The most common way to look at election shifts is to look at county-level swing. This is the standard, accepted tool in virtually every form of analysis, and it is quite easy to compute: for each county, simply take the margin in one election and subtract it from the margin in another. For instance, if a county was R+50 in 2012 and R+40 in 2020, it means it swung ten points to the left. And if you looked at the 2012 to 2020 presidential swing through this metric, you’d be able to gather some very valuable insights regarding which areas were trending Democratic and which ones were trending Republican.
Where this tool begins to break down, however, is in long-term trend assessment over longer periods of time. Counties are not static, atomic units; they constantly gain or lose population, and the partisanship of new voters does not always match a county’s existing partisan lean. A case study of this may be found in North Carolina, where the solidly Republican Johnston County (a Raleigh exurb) swung 3 points to the left from 2012. However, Johnston was actually growing rapidly in size, gaining nearly 50,000 people between 2010 and 2020. The new voters migrating in may not have been as conservative as the rest of the county, but they were still more Republican than they were Democratic. And so the trends in this county actually favored Republicans — the county did swing marginally to the left, but this was not enough to offset the sheer influx of Republican voters, and so it actually began giving the GOP more raw votes than ever before, even relative to the statewide total.
On the flip side, it could be the case that a county swinging Republican actually is trending favorably towards Democrats, simply because it is depopulating so quickly that it is no longer able to give Republicans the same vote margin relative to the state that it once could. Let’s consider the case of Sherman County in north Texas as an example. This sparsely populated county swung 4 points to the right between 2012 and 2020, but it actually saw its population decline significantly, losing 8% of its people between 2010 and 2020 even as Texas grew by 16%. Thus, this county’s trends were actually quite favorable for Democrats, in a sense — although voters there wouldn’t be backing Joe Biden anytime soon, because the county’s population is declining far faster relative to Texas’ statewide population than it is swinging right.
These case studies show how looking at a swing map and gauging trends simply through looking at county-level swings might be a bit misleading. It is not enough to say that a county is trending Democratic because it swung slightly to the left, because population trends must be accounted for as well. To get around this, several people use the concept of raw vote margin — however, by itself, this is also misleading, because if an election has universally higher turnout, then most areas would be shown as giving more raw votes to a party. That, of course, says less about the trends in an area than it does about turnout itself.
To address this, we devised a one-size-fits-all metric that considers both population trends and margin swing in determining whether a county has favorable trends for Democrats or Republicans, and we’ll use this metric to examine the shift between the 2012 and the 2020 presidential elections. For brevity, we leave the mathematical calculations here to the footnotes, where we describe them in greater detail, but the effects of our metric are as follows:
- Red counties that grew faster (in terms of total votes cast) than the state did will trend right by more than the raw margin swing.
- Red counties that grew slower than the state did will trend right by less than the raw margin swing.
- Blue counties that grew faster than the state did will trend left by more than the raw margin swing.
- Blue counties that grew slower than the state did will trend left by less than the raw margin swing.
The growth-adjusted picture can explain a few things that may be glossed over by the initial swing map presented at the beginning of the article, and when you compare the two maps, some interesting differences emerge. For instance, Texas and Kansas begin to look a lot more favorable to Democrats, with the depopulation of heavily Republican rural areas in both states becoming much more stark and on display with the adjusted map. In Texas, in particular, the rural depopulation provides a powerful Democratic supplement to the leftward swing of suburbanites in the rapidly-growing metropolitan areas of Dallas, Fort Worth, Austin, San Antonio, and Houston, and it adds to the cadre of reasons that Texas is arguably the most promising medium-term project for Democrats.
Meanwhile, North Carolina looks a shade more favorable to Republicans than a swing map would indicate; this is because some of the Republican strongholds, such as exurbs around Raleigh and Charlotte or counties like Brunswick, have been growing at a much faster rate than the rest of the state. This means that Republicans are netting more and more votes out of the deep-red exurbs than ever before, even though their margin may be shrinking. This helps counter the heavy leftward swings seen in the cities and other suburban areas, and these counteracting trends are why it is more likely than not that North Carolina will be purple for the foreseeable future.
Lastly, perhaps the most interesting case study is Utah, where the explosive growth of the Salt Lake metro area has further diluted the voting power of the state’s deep-red rurals. The state had already swung sharply left between 2012 and 2020 due to the contrast between Mitt Romney (a member of the Mormon Church) and Donald Trump (a comparatively bad fit for Mormon conservatism) — however, our metric shows that this swing left is more favorable for Democrats than one may think, due to the population trends in much of the state. While it is a stretch to expect blue Utah anytime soon, it is these trends that led to Democrats flipping UT-04 blue with Ben McAdams. Any possible Democratic statewide victory in the distant future, even if it is two decades from now, would be powered by the swing and the population growth seen here.
The electorate isn’t the same across two presidential elections, and subtracting margin can often mislead. The metric we’ve proposed isn’t perfect, but it does adjust for population changes in a way that no other metric we’ve seen does yet. It is important to note, however, that this map is still not all that different from the swing map in the majority of scenarios. The midwestern and working-class swing right is still prominent and striking, as are the leftward sprints of the Atlanta and Phoenix metros and the rightward swings of south Texas and south Florida. Because at the end of the day, we aren’t arguing for the abolition of swing maps; they’re still very useful tools and the industry standard for a reason. We’re simply proposing a better way to contextualize margin shifts with population changes, and that is essential for any long-term trend analysis.
Ed. Note: For comparison’s sake, we attach the turnout-adjusted swing map, the raw swing map, and the difference between the two (turnout adjusted minus raw swing) below for interested readers to toggle through.
Mathematical Details: Traditional percentage margin is calculated by taking the raw vote margin for a party and dividing it by the votes cast, on a per-county basis. We do something a bit different: we divide the 2020 raw vote margin by the expected votes cast based on the results of the 2012 election. This is calculated as follows: multiply the votes cast in 2012 by the turnout spike experienced by the state between 2012 and 2020, essentially scaling the 2012 votes cast up by the statewide turnout increase. Let this variable equal expected_votes_2020.
Then, the swing is calculated as follows: adjusted_swing = raw_vote_margin_2020/expected_votes_2020 – raw_vote_margin_2012/votes_cast_2012.
Doing this means that Republican areas that grow faster than expected will be more friendly towards Republicans than the traditional “swing” metric may indicate, as those red areas are more likely than not to be adding Republican-leaning voters, even if they are not as Republican as the rest of the county. By similar logic, Democratic areas that grow faster than expected will be more friendly towards Democrats. Lastly, Republican areas that are bleeding population or growing slower than the rest of the state will be adjusted to be more friendly towards Democrats, because Republicans are likely losing votes in those areas.
It is important to note that we gauge growth in terms of votes cast rather than in terms of population added. We think this is a more robust way to measure growth in elections, as it is among the easiest ways to directly gauge the impact of new voters on the electorate, because a new voter isn’t of much use to a party until they actually cast a ballot. Doing this also makes the metric more generalizable to different types of cycles and elections (i.e. Senate elections), where elections are more subject to wild swings based on turnout due to the lower salience of the elections.
I’m a software engineer and a computer scientist (UC Berkeley class of 2019 BA, class of 2020 MS) who has an interest in machine learning, politics, and electoral data. I’m a partner at Split Ticket, handle our Senate races, and make many kinds of electoral models.