Revisiting What Happened in the 2018 Election

An archived version of the data used in Catalist's 2018 analysis is available here2018 analyses should not be used for comparative purposes now that updated data are available. Please refer to the latest What Happened analysis for the most up-to-date estimates.

May 21, 2019

Author: Yair Ghitza, Chief Scientist

Immediately following the 2018 election, we published an analysis of demographic voting patterns, showing our best estimates of what happened in the election and putting it into context compared to 2016 and 2014. That analysis was preliminary, based on various sets of pre-election data and precinct-level election results from a few states, that were available at the time1The methods used are described in a preview post.. The goal was to put data out quickly, and to update the analysis as more data became available.

Since then, we’ve collected much more data — precinct results from more states and, importantly, individual-level vote history records from Secretaries of State around the country. This analysis updates the earlier work and adds to it in a number of ways. Most of the results we showed remain the same as in the earlier analysis, but there are some changes.

The rest of this post will detail what the data tells us in a number of ways — first by revisiting demographic trends and adding important geographic pieces. After that, we’ll use the data to explore an important and much-discussed question: how much of the change from 2016 was due to different people voting vs. the same people changing their vote choice? Lastly, for the more technically inclined, we include an appendix to discuss how our preliminary estimates compare to the updated data.

  1. Turnout increased dramatically compared to past midterms, and the composition of the 2018 electorate resembled recent Presidential electorates much more than recent midterms. Young voters and voters of color, particularly Latinx voters, were a substantially larger share of the electorate than in past midterms. White non-college voters and people we’ve historically modeled as Republican supporters were a smaller share. The 2018 electorate was similar to 2016, with the exception of age: midterm electorates are older than Presidential electorates generally speaking, and 2018 ended up somewhere between 2016 and 2014 in this regard.
  2. Comparing the 2018 Congressional results to the 2016 Presidential election, Democrats gained about 5 points in margin, but Democratic gains were uneven across different parts of the electorate. 2Hillary Clinton won the 2016 popular vote by 2.1 points. The Democratic margin for 2018 was 8.6 points, but there were many uncontested races that Democrats won 100–0, skewing comparisons between years. We project election results for uncontested races to make results between years more comparable; the “projected” margin for Democrats was 7.3 points. See here for more detail.We saw the biggest gains among young white voters, white college voters, and people we’ve historically modeled as neither Democratic nor Republican — these aren’t exactly “Independents” but we think this is a meaningful group.
  3. Democratic gains were uneven across geography too, both at the Congressional level and in statewide elections. There has been a lot of attention paid to the Democratic victories in suburban areas, but we find that Democratic gains were actually largest in rural areas. These gains weren’t enough to get over 50% and win seats in many rural districts, so they have escaped much of the mainstream election analysis to this point. These changes are nonetheless important, particularly because they were large in many of the midwest battleground states that will no doubt be important in 2020.
  4. Compared to 2010 and 2014, 2018 saw (a) less Presidential dropoff (people who voted in the Presidential election and didn’t vote in the midterm), and (b) more midterm surge voters (people who didn’t vote in the last Presidential but did vote this time). Voting patterns for all of these groups show more Democratic enthusiasm and support than past years.
  5. Thinking about the change from 2016 to 2018, it is clear that both mobilization and persuasion were critically important in producing this scale of victory for Democrats. When it comes to turnout, the composition of the electorate roughly “broke even” with 2016, much different than the past two midterms. But “breaking even” doesn’t explain the amount and geography of gains that Democrats saw. A large portion of gains came from people who voted in both elections, switching from supporting Trump in 2016 to supporting Democrats in 2018. We show some of the math behind this, including how that conclusion changes in different areas of the country.
  6. Looking ahead to 2020, it is reasonable to expect another historic level of turnout, perhaps approaching 160 million votes or more. It is not safe, however, to assume that Democratic gains from 2016 to 2018 will hold. Republican gains in 2010 and 2014 bounced back to Democrats 2 years later (at least in the national popular vote in 2016), and Democrats should be aware this is a distinct possibility going into 2020 as well.

As always, it is worth emphasizing that many of the data points shown here are estimates, based at least partly on survey data and statistical modeling. In other words, there remains uncertainty around these results. For more information, see here and the section on statistical methods below.

Composition of the Electorate and Voter Turnout

The composition of the 2018 electorate resembled recent Presidential elections more than recent midterms.


Turnout rates by age, across selected recent elections.


Vote Choice

The major turnout surge in 2018 was only part of the story. Democrats saw gains across many different areas of the country, due to both turnout and changing vote choice, i.e., people switching from voting for Donald Trump in 2016 to voting for Democrats in 2018. We will examine turnout versus vote choice more explicitly later on, but first this section lays out baseline data for how different groups voted in recent elections.

Democrats gained 5 points in margin nationally, but gains were uneven across demographic groups.



Democratic gains were also geographically uneven across the country. In our view, these trends have been fairly widely mischaracterized.4This section is built on joint research that we conducted with People’s Action. Most people are probably familiar with the large-scale rural/suburban/urban story from 2016. Despite rural areas being more Republican even before this election, the change from 2012 to 2016 was dramatic: rural areas became even more Republican, suburban areas became more Democratic, and urban areas largely stayed the same, with small shifts towards Democrats:

Rural America shifted towards Republicans from 2012 to 2016.


Media reports have largely described 2018 as a continuation of this trend, focusing mainly on suburbs that flipped Congressional seats from Republican to Democratic, and saying that the urban/rural divide is bigger than ever. But this misses a critical part of the story: rural areas largely moved in a Democratic direction, often by even larger margins than the suburbs.5The maps show census tracts, using precinct-level election results where available, and data projected from individual-level voter registration data and district-level Congressional results elsewhere.

Despite media narratives to the contrary, rural American bounced back towards Democrats in 2018.


The trend lines below show the year-to-year trends more explicitly. We categorize every census tract in the country as urban, suburban, or rural, based on population density. The most urban (dense) tracts are shown on the left, and the most rural tracts are shown on the right. From 2012–2016, rural areas became more Republican by about 11 points in margin. From 2016 to 2018, there was a major bounce-back, with the same areas becoming about 6 points more Democratic. Suburban areas trended Democratic in both elections, and urban areas moved towards Democrats by a relatively small amount.

The urban/suburban/rural trends, shown more explicitly.


A state-by-state analysis is outside the scope of this post, but it is important to realize that this trend was not consistent across the entire country. It was strong across the Midwest, but not always seen in the South. The data above shows changes from 2016 President to 2018 Congressional, but the trend was also seen in many different statewide elections, and can be seen using publicly available data, as was done here. Some of this was undoubtedly influenced by incumbency and other local factors that may not necessarily translate in a nationalized election like the 2020 Presidential race. When doing a national analysis like this, by necessity we are capturing national trends, while averaging over some important state-by-state distinctions. Suffice it to say, we think it is important to understand what drove these changes, in order to understand potential changes in 2020 Democratic and Republican coalitions.

Presidential Dropoff and Midterm Surge

Traditional election analysis, including the earlier sections of this post, deal with fairly standard sets of group-level aggregate data. Who did Latinx voters support (as a group), and how did Precinct X vote (as a group)? Using individual-level voter registration data, collected over time, opens up the possibility for different groups that are defined by their recorded behavior.

  1. People who voted both times
  2. People who voted neither time
  3. People who voted in 2016 but not 2018 (Presidential dropoff voters)
  4. People who didn’t vote in 2016 but did vote in 2018 (midterm surge voters)

Higher turnout was associated with relatively low voter dropoff from 2016, and a large number of new voters.


Putting It Together: Different Electorate and Different Vote Choice

[UPDATE: In 2020, Catalist refrained from publishing a decomposition of the electorate similar to the one it provided in 2018. An FAQ in the What Happened 2020 analysis explains this reasoning:

In our analysis of the 2018 election, we tried to explicitly decompose the impact of differentia turnout versus changing vote choice to explain changing election results from one year to the next. We decided against publishing a similar analysis in this report, for two reasons:

  1. Our report in 2018 repeatedly and strongly emphasized that both turnout and vote choice played an important role in Democrats’ victory: (a) the spike in Democratic turnout compared to past midterms made the electorate demographically similar to 2016 which was incredibly important, and (b) changing vote choice among 2016 voters pushed things even more in the Democratic direction. Point (a) was lost to many, because our “decomposition” calculation used 2016 as a baseline, and was explicitly comparing that 2016 (Presidential) election to the 2018 (midtem election).
  2.  In retrospect, we consider the “percent impact” number we developed to be a noisy indicator of the impact of persuasion vs. mobilization. The number divides one small number (the impact of turnout or changing vote choice, based on survey-driven estimates) over another small number (the change from one election to the next). From 2016 to 2020, we would be trying to decompose a 2-point margin change (the denominator), using a numerator with too much uncertainty around it. In other words, small changes in our estimates could lead to large-looking changes in the “percent impact” numbers.

The original 2018 decomposition analysis below is preserved for transparency.]

Because of these two factors, we think it is more prudent to lay out the various factors in more descriptive terms, as we’ve done in this report. For more information on why we consider some of these numbers to have substantial uncertainty around them, see details in the New Generation of Voters section. Overall, we know that people’s enthusiasm to vote and their propensity to cross over to vote for a different party or candidate than they have previously can be closely related. We’re interested in developing more sophisticated and rigorous methods for decomposing this effect, and are interested in working with researchers who have interest in studying this more going forward.

As different years bring different election results, many people have debated the extent to which these changes are driven by (a) differential turnout or (b) changing vote choice.

  1. President to midterm dropoff voters: how much loss was due to Democrats who voted in 2012 but dropped off and didn’t vote in 2014?
  2. New, midterm surge voters: how many new votes did Republicans gain due to new voters who showed up in 2014 but not in 2012?
  3. Changing vote choice among people who voted both times: did this happen, and to what extent?

Decomposing the change in total outcome (margin) from 2016 to 2018. Turnout was important in essentially breaking even with 2016, but Democratic gains were also largely driven by voters who voted for Trump in 2016 and voted Democratic in 2018.


Decomposing the change in total outcome (margin) from 2016 to 2018, for different statewide and Congressional elections. Dynamics changed from place to place, but overall, many of the places that saw big Democratic gains had a large component of vote choice, i.e., Trump to 2018 Democratic voters. Geographies are sized by how competitive they were in 2018.10Congressional races are sized smaller than statewide races to keep things fairly visible.


Looking Ahead to 2020


This analysis has many implications for the 2020 Presidential election, frankly too many to be covered in detail in this post. We’ll briefly focus on three topline takeaways here.

Expect a historic number of votes in 2020.


First, on turnout: there are few signs that the overwhelming enthusiasm of 2018 is slowing down. 2018 turnout reached 51% of the citizen voting-age population, 14 points higher than 2014. 2016 turnout was 61%. If enthusiasm continues, how high can it get? It is unreasonable to expect a 14 point boost up to 75%, but is 70% reasonable? Here we show that turnout could easily reach 155 to 160 million votes, due to a boost in the turnout rate and the steadily increasing population size, which could reach around 240 million people in 2020.

Decomposing the change in total outcome (margin), for the last two Presidential swings. States are sized by how competitive they were in the latter election.


Republican gains in the past two midterms swung back towards Democrats in the following Presidential election. The diagonal line shows 100% “bounce-back” from one election to the next. In both cases (key groups on the left, a random sample of 10,000 individual-level estimates on the right), there was a substantial swing back.


Appendix: Notes on Statistical Methods


e published our preliminary estimates the week after the election, and since then have been collecting various new pieces of data to verify and update what we saw back then. Looking back on the preliminary data and analysis, we think that the estimates were very accurate in many cases and slightly off in some other cases. Substantively, our judgment is that the analysis holds up, with some important caveats.

Appendix: Decomposing Changes in Turnout and Vote Choice

The calculations described earlier to decompose the margin change are shown below. One complication is third party voting, which was higher in 2016 than other years. Different versions of these calculations were done, by excluding third party voters or projecting their two-party vote share. Results remained the same under these different sets of calculations.

Calculations for margin decomposition, 2012 to 2014 (top) and 2016 to 2018 (bottom).