No truth stands alone.

Pandemic Politics: Incumbent Governments

This is the second part of our Pandemic Politics series (parts 1, 3, 4), in which we explore the effects of a pandemic in coronavirus elections. Here, we examine the received wisdom that the COVID-19 pandemic has been good for incumbent governments I should note that analysis in that last article is of appallingly low quality. They mix by-elections (which have very different electoral conditions from regular elections) with full-scale elections, then twist their data into more shapes than balloon twisters to fit their preferred conclusion.

This ranges from using terms such as “86% (of electorates who held elections during the pandemic) have been retained by incumbents” (ignoring the fact that incumbents tend to win re-election anyway – e.g. in the US 7/10 post-war Presidents who have sought re-election have won), to using percentages instead of frequencies to make effects sound larger than they really are: “56% of them received a swing towards them” translates to 16/29 incumbents receiving a swing to them, just 1 shy of 50-50 or pure chance.

In their final attempt to waterboard the data into showing pandemic incumbents are favoured to win, they calculate an average swing for incumbents, except they only calculate that average from incumbents who won re-election, which is kind of like concluding all religious people eventually become atheists by polling the American Atheists Convention.
, as well as examining if COVID-19 caseload has a significant impact on the swing in incumbent governments’ vote.

As before, a caveat: the elections held in 2020 may not be a perfectly representative sample of all democracies. There are few national legislative elections in Europe in 2020, for example. Additionally, each electorate will have its own political peculiarities which affects the outcome; hence this should be taken as a ‘baseline’ of how voter behaviour has changed during the pandemic, instead of a conclusive “this is how pandemics have affected incumbent governments”.

Hypotheses on the impact of COVID-19 on incumbent governments’ vote

A few potential hypotheses on what we might expect to find:

  1. No effect: COVID-19 may have been an issue in several elections, but on average the presence of COVID-19 (as well as the size of the COVID-19 caseload) has had relatively little effect, with each electorate’s political peculiarities instead taking precedence. (i.e. the null hypothesis)
  2. Positive: COVID-19 may have produced a “rally-around-the-flag” effect, where incumbents around the world get a boost from national crises as the population rallies around the government.
  3. Negative: On the other hand, COVID-19 may have exposed problems with healthcare systems around the world, producing a backlash against governing parties/coalitions as they struggle to deal with the first major pandemic in a decade.
  4. Antagonistic: In the elections held this year, the incumbent governments up for re-election may have been slated for a big swing to/against them, but COVID-19’s effects negated or reversed this swing. This can only be assessed using polling before/after the pandemic, which we plan to do in future Pandemic Politics pieces.

The criteria we used to decide which elections to include in this analysis:

  • Elections were generally agreed to be fair and impartial, and held in nations with a Democracy Index >= 6 (if no DI available, author judgement is used)
  • Election ended 11/March/2020 or later (the date WHO officially declared COVID19 a pandemic)
  • The governing party/coalition must be reasonably apparent.
  • The governing party/coalition remained relatively constant (i.e. if a component party switched coalitions, that election was excluded)
  • Opposition or other significant parties (won >= 10% of vote at last fair election, if no fair elections author judgement is used) did not boycott election
  • To minimise the impact of personality politics, legislative elections were used where possible, instead of executive elections.
  • Election has to be a general election of all seats in the legislature, not a by-election.
  • Where there were separate constituency and proportional elections held simultaneously, the results of the proportional election were used.

With all of the groundwork out of the way, let’s have a look at the data:

On average, the pandemic doesn’t appear to have had much impact on voters’ preferences for incumbent govts

(if you’re on a mobile device, scroll right for full data or turn your device landscape)

Nation/subdivisionRegionSwingPrev govt.2020 govt.DateSystem
South KoreaAsia7.8625.533.3604/15PR
SurinameSouth America-21.5245.4923.9705/25PR
Saint Kitts and NevisNorth America5.9248.9354.8506/05Plurality
AnguillaNorth America-19.3254.4435.1206/29Mixed
Dominican RepublicNorth America-9.3841.7932.4107/05PR
Trinidad and TobagoNorth America-2.651.6849.0808/10Plurality
Northern TerritoryOceania-2.7742.239.4308/22Majoritarian
JamaicaSouth America6.9950.0857.0709/03Plurality
New BrunswickNorth America7.4531.8939.3409/14Plurality
BermudaNorth America3.2158.8862.0910/01Plurality
Australian Capital TerritoryOceania2.648.751.310/17PR
New ZealandOceania13.1236.8950.0110/17PR
British ColumbiaNorth America7.4140.2947.710/24Plurality
SaskatchewanNorth America-1.4162.5361.1210/26Plurality
United States of AmericaNorth America0.8146.0946.911/03Plurality

From the data, it seems like hypotheses 2 and 3 (the pandemic having a positive/negative effect on incumbent governments’ vote) can be quite easily eliminated. On average, incumbent governments have done about as well during the pandemic as they have before it, with no evidence to suggest that the pandemic has boosted or hurt governments’ chances on net. For those worried about the effect of the outliers of Anguilla and Suriname on the average, the median swing is +0.8% – barely any different. From a statistical standpoint, a one-sample t-test on the swings gives a p-value of 0.7129, which means it’s very likely we would have gotten this data if the net effect of the pandemic was 0. Graphing the swings in incumbent government vote seen this year, we can see that the change looks pretty similar to what you might expect if the pandemic had no impact at all:

Histogram of incumbent govt swings in 2020 pandemic elections.
Histogram of incumbent govt swings in pandemic elections, along with a t-distribution fitted using maximum-likelihood. Note that there appears to be little shift or skew to either side.

Of course, the reason the average/median govt vote swing is basically 0 might be that voters reward governments who managed to contain the pandemic, while ousting governments who are perceived to have failed to do so. If this is the case, we would expect to see a swing against governments in electorates with high caseload, while voters in electorates with low caseload would swing towards their government; however, there is no strong evidence for this.

COVID-19 caseload versus incumbent government swing
Scatterplot of COVID-19 cases per 1m pop. v incumbent govt swing
Graph of COVID caseload per 1 million population versus incumbent govt vote swing, along with a linear regression fitted using ordinary least-squares. That massive outlier to the right is the USA.

As can be seen from a graph of incumbent govt swing vs COVID caseload, it appears that there doesn’t seem to be any strong correlation between a government’s performance on COVID-19 containment and its subsequent electoral performance. However, this may well have been an artifact of the massive outlier to the right (the USA), or an issue with electorates with small populations. We calculated COVID-19 caseload as COVID-19 cases per 1 million population. In smaller regions/nations, this may have artificially inflated the COVID-19 caseload, as 10 imported cases in a population of 10,000 would be the same as 1,000 community cases in a population of 1 million by this method, even though an area with just 10 cases is doing a pretty good job of containment while 1,000 cases probably indicates some kind of containment failure. If we exclude the USA and electorates with a population of less than 1 million, we get the following graph:

Scatterplot of COVID-19 cases per 1m pop. v incumbent govt swing, outliers and small-pop electorates excluded.
Graph of COVID caseload per 1 million population versus incumbent govt vote swing, along with a linear regression fitted using ordinary least-squares. Outlier (USA) and electorates with a population of less than 1 million excluded.

With those outliers excluded, there’s a much stronger negative correlation between COVID-19 caseload and the swing to/from incumbent governments. However, examining the graph more closely, most of the correlation seems to be driven by a few points in the low-caseload (< 1000 / mil) and high-caseload regions (> 7000 / mil), as well the exclusion of some low-population electorates which had both very low caseloads as well as massive swings against the incumbent government (Anguilla and Suriname). This is reflected in the accompanying statistical analysis for the model, which concludes that there is very weak evidence for the hypothesis that increased COVID-19 caseload damages incumbent government vote. p-value = 0.12; for reference the most commonly used p-value cutoffs require a p-value of less than 0.10 to be considered “weak evidence”, while p < 0.05 is usually used as the cutoff for statistical significance. This does not mean that increasing COVID-19 caseload has no impact at all on incumbent governments’ performances, nor does it mean that incumbent governments who did well electorally must have controlled COVID-19 well.

The most plausible interpretation of this data is probably that COVID-19 may have a negative impact on incumbent governments’ vote; however the effect is not as large as might be naïvely assumed, and more evidence is required to demonstrate this.

Implications for electoral democracy

One of the reasons why academics and most citizens believe some variant of participatory democracy produces the best outcomes for the nation (for now) is that democracy allows the populace to align a ruler’s incentives with that of the good of the nation, as broadly perceived by the populace. One would assume that, in a global pandemic, governments who have successfully kept COVID-19 caseloads low would be rewarded by voters, or at least that governments in nations with high COVID-19 caseloads would be ousted in landslides.

However, as the analysis above shows, evidence for the notion that high COVID-19 caseloads costs incumbents re-election is mixed at best. While it’s plausible that some governments have made gains thanks to their successful pandemic response, as the first graph of caseload v swing shows, a government can successfully keep COVID-19 caseloads low and get ousted by the electorate (Anguilla While Suriname also had a low COVID-19 caseload and experienced a massive swing against the incumbent govt, Suriname is experiencing an economic crisis which may have been a greater issue for voters.) while a government with one of the highest COVID-19 caseloads even as their neighbour successfully contained the virus nearly hangs on (USA).

Although there are definitely country-specific factors influencing each government’s vote swing, the inelasticity of such swings to their government’s objective performance on COVID-19 containment suggests that the incentives of democratically-elected governments may not be aligned with the good of the nation, reducing the effectiveness of democracy as a system of selecting governments. In particular, in many of the Western democracies which form the plurality of the data used here, due to hyper-partisanship and ideological polarisation, landslide elections have become less likely, which suggests voters are less responsive to once game-changing events such as a devastating pandemic. We don’t pretend to have any easy solutions to this, although some voting reforms such as proportional representation may help voters to express their honest choice.

Where to from here?

As the data above shows, there’s no evidence for the notion that the pandemic helped incumbent governments to stay in office across the world, and weak evidence for the theory that an effective COVID-19 response helped governments or that a poor response doomed them. However, as before, this does not inherently imply that COVID-19 had no effect at all on voter preference, or that an electorate will completely ignore a government’s COVID-19 response in its voting decisions.

While the data seems to fairly clearly refute hypotheses 2 and 3 (COVID-19 boosts/hurts incumbent governments respectively), it is entirely possible that a government was on track to lose or win much more narrowly in 2020 had it not been for COVID-19. A good example of this is New Zealand, whose major governing party (NZ Labour) was fairly close in opinion polls to its main rival, the National Party for most of 2019, only to surge dramatically once it became clear that the nation had successfully contained COVID-19. Although there is some very limited evidence for the hypothesis that in New Zealand, the Prime Minister’s party tends to gain a swing to them at their first run, so that should be priced in as a prior here. To examine hypothesis 4 (the idea that the govts up for re-election this year were slated for a big swing, but COVID-19’s electoral effects cancelled out that swing), we will be using polling to examine time-based changes in voting intention in future Pandemic Politics pieces for both minor parties and incumbent governments.

Add Your Comment

* Indicates Required Field

Your email address will not be published.



This site uses Akismet to reduce spam. Learn how your comment data is processed.

× Home Politics Projects

You appear to be using an outdated browser, for which this site is not optimised.

For your security, we strongly recommend you download a newer browser.

Although any of the latest browsers will do, we suggest the latest version of Firefox.

Download Firefox here

(Click on the button in the top-right to close this reminder)