America’s first (modern) election with Approval Voting

On June 9, Fargo (ND) held its first election under Approval Voting. It was the first such exercise for a U.S. public election, unless we count a variant used in the early republic. Fargo’s precinct-level data are available here. What follows is an attempt to make sense of those data.

[Note of 2020-12-23: I noticed that the link above (to the election data) had somehow become a second link to the “variant” paper. Here is a new link to the data page, and here is a link to a fresh data export. These data still do not include the total number of ballots.]

The election was nonpartisan, for two seats on the City Commission. There were seven declared candidates. Two (Strand and Grindberg) were incumbents. A third (Preston) had served in the 1990s. The winners were Strand and Preston. More information here.

According to this article, voters were asked to “Vote for ALL of the names you approve of.” This appears to conflict with what is said on the state website: “Vote For 2.” If I recall correctly from a Twitter conversation, the article has it right (not what’s implied on the state website).

One issue in approval voting is the extent to which voters “bullet vote” — or “truncate their approvals,” per the literature. I do not know this literature as well as I should. I do know, however, that this gets into larger issues about whether approval removes the need to vote strategically. We cannot tell, from the official data, how many voters “approved” of just one or two candidates. That is because the data do not include the total number of ballots cast, valid or otherwise.

A related issue is the extent to which approval voting avoids “center squeeze.” What does this mean? We start with a uni-dimensional model of politics. We then add in multi-candidate competition. (It may be that multi-candidate competition implies we need more than one dimension, but that is a separate issue.) As long as support for the centrist(s) is less than that for extremists, voters can be expected to desert centrists for extremists. This is said to hold for plurality, runoff, and instant runoff. With approval voting, however, there is no need to desert. Examples of this logic are here, here, and here.

Did the Fargo election reflect center squeeze? Part of the answer involves saying how much is too much. Approval advocates (see links above) tend offer the following standard: one of the extremists beats the centrist(s), even though most voters might have preferred the centrist(s). We do not have access to voters’ sincere preferences. But we can project the entire election onto a line — or more than one line, if that’s what comes from the data. I tried to do so using principal-components analysis. Observations are precincts, variables are candidates, and cell entries represent the percentage of votes (or “approvals”) for each candidate in that precinct. We are working with a 20×8 matrix (the last column of which is write-in).

Caveat: better ways to do this involve ballot-level data and formal models of voter decision. Nevertheless, the PCA with precincts may be instructive.

My first graph suggests a two-dimensional space (not one dimension, as in center-squeeze theory). Bars reflect the shares of overall variance in the data explained by each dimension.

My second graph helps think about what defines the space. The length of each arrow along each dimension suggests how much each candidate (named at the end of each arrow) is related to that dimension. The first dimension clearly captures competition between incumbents (Strand and Grindberg). It also captures Preston, the former commissioner, who appears very close to Strand. The second dimension looks like a sort of anti-establishment one, picking up votes for candidates without (detectable) history on the Commission. But the first dimension also picks up division among these other candidates.

My last graph arrays candidates from left to right, along the main dimension (using their factor loadings). This is not a statement about how substantively liberal or conservative anyone is. It is simply the first dimension that emerges from the data. Note that the two winners, Strand and Preston, anchor the “right” of this space. Based on this graph and the preceding one, it really looks like they campaigned as a team. I’d be interested to know if they did, and if so, what that looked like.

So, did the center get squeezed? By the winner metric, yes, at least given these data. By a voter-desertion metric, it is impossible to say. We don’t have access to preference orderings, nor to the data on “bullet voting.” And I am reluctant to talk about a “center” at all. What I see in these data are two opposing coalitions — one defined by Grindberg, then a second by Strand/Preston.

Why Democrats cannot change the rules even though it would be good for them: Presidential edition

The impetus for this post is a recent, casual discussion with a Green Party voter. I suggested that their cause might be well-served by withdrawing Howie Hawkins and instead endorsing Biden. The alternative is to run a spoiler campaign (in appearance or reality).

The result would be, in essence, a fusion candidacy. This would signal willingness to govern — and contest office, which would need to happen in our presidential system — in coalition with Democrats. In turn, the Greens might secure support for the sorts of reforms they are likely to want, e.g., a larger US House, instant runoff voting, you name it. And even if those reforms were not forthcoming, Greens at least would have signaled that they are open to coalition — a fancy way of saying “playing nice.”

The response was as you might expect. The only way that Democrats will come to back reform is by facing threats from spoilers.

The rest of this post focuses on my interlocutor’s theory of reform. As far as Democrats go, some are indeed open to the idea of coalition.

Below is a list of the top ten closet states in the 2016 presidential election, via US News, verbatim. I have added asterisks to those states with divided government, i.e., where Republicans control enough of government to block instant runoff voting. The data on party control are from Ballotpedia.

My post assumes that Republicans are unanimous in their opposition to IRV, based on expectations about the 2020 election. Maybe true; maybe not.

Further, I am not saying that IRV even would matter here. All chatter I’ve seen so far suggests less third-party voting this year than we saw in 2016.

A fuller analysis might cover all 50 states, but this is a good start. The smallest margin in this list is 0.3 percentage points (Michigan). The largest margin is 3.9 points (Arizona). Just for argument, let’s say my friend’s theory of reform does not operate at larger victory margins.

Republicans control enough government to block reform in 8 of these 10 states. There are just two Democratic trifectas. One is in Maine, which will use instant runoff this November. The other is in Nevada. I don’t know what is happening in state government there, but Nevadans for Election Reform did try for a ballot measure.

The list

1. Michigan 0.3 percent*

Trump 47.6 percent, Clinton 47.3 percent

Difference: 13,080 votes

2. New Hampshire 0.4 percent*

Clinton 47.6 percent, Trump 47.2 percent

Difference: 2,701 votes

3. Wisconsin 1 percent*

Trump 47.9 percent, Clinton 46.9 percent

Difference: 27,257 votes

4. Pennsylvania 1.2 percent*

Trump 48.8 percent, Clinton 47.6 percent

Difference: 68,236 votes (99 percent reporting)

5. Florida 1.2 percent (R trifecta)

Trump 49 percent, Clinton 47.8 percent

Difference: 114,455 votes

6. Minnesota 1.5 percent*

Clinton 46.4 percent, Trump 44.9 percent

Difference: 44,470 votes

7. Nevada 2.4 percent (D trifecta)

Clinton 47.9 percent, Trump 45.5 percent

Difference: 26,434 votes

8. Maine 2.7 percent (D trifecta)

Clinton 47.9 percent, Trump 45.2 percent

Difference: 19,995 votes

9. North Carolina 3.8 percent*

Trump 49.9 percent, Clinton 46.1 percent

Difference: 177,009 votes

10. Arizona 3.9 percent (R trifecta)

Trump 49.3 percent, Clinton 45.4 percent

Difference: 91,682 votes

Thoughts on Massachusetts Question 2

This fall, Massachusetts voters will decide on preferential (aka ranked choice) voting. Political science is polarizing on the larger issue, although there is an impressive and diverse list of scholarly endorsers, at least in this case. Here are some observations on the Massachusetts measure.

First, let’s be precise about what’s on the ballot. RCV will apply to “all state and federal elections in Massachusetts, both primary and general elections,” but not presidential elections (per the Yes on 2 website).

The measure would not affect the composition of primary electorates. In other words, it would not let Republicans influence Democratic nominations, nor vice-versa, any more than the current rules. Under those rules, according to the group Open Primaries, unaffiliated voters already may vote in the primary of either party. Registered Republicans and Democrats, however, must vote in their respective parties’ primaries (or not at all). RCV in Massachusetts would not change this rules constellation.

Other recent RCV proposals would change primary electorates. These proposals incorporate California-style “top two,” better understood as a nonpartisan two-round system. Such systems do not let a party advance a single candidate (or slate) to the decisive election. My understanding of “top two” is that it has been hard on Democratic Party establishments, and organized labor in particular. One can imagine those fortunes changing, but, for now, this seems to be the big effect.

Also, it is crucial to understand the reform coalition itself. In Massachusetts, as far as we can tell, it is Libertarians, “libertarians,” and a good chunk of the Democratic Party establishment. The measure therefore reflects a broader, ongoing realignment in American politics. I suspect these groups would coordinate anyway, in the absence of a ranked ballot, given the current state of the Republican Party brand.

One common argument in reform circles is that RCV stimulates candidate entry, and victory, by women and people of color. The fact that it worked out this way in the California Bay Area may have more to do with the reform coalition than the reform itself. One of my current research projects is to see whether the reform itself plausibly has this effect. My priors really are not sharp. On the one hand, the Alternative Vote is a very modest reform. Far more important are district magnitude and the size of an assembly. On the other hand, at least recently, women have tended to do well in both single-seat RCV and runoff systems. The runoff evidence is from Cynthia McClintock and Joseph Cerrone, who presented yesterday in my APSA panel. Their evidence comes from Latin America and shows that, in recent years, women have done better under runoffs than in single-round systems. The RCV evidence is from Represent Women (and does not cover non-RCV cases). Finally, some have argued that MA needs RCV because its primaries already are crowded. Maybe that says something about the effect of reform itself.

It also is worth noting that, historically, in foreign countries, single-seat RCV has been used to stop the economic left from winning. British Columbia in the 1950s and Australia in the 1910s are glaringly obvious cases of this. Again, however, this may have more to do with reform coalitions than the substance of reform itself. A similar reform (runoffs) once was used to insulate a liberal-labor coalition. Many left-labor-type groups formally support the MA measure: an SEIU local, Our Revolution, etc.

Overall, the Massachusetts package feels pretty “safe.” If anything, it seems like it will let people do what they plan to do anyway.