Talking Turnout and The 2022 Midterms

The 2022 US Midterm general elections are upon us. Irrespective of the electoral context, turnout is significant. The greater the turnout the greater one is able to claim that the result is representatively legitimate. Amidst the mire of loud voices, partisanship and polarization, the purpose of this piece will be to analyse and evaluate past US midterm general election turnout rates with an eye to November 8th. Firstly, I will discuss how turnout is measured. This will serve the function of clarifying for the reader exactly what I mean when I say ‘turnout’ and the various mechanisms used to determine this. Subsequently, midterm turnout rates from 1986-2018 will provide the basis for a dataset that will undergo analysis. Engaging with this will allow us to use such data in order to examine probabilities for what may indeed happen this year on the national level. Once this has been achieved, a rough ‘ball-park’ prediction figure will be presented, explained and evaluated, prior to a discussion about early voting, turnout myths and some concluding thoughts.

Measuring Turnout

Before we begin to analyse the data, we must discuss how we actually measure turnout. There are two ‘operationalisations’ that are often employed when discussing and measuring turnout rates. The first concerns turnout as a percentage of Voting Age Population (VAP). Here, VAP measures the percentage of the population over the minimal legal voting age that turned out to vote. The second measures turnout as a percentage of the Voting Eligible Population (VEP), whereby the percentage of the legally eligible voting population that cast a ballot is recorded. Although VAP can be useful for grasping certain broad socio-political demographic changes in the population, VAP includes ineligible voters in its calculations, for instance those without full citizenship, felons and so on. The purpose of VAP is to simply provide an indicator of the share of the voting-age population that turn out to vote, revealing the share of legal adults that participate in electoral processes. As a result of such a broad-based inclusion of those who are ineligible to vote, VEP will be used here as the preferred measure of turnout, precisely because it does not entail such a limitation.[1]

Midterm Turnout in Data (1986-2018)

So, what can the data tell us? Firstly, we must recall that the 2022 general elections are midterms. This is significant as there is a clear distinction between turnout for General Elections in years with a presidential race as opposed to those without. From the 1986 elections to the present day, by looking at Figures 1a, 1b and 1c in the Appendix below, we can see that the highest turnout year for any midterm was in 2018, where turnout sat at 50% VEP. The lowest turnout across the same period in the years with presidential races, however, was in 1996, garnering 51.7% VEP. This perfectly shows that the turnout rates of midterm general elections are habitually lower that those which include presidential races, perhaps unsurprisingly. In fact, if we compare the average turnout for midterm general elections and those with presidential races since 1988, a 17.68-point difference emerges between them. This indicates that when analysing the turnout data of recent general elections, we must only include those of Midterm election years so to avoid negatively skewing the dataset.

Between 1986 and 2018, the national average turnout of US midterms sat at 40.8%. Across the nine midterm general elections in this period, the lowest recorded turnout was in 2014, achieving just 36.7% VEP. The highest was the succeeding Midterm election, at the mid-point of President Donald Trump’s administration in 2018, reaching the dizzying heights of 50% VEP. If we were to exclude 2018 turnout from our dataset, what we can observe is that average midterm turnout from 1986-2014 would sit just a little bit lower, at 39.65%, but with a standard deviation (σ) far smaller. Thus, in relation to the data for the period 1986-2014, 2018 turnout excelled far beyond +3σ. In fact, a 50% turnout rate would have returned a p-value threshold of 5x1010 at this point - a probability of 5 in 10 billion.[2] This is of course merely statistics alone and looking at data in this way cannot take into account the context of any election. This significantly explains why 2018 turnout was so high and broke through the ceiling of probability, due to its context. Importantly, it is significant to mention that 2018 could be a landmark year in the modern history of US midterm turnout. This election, in 2022, will determine whether or not the kind of turnout seen in 2018 was an anomaly, bound up with the polarizing context of the then president’s administration, or whether it is part of a new phenomenon that is here to stay.

What Are the Odds of That?

Given the 1986-2018 dataset, what are some probabilities for US midterm general election turnout in 2022?[3] Should we expect to call a handyman to re-plaster the probability ceiling again? Under a normal distribution curve, the probability that the 2022 turnout drops beneath the 2014 low sits at 12.89%[4]. That it will go beyond the 2018 record returns a probability of 56 in 10,000.[5] Thus, even if the 2018 turnout rate is overtaken, it should not be considered quite so much of an event as in 2018 precisely because the odds of achieving such a height were far longer. Nonetheless, in order for this to take place, the rate of turnout would have to safely break far beyond +2σ from the mean of the period. Subsequently, such a probability is far too low to statistically assume that this will happen, especially as we are yet to know if the 2018 turnout rate was an anomaly, simple outlier or the beginning of a new chapter in higher expectations.

If we know that we should not statistically assume that the 2022 turnout rate will be higher than 50%, where should we expect it to sit? As we can see from Figure 1b, although there is a positive correlation between turnout and time, this positive relationship is not deemed statistically significant enough to suppose that it would likely continue into the future.[6] Although lacking significance for a precise projection, there is still a moderate positive correlation. Equally, until the sharp volatility of 2014 and 2018, midterm turnout was rising with a highly statistically significant correlation and a strong coefficient of determination.[7] If we include 2018 to this mix, it would be safe therefore to infer that there is a strong prospect that turnout will rise above the 2010 level of 41.8%. Let’s taper this down a  bit. The likelihood that turnout will fall between the lowest (36.7%) and highest (50%) points in the dataset is a reassuring 86.55%.[8] If we narrow this even further, we can see that there is a 58.18% probability of the turnout rate sitting between 40-50%.[9] Making this aperture of estimation even tighter, 46.41% of the time the turnout rate falls between 40-45% and only 11.77% of the time does it rest between 45-50%.[10] Hence, statistically, we should assume the likelihood of the 2022 turnout rate sitting between 40-50%, with an emphasis on 40-45%, in keeping with broad-based trends.

Speculative Forecasting

Although the following is pure conjecture, resting on inclination, instinct and statistical inference, my personal estimate is that the 2022 turnout will return somewhere between 42-48%, of which there is roughly a little more than a 1 in 3 probability.[11] Overall, prediction is based on statistical likelihoods and prospects. As far as midterm turnout is concerned, the outlying turnout rates of 2014 and 2018 have made statistical prediction for this general election much more complicated. Nonetheless, this range (42-48%) would be statistically in keeping with the long-term trends in the data. Additionally, this does not assume that the turnout rate of 2018 will be lapsed, which we must assume is still an outlier until evidence to the contrary is presented, because of how far it breaks from past data in the set. However, taking this into consideration, such a ‘prediction’ could still be conceptually classified as somewhere between an ‘educated guess’ and statistical inference – there is always a chance that statistical probability is thrown to the wind, as with the past two midterms.

Early Voting

In fact, a number of outlets are already predicting that this is likely to be the case and the 2018 turnout matched.[12] There are a number of variables and factors that have been discussed as potentially effecting voter turnout at US general elections: the structure of electoral competition[13], ‘spectator’ participation in competitive environments[14], campaigning in swing seats[15], education[16], the effect of television[17], public administration of elections[18], and the effects of voter ID legislation and restrictions[19], to name but a handful. In this case, a much-discussed indicator of expected high turnout is early voting.

Early voting has been a talking point in the lead up to this election as, at the time of writing, it is thought that early voting for the 2022 Midterms has far surpassed prior elections, indicating that we should expect turnout to be high. On the morning of the 31st of October 2022, there have been reports of as many as 21,200,568 early votes cast, “of which 12,851,641 were mail-in ballots returned and 8,348,927 ballots cast in person so far. Last Monday, Oct. 24, the count was 8,018,219.”[20] Interestingly, according to one poll, 28% of registered voters said they were ‘extremely enthusiastic’ about voting in this year’s midterm election, just 2% lower than at the same time four years ago - the highest recorded - and 13% higher than in 2014 respectively - the lowest recorded.[21] This, at least in the immediacy of short-term correlation, clearly mirrors contemporary turnout trends rather closely. Therefore, if these polls are correct as a reliable indicator, we may see a turnout rate similar to 2018.

Looking at some specific states, we can see that there is a chance turnout will actually exceed 2018. For instance, in Ohio over 943,000 people either actually voted or requested absentee ballots in the initial week of early voting; an increase of 2.7% over that of the same period in 2018.[22] In Pennsylvania, more than 685,000 ballots had been cast through the mail as of Wednesday the 26th of October, which is four times more than the roughly 167,000 that were cast in 2018.[23] Similar scenarios are being reported in North Carolina, Georgia, Texas and Florida, to name but a few.

Nevertheless, before we consider this outcome a foregone conclusion, perhaps some doubt on this should be cast. As one reporter for The Wall Street Journal has pointed out, it would be a mistake to directly compare the early voting figures of 2022 with that of 2018 due to their differing cultural and statutory contexts.[24] Like a number of political and social phenomena, the effect of the COVID-19 Pandemic has been significant in adapting our norms and practices. Elections have not been exempted from this. In order to combat any anxieties related to potentially increasing the rate of COVID-19 infection, in 2020 23 states made changes to their statutory voting procedures in order to accommodate the epidemiological context, ultimately providing new avenues of electoral participation for over 66 million eligible voters – 84% of which would be able to vote by mail.[25] In most of these cases, this concerned mail-in or absentee voting. For many across the country, for the first time, democratic electoral participation became easier. Thus, on top of the already polarizing and competitive context of the election, widespread adaptations to state electoral law were ultimately a significant factor in adding to ‘concrete’ cultural practices and norms of democratic participation; manifesting through an increasing rate of turnout.[26]

Early voting may be an initial indicator of high turnout. Nonetheless, by comparing the early voting statistics of 2018 to those of 2022, one ignores that the normative culture of early voting has indeed adapted in the interim. We just may simply find that a greater percentage of the electorate have changed their electoral preferences.[27] In this, I am not necessarily suggesting that partisan preferences have adapted, i.e., who one prefers to vote for. Rather, I am suggesting that how one prefers to vote may have altered – one more instance of the so-called ‘new normal’. Consequently, we may simply see in-person voting on Election Day plummet if this is the case. In this scenario, early voting as an initial indicator of turnout becomes defunct, as new norms and practices set in. Nevertheless, this is the first post-pandemic general election in the US, and so we really will just have to wait and see.

Turnout Myths

There are some trends we can almost be certain that the pandemic has not changed one iota. As far as turnout is concerned, there is one point that I would like to make abundantly clear before I lay out some final thoughts. Many often articulate that: ‘If only more people had turned out, we would have won’. What is implied here is: ‘If only more people had turned out for us, we would have won’. This is an obvious assertion, but the latter formulation of this sentiment does not include an implicit conflation of a general rise in turnout with a rise in popularity for any given party. The former, however, does. This fable - that turnout on the national level is tied to the fortunes of any particular party (usually the Democrats) – has been debunked.

In their ‘The Turnout Myth’, Daron Shaw and John Petrocik explored this notion. On the whole, their aggregated data analysis clearly led them to conclude that: “within the range of turnout variation that we experience in this county, there is no systematic link between election outcomes and turnout levels”, statistically hammering home the claim that: “there is no axiomatic connection between the turnout rate and the partisan outcome of an election, although large majorities of scholars, media commentators, and average citizens believe otherwise”.[28] In this manner, an increase in national turnout should only broadly strengthen the democratic and representative legitimacy of the winning candidate(s). Simply, there are no statistically significant patterns displaying how variations in turnout benefit or limit either one of the mainstream political parties on the national stage. It would be prudent to keep this in mind.

Concluding Thoughts

So, all in all, the rate of turnout is significant for the 2022 US midterm general election. From the 1986-2018 dataset of national turnout at US Midterm general elections: (a) the average turnout rate sits at 40.8%; (b) this has a standard deviation of 3.6237; (c) there is not a statistically significant relationship with time, even though there is a moderate positive correlation, where R = 0.5451; (d) Since 2010, we have seen both the basement-level lows of 2014, returning 36.7%, and the altitudinal highs of 2018, returning 50%. Given this data, the probability that the 2022 rate will sit: (1) between the highest and lowest points of the dataset is 86.5%; (2) below the 2014 rate is 12.89%; (3) above the 2018 rate is 0.44%; and lastly (4) between the turnout rate of 2010 (41.8%) and 2018 is 38.57%.

Ultimately, at least as far as turnout is concerned, the 2022 Midterm will return one of four outcomes. Firstly, a thorough and cyclical volatile turnout rate could emerge as the dominant new paradigm. Here, if turnout drops significantly to around its 2014 level, then a certain trendless instability will be cemented as the new phenomenon of turnout, whereby the rate of turnout can be seen to have widely oscillated since 2010 with every passing midterm election.

Secondly, we could see the turnout rate return to its pre-2014 trends, falling between 40-45%. This would indicate that the 2014-2022 period was an anomalous blip in the long-term pattern, defined initially by voter apathy during the final Obama administration perhaps, and then by partisan or ideological polarisation in 2018. In both this and the previous scenario, the task of pollsters and political scientists alike would then become explaining how and why early voting was so high when turnout would have declined from 2018.

Thirdly, the turnout rate could land somewhere around the 2018 benchmark, which, as we have discussed, early voting data would suggest. In this scenario, far higher turnout rates become a normalised phenomenon, for whatever particular reason – be that political polarisation, education, administrative changes, and so on. 2018 would become less of an anomaly in this case and more of an inception point, marking the beginning of a period of higher turnout.

The last scenario is that the 2018 record could be broken. We should recall that this is statistically unlikely, occurring only 44 times in 10,000. This being said, if this does occur – and media hype concerning high early voting is not just sensationalism – the US can thoroughly be thought of as entering a new era of higher voter participation at Midterm elections, leaving us wondering just how high turnout can get. Nonetheless, this is a question for another day, given the right circumstances – naturally.

Therefore, to conclude, as discussed above, my personal estimation is that the second of these scenarios will be accurate. Given the statistical trends and probabilities discussed here, a safe judgement is to assume that turnout will fall from its 2018 level but still be far higher than in 2010, between 42-48%. This would be a safe estimation. Nevertheless, don’t hold me to that. Early voting is up, and previous turnout volatility makes long term trends less reliable and short-term trends only momentarily relevant, being so vastly distinct. Whatever the outcome, be it (a) falling to a new low, (b) falling but not so significantly, (c) continuing from the 2018 record, or (d) exceeding it, there will indeed be questions to ask.



Figure 1a – United States General Elections VEP Turnout Rate (%) (1986-2020)[29]

Figure 1b – United States Midterm General Election Turnout Rate (%) (1986-2018)[30]

Figure 1c – United States General Election Turnout Rate in Years with Presidential Races (%) (1988-2020)[31]

[1] Thomas Holbrook and Brianne Heidbreder (2010) ‘Does Measurement Matter? The Case of VAP and VEP in Models of Voter Turnout in the United States’, State Politics and Policy Quarterly, 10(2), pp. 157-179; Michael P. McDonald (2022) ‘What is the voting-age population (VAP) and the voting-eligible population (VEP)?’, United States Elections Project, denominator [Accessed 28th October 2022].

[2] μ = 39.65, σ = 1.69411, z = 6.1094.

[3] μ = 40.8, σ = 3.62369.

[4] P = 0.1289.

[5] P = 0.0056.

[6] R = 0.54509, P < 0.12908.

[7] R = 0.8539, r2 = 0.7291, P < 0.0145

[8] P = 0.8655.

[9] P = 0.5818.

[10] P = 0.4641, P = 0.1177.

[11] P = 0.3468.

[12] Harry Enten (October 20th 2022) ‘Why very high turnout is likely this midterm’. CNN Politics. (Accessed 31st October 2022).

[13] See: Daniel E. Bergan, Alan S. Gerber, Donald P. Green, and Costas Panagopoulos (2005) ‘Grassroots Mobilization and Voter Turnout in 2004.’ Public Opinion Quarterly, 69(5): 760-777; Costas Panagopoulos and Peter Francia (2009) ‘Grassroots Mobilization in the 2008 Presidential Election.’ Journal of Political Marketing, 8(4): 315-333; Costas Panagopoulos (2011) ‘Voter Turnout in the 2010 Congressional Midterm Elections’. PS: Political Science and Politics, 44(2): 317-319.

[14] Keena Lipsitz (2009) ‘The Consequences of Battleground and ‘Spectator’ State Residency or Political Participation’. Political Behaviour, 31(2): 187-209.

[15] Costas Panagopoulos (2009) ‘Campaign Dynamics in Battleground and Non-battleground States.’ Public Opinion Quarterly, 73(1): 119-130.

[16] Rachel Milstein Sondheimer and Donald P. Green (2010) ‘Using Experiments to Estimate the Effects of Education on Voter Turnout’. American Journal of Political Science, 54(1): 174–189.

[17] Matthew Gentzkow (2006) ‘Television and Voter Turnout’. The Quarterly Journal of Economics, 121(3): 931–972.

[18] Barry C. Burden and Jacob R. Neiheisel (2013) ‘Election Administration and the Pure Effect of Voter Registration on Turnout.’ Political Research Quarterly, 66(1): 77–90.

[19] Jason D. Mycoff, Michael W. Wagner, and David C. Wilson (2009) ‘The Empirical Effects of Voter-ID Laws: Present or Absent?’. PS: Political Science and Politics, 42(1): 121–126; Rene R. Rocha and Tetsuya Matsubayashi (2014) ‘The Politics of Race and Voter ID Laws in the States: The Return of Jim Crow?’. Political Research Quarterly, 67(3): 666–679; Rick Hassen (2020) Election Meltdown: Dirty Tricks, Distrust, and the Threat to American Democracy. New Haven, CT: Yale University Press, p. 41.

[20] Isabella Murray (October 31st 2022) “Midterm elections early voting updates: Turnout surpasses prior years”. ABC News. tory?id=92022629 [Accessed 31st October 2022].

[21] Harry Enten (October 20th 2022) ‘Why very high turnout is likely this midterm’. CNN Politics. https:// [Accessed 31st October 2022].

[22] Adam Gabbatt (23rd October 2022) “US midterm elections: early voting on track to match 2018 record”. The Guardian. [Accessed 31st October].

[24] Ibid.

[25] Kate Rabinowitz and Brittany Renee Mayes (September 25th 2020) “At least 84% of American voters can cast ballots by mail in the fall”. The Washington Post. 2020/politics/vote-by-mail-states/ [Accessed November 1st 2022].

[26] Michael P. McDonald (2022) From Pandemic to Insurrection: Voting in the 2020 US Presidential Election. Berlin and Boston, MA: De Gruyter.

[27] For data recording that Democrat voters make up a greater percentage of early voters than Republican voters, see: Michael P. McDonald (23rd November 2020) “2020 General Election Early Vote Statistics”. U.S. Elections Project. [Accessed November 1st 2022]; (31st October 2022) “2022 General Election Early Vote Statistics”. U.S. Elections Project. [Accessed 1st November 2022].

[28] Daron Shaw and John Petrocik (2020) The Turnout Myth: Voting Rates and Partisan Outcomes in American National Elections, Oxford: Oxford University Press, pp. 11, 4.

[29] Data taken from: Michael P. McDonald (2022) ‘National General Election VEP Turnout Rates, 1789-Present’. US Elections Project. [Accessed 28th October 2022]. 

[30] Ibid

[31] Ibid