County Elasticity: A Metric for Quantifying the Bipartisan Voting Tendencies of a County


The concept of a "swing state" is thought to be an easily-understood notion in politics -- it's a state that could tip either way in any given election. However, one of the biggest misassumptions in politics is the notion that swing states are all comprised of moderate swing voters. While this is certainly true for some states, like Wisconsin, it definitely isn't the case for all of them; in fact, several swing states, like Florida and North Carolina, are home to relatively few swing voters. The closely contested nature of these states comes instead from a relatively equal set of committed partisans on each side, and election victories in such states generally go to the candidate that can turn out the most voters on their side.

This is a tough thing to measure, however -- how can we understand which category certain states fall in? One thing we could examine is the tendency for the state's counties to swing between parties across elections -- for example, a county could vote for the Republican by a 5 point margin for Governor and vote for the Democrat by a 10 point margin for President, indicating a high degree of openness to voting for any candidate, regardless of party. Examining this tendency should thus help us gauge the partisan loyalty of a county across offices and would thus provide a rough but vote-based estimate of the types of voters in an area and their tendency to swing between parties. With context, this would greatly help us in identifying how persuadable voters in counties really are.

Understanding the composition of voters in areas is an extremely important thing for candidates, as it helps them figure out what resources to devote to where and aids in setting campaign strategy for maximizing the ultimate goal: votes. For example, Joe Biden would likely be wasting his time airing ads about bipartisanship in a county like Florida's Broward County that already is 66% Democratic -- he'd be much better off airing those ads in Wisconsin, where more swing voters exist, and investing in a heavy turnout machine in Florida to turn out his base instead (a strategy that won Obama the state twice) .

To measure this tendency, we'll introduce a concept called "elasticity" for counties. This concept measures the deviations in a county's percentage-based vote margin across a set of elections, which we use as a proxy for the openness of a county's voters to voting for candidates across the political spectrum, regardless of political affiliation.

The metric is computed as follows: For any set of elections {A, B, C, D, ...}, we plot the (Republican, Democrat) vote by county on an (x, y) coordinate scale and compute the pairwise Euclidean distance between all points. These distances are then all summed to obtain an elasticity score for the county.*

It is important to note that what this metric measures is the vote-based electoral "bipartisanship" of counties across offices -- i.e. "How much has this county's vote varied across elections?". The underlying idea is that counties with voters that are more open to splitting tickets (or voting for different parties across races) would have a far higher elasticity score, as their vote margins should vary a fair amount between elections. For example, states like Montana are safe Republican at the presidential level, but are highly elastic in down-ballot races -- in fact, come January, Montana may have 2 Democratic senators and a Democratic governor, even though Trump should win the state by double digits.


Let's consider a few case studies now. The 4 elections we will consider in computing elasticity are the two most recent presidential elections, the most recent Senate election, and the most recent gubernatorial election. Elasticity is clipped to be between a 0-100 scale, with 0 indicating the least elasticity and 100 representing the most.

First, we'll examine Florida and Wisconsin, two of the most well-known swing states in the nation, as a case study. As a rule of thumb, the elasticity of a county is directly proportional to the intensity of its shading -- darker counties are more elastic.

It's clear that Wisconsin and Florida lie on opposite ends of the elasticity spectrum; Wisconsin appears to be as elastic as Florida is inelastic. In Florida, therefore, elections generally come down to each side turning out their respective bases in as large numbers as possible and are far less dependent on swing voters than one may think. Wisconsin, meanwhile, appears to fit the "swing state with swing voters" profile that so often finds its way into cable punditry, as the individual counties are far more open to voting for candidates across the political spectrum.

The elasticity of a county is especially useful for minority party officials in states that traditionally vote the opposite way (i.e. red-state Democrats and blue-state Republicans). For example, an analysis of Colorado shows that the state is highly inelastic, meaning that Senator Cory Gardner (R-CO) is in deep trouble for this coming election.
In an election like 2020, turnout will almost certainly be extremely high (unlike, say, 2014). So for Cory Gardner, a Republican running in a blue state, it's not about depressing Democratic turnout and having higher Republican turnout -- that's impossible in a presidential election like 2020. Gardner will instead need swing and crossover voters to give him a path to re-election.

You would expect, if these voters still exist, that they would have been present at the statewide level in 2018, or maybe even in Senator Michael Bennett's (D-CO) 2016 election. These are voters that would split tickets at state level for the right candidate -- voters who would vote Democratic at the presidential level because they hate Trump, but would vote for a down-ballot Republican because they like his proposals for the state, or his ability to serve as a check on a potential Democratic presidential administration.

The problem for Gardner? Those voters just don't seem to exist in Colorado anymore. The elasticity metric shows that all of the 11 most populous counties (El Paso, Denver, Araphoe, Jefferson, Adams, Larimer, Boulder, Douglas, Weld, Pueblo, and Mesa) are highly inelastic. The only elastic counties across the 4 elections mentioned are sparsely populated and historically Republican ones with fewer than 5,000 people in them, and this is not good news for the incumbent senator.

Finally, we can also use this to understand historical trends in a state. California** was once a swing state with a strong, vibrant, and extremely powerful Republican party, and between 1983-2011, it sent exactly one Democratic governor to Sacramento (and ended up recalling him mere months into his second term), even as the state turned reliably blue at the Senate and Presidential level from 1992 onwards. However, in the last ten years, the state GOP has slowly become extinct, and the political calculus in the state has shifted towards a battle between progressive and moderate Democrats. We see that the elasticity of California has drastically dropped from 2004-2018 -- the state was once fairly open to electing candidates of any partisan affiliation, but that tendency has now almost entirely disappeared. Santa Clara County, for example, voted for Schwarzenegger (R) in 2006 by a 10% margin and Obama (D) by a 41% margin, but backed Clinton and Newsom (both Democrats) by over 40 points in 2016 and 2018, respectively.

It is at this point that we wish to highlight something: elasticity is a highly useful first flag that gives information about a county's voting patterns, but it can also be misleading if used as the only metric -- although this is far less common, the same macro-level polarization along partisan lines at state level could, in theory, exist at county level too, meaning that a county may also be highly polarized along partisan lines and yet have a high elasticity metric based on which bases turned out for which elections. Therefore, while elasticity helps provide extremely useful information about a county's voting history and helps give a metric to quantify the recent behavior of voters on a county-level basis, it is important to remember that this must be combined with analysis of the county's demographics, composition, and trends to truly draw conclusions.

For example, let's take the state of Kentucky. Although it is, by a considerable margin, a very Republican state, it is not impossible for Democrats to win, and this is shown by Andy Beshear's razor-thin victory over a deeply unpopular Matt Bevin in 2019. Given the immense deviation from the partisan voting index required to achieve such a victory, Kentucky's counties would thus have a very high elasticity metric for the aforementioned set of statewide elections, but it would be a mistake to assume that it's a swing state -- Pike County going from R+63 to R+12 in an election will mean it has a high degree of elasticity, but that's certainly not the same as the county being competitive! Moreover, we know that Beshear's 2019 victory was largely buoyed by a popular statewide reputation, with his father being a former governor and the candidate himself being the Kentucky Attorney General, and his running against an extremely unpopular candidate.

However, it would be a mistake to discount this election entirely; whatever the surrounding circumstances may be, Beshear is still a Democrat who won in a deeply Republican state, and that has to count for something when considering voter behavior. When viewed in context, a case like this thus indicates areas where Democrats may still be able to do well enough, given the right set of circumstances -- indeed, eastern Kentucky is an area with a lot of ancestral Democratic strength, a fact borne out by the map above, and the right candidate may be able to peel off enough votes. In a place in which there aren't that many Democrats left to begin with, this matters a lot -- even driving down the Republican margin of victory by a decent amount is vital, and this metric highlights areas in which they should be able to do that. Thus, we still get a lot of useful information out of examining a state like Kentucky. Elasticity, when combined with priors about the state's demographics and electoral circumstances, can thus still tell us a lot about an area and its electoral tendencies.


The concept of elasticity, as presented in this article, shows us the tendencies of voters in a state on a county-level basis, and helps us examine their tendency to split tickets for candidates of different parties across elections. It's important to note that a metric like this is not at all the be-all, end-all for gauging the partisan tendencies of a district; it is instead merely a useful technique to help flag counties of attention based on prior voting patterns. However, in context, a metric like this helps gives candidates and campaigns an understanding of how to devote and allocate resources across counties in order to maximize votes gained. For example, a county with high elasticity is likely one in which many swing voters reside, meaning it is probably one in which the candidate would spend time preaching the benefits of bipartisanship and moderation, whereas a highly inelastic area like southeast Florida would be one in which the candidate would do better to invest in turnout operations that would bring their base to the polls instead.

*For those of you that are more graph theory-oriented, another way to think of this is as a fully connected graph consisting of x nodes, where the node coordinates are their respective (R, D) vote totals. The edge length between each pair of nodes would be the Euclidean distance between them. The elasticity score for a county would then be the sum of all graph edges.

**As California switched to a jungle-primary system, locking the Republicans out of the statewide November elections entirely from 2016 onwards, the 2012 Senate election was utilized for computing the elasticity metric of California between 2012-2018.

Author's note: a version of this piece, with more graphics and applications to 2020 swing states, was published in Sabato's Crystal Ball. Click here to read it.


  1. This is such a well-written article! I'm glad that you were able to turn your 50-tweet tweetstorm (which was very intense) into a coherent article that was honestly really fun and informative to read. Maybe you should make a version of this that provides insights for what Biden's campaign should do to maximize how they can spend their time and resources, and send it to Biden (as well as the people over at Vote Save America), I think they'd find it interesting!

  2. Very interesting metric to develop! However, I would say that the metric of calculating the Euclidean distance will conflate elasticity with one-time structural shifts. The map of Wisconsin is telling here - much of the state is quite swingy (really with Milwaukee county as the exception), and southwest and Northern Wisconsin had big structural shifts towards Republicans during this period.

    Perhaps one way to counter balance is to use county-level demographics to (i) identify is the county composed of types of voters who are swingy everywhere, and (ii) use those demographics to subtract out predicted structural shifts.

    Unfortunately with any backward looking variable, elasticity might change substantially over time, so using more than 15 years of data might not be reliable. But, if you're using short periods, you will mix up structural shifts with elasticity.

    1. You're absolutely right that we may end up with interference from structural shifts. The issue is that those shifts are often kind of hard to identify right away, and "reversion to the mean" is a very real concept electorally that has some sway as well. This metric is only meant to add on give a more complete picture of an area instead of serving as a "one-size-fits-all" measure. I only try to quantify how an area has voted over a certain period of time and see if we can augment our knowledge on an area with this information to better gauge how the state will vote. Essentially, even with a static metric, the conclusions would vary based on the state and the demographic and circumstantial priors.

      I'd love to add in demographic data. Unfortunately, I'm not quite sure where to find it. I have added turnout, though; I'll see if I can do another post about that soon!

    2. You could use various economic variables as co-variates for reducing the variance in elasticity. I believe that data is available at the county level from the US Department of Labor or Census.

  3. It would also be interesting to see if elasticity is a function of education, economic status and/or religiosity. My hypotheses would be that elasticity would decrease with religiosity and increase with education and wealth, the later two being highly correlated.


Post a Comment

Popular Posts