provide for an unfair method of evaluating choices.

2. Failure of party leaders and officials to educate the public in the use of the preferential feature, due partly to their opposition to it as a complicated device and one the results of which could not be easily anticipated.

Weeks’ article contains a table of the 11 states that used IRV or the Bucklin system within primaries. Four had gone to runoffs by 1931. Six had gone to plurality. One more (Maryland) continued using a ranked-ballot system through 1937.

I touched on these issues here (pp. 177-180). The topic would make for an interesting research project.