No truth stands alone.

Pandemic Politics: Incumbent Governments Voting Intention

In our previous Pandemic Politics pieces, we demonstrated that incumbent governments who went to re-election during the pandemic did not do markedly better than at their last election, that the minor party vote was not lower (or significantly higher) at pandemic elections, and that any effects of the pandemic on the share of voters interested in voting for a minor party faded either by election day or by the end of the year. Here we will examine whether or not incumbent governments received a boost during the COVID-19 pandemic, and if so, whether they are still benefiting from an improved vote compared to before the pandemic.

Although we’ve previously established that the average incumbent government did not receive a swing to it in pandemic elections, we also noted that it might be the case that incumbents were on track to do worse if not for the pandemic (hypothesis 4). To examine whether or not this is true, we have to examine public opinion polls to analyse the trends in whether or not people intended to vote for the governing party (referred to as voting intention) before and during the pandemic.

As I’ve noted before, using opinion polls as opposed to election results introduces several potential issues into our analysis (click/tap each for a full explanation):

  1. This is, broadly speaking, the many variations of polling error which many people have become familar with in recent years.

    Polling error can have many sources, ranging from sampling error (randomly finding more/less voters for party X in a sample than in the population), unrepresentative samples stemming from improper or failing to weight data by relevant demographics, to more prosaic issues such as pollster herding (where there are abnormally few outlier polls) and non-response bias (where people who don’t respond to polls voting significantly differently from those who do despite sharing common demographic characteristics).

    Unlike the elections we looked at in our previous pieces, polling is not as comprehensive a tool in assessing public opinion. Despite that, it still remains a very useful tool for discerning trends in public opinion (and in this case, for assessing the impact of a pandemic on electoral politics) – even if polls in one country are off, if we see that most incumbent governments around the world have gained in polls a month into the pandemic, we can conclude that the pandemic likely boosted voting intention for incumbent governments, even if we don’t know the exact increase.
  2. Generally speaking, countries with regular polling of voting intention tend to be disproportionately Western, educated, and more populous than democracies as a whole.

    Such countries are more likely to have ideology-based parties (where a party is organised around a core ideology e.g. progressivism, conservatism, liberalism etc) than personality-based parties (where a political party is primarily organised around supporting an individual’s aims). Furthermore, they are also significantly more likely to use a proportional-representation electoral system instead of single-winner electoral systems, meaning that the vote for any one party is likely to be lower in our sample than it might be if we could poll every democracy.
  3. Where we were able to match up pre-pandemic-election polling to an election result, polling of incumbents didn’t seem to do very well in the pandemic, with an average error of about 2.7% thanks to massive whoppers like South Korea (6.6%) and Croatia (10.6%). At the same time, there doesn’t seem to be a systematic skew towards or against incumbents during the pandemic (8 of 19 under-estimated the incumbent party); hence we can still analyse a large enough sample of polls to see how voting intention has changed during the pandemic (even if we’re not confident in what the exact voting intention is).

    Polling during the pandemic may produce systematic non-response biases by making it more likely that some voters respond to polling (e.g. voters whose jobs allow them to work from home may be overrepresented in landline polling) than others. In particular, a hypothesis about the cause of polling error in the 2020 US elections suggests that polling error and coronavirus cases may have been correlated, which would suggest that the pandemic may have made it harder for pollsters to contact a representative sample.
  4. In some ways, this is a subset of point 1, but it’s worth pointing out that polling too far out from an election may not be accurate.

    For example, when attempting to forecast the final result, fundamentals models (which don’t poll voters but rather attempt to model how people will vote given the state the country is in e.g. economy, peace/war, crime etc) have been shown to be more accurate than polls up until a few weeks before the election. This is likely because polls taken too far out from an election tend to have a larger subset of their sample who haven’t thought much about politics and thus whose opinions on the party they intend to vote for may shift rapidly once they tune into the election campaign. Furthermore, there are many events which can cause a temporary bounce in polling which fails to translate to a markedly higher vote on Election Day (in the USA, convention bounces, terrorist attacks etc); pandemics may be one such event.

    Even in a country whose government is widely recognised as having successfully contained COVID-19 and whose 2020 re-election is often credited to said containment (New Zealand), some reversion to the mean in their polling from the massive highs seen in the early months of the pandemic is fairly evident.

Despite these issues, polls are able to provide a snapshot of public opinion during certain periods of time which may not be covered by elections. Hence, we can use polls to perform a direct, apples-to-apples comparison of voting intention between different time periods (e.g. before the pandemic and during the pandemic), whereas elections are usually held at different times and on different schedules, making direct comparison harder. To determine which countries’ polling to use, we used the following criteria:

  • Elections are generally agreed to be fair and impartial, and held in nations with a Democracy Index >= 6 as of 2020
  • Opposition or other significant parties (won >= 5% of vote at last fair election) did not boycott election
  • Need to be able to determine voting intention from given polling data
  • Polling data had to be available for the relevant periods under investigation
    • At minimum, there had to be polling data from Nov/Dec 2019, polling from Apr 2020 as well as polling from Apr 2021 [if no 2020 election] / final pre-election polling [if the country held an election in 2020]
  • Must be polling for a general election and not a by-election
  • Where possible, legislative elections were used instead of executive elections.
  • Where there were separate constituency and proportional polls conducted simultaneously, the results of the proportional election were used if possible.
  • Must be reasonably able to determine incumbent government.
  • The identity of the governing party or coalition must not have changed since a government was first formed after the last election. Exceptions are made for governing coalitions where one minor party was swapped out for another (e.g. if a government consisted of parties A (35%), B (10%) and C (8%), but party C was swapped out for party D, that would still be included in our analysis).
  • Where government was formed through a loose coalition (as opposed to a formal, long-term electoral alliance, e.g. the Liberal-National Coalition in Australia), the governing party whose vote was used for the analysis was determined as follows:
    1. If applicable, the voting intention of the party which provided the head of the executive branch or head of state
    2. If no such party was in the governing coalition, then the voting intention of the party which had the most seats in the legislative branch in which government was formed (e.g. the House of Commons in the UK)
    3. If there was a tie, it was resolved in favour of the party with the most votes.
  • Countries with elections held during or before Apr 2020 were excluded for the analysis of pandemic elections.

The pandemic initially boosted voting intention for incumbents around the world

(if you’re on a mobile device, scroll right for full data or turn your device landscape. Click the Previous and Next buttons to view all data.)

Nation/subdivisionSwingEnd 20191 month into pandemicRegionPandemic election?
South Korea4.73741.7AsiaYes
Dominican Republic-1.93129.1North AmericaYes
New Brunswick-43733North AmericaYes
British Columbia73542North AmericaYes
New Zealand163955OceaniaYes
Saskatchewan-3.15551.9North AmericaYes
Jamaica15657South AmericaYes
United States of America14041North AmericaYes
Newfoundland and Labrador184260North AmericaYes
Canada7.831.539.3North AmericaNo
United Kingdom7.34451.3EuropeNo
Czech Republic3.431.635EuropeNo
Brazil-5.73226.3South AmericaNo

On average, incumbents gained about 4.7% during the first month of the pandemic, with 30 of 39 governments seeing an improvement in their voting-intention figures compared to their polling at the end of 2019. You can pretty clearly observe how voting intention for most incumbent parties went up below:

Histogram of changes in incumbent government voting intention 1 month into the pandemic
Histogram of changes in incumbent government voting intention 1 month into the pandemic (April 2020), along with a Gosset’s t-distribution fitted using maximum-likelihood estimation. Data was combined from nations who held an election during the pandemic and those which did not as there isn’t much difference between the two on this metric. All data collated available here.

The shift in voting intention is highly statistically significant (p < 0.0001), meaning that it is almost certain that there was a shift towards support for incumbent governments in the first month of the pandemic. This is consistent with research suggesting that during times of crisis, a population will tend to rally around the incumbent government (also known as the rally round the flag effect) for at least a short period of time. Most governments received a similar boost in the share of people intending to vote for them, with very little correlation to the COVID-19 caseload of the country:

Scatterplot of COVID-19 caseload vs change in incumbent govt polling
Scatterplot of COVID-19 caseload versus change in incumbent government voting intention as of April 2020, along with a linear regression (in red) fitted using least-squares.

Interestingly, despite research suggesting that the rally-round-the-flag effect usually fades by election day (unless the event occurs very close to an election), incumbent governments seem to have actually made further gains by election day. While the pandemic bounce either partially or completely faded by election day for six governments (New Zealand, Croatia, Jamaica, Netherlands, Newfoundland and Bulgaria), another nine governments saw a further boost in their voting-intention figures by election day:

For incumbents who went to re-election during the pandemic, the pandemic bounce mostly stayed in place up until election day

(if you’re on a mobile device, scroll right for full data or turn your device landscape. Click the Previous and Next buttons to view all data.)

Nation/subdivisionSwing1 month into pandemicFinal pollsPolling errorElection date
South Korea1.641.743.36.62020-04-15
Dominican Republic-1.929.127.24.32020-07-05
New Brunswick433372.32020-09-14
New Zealand-9.15545.94.12020-10-17
British Columbia64248-0.32020-10-24
United States of America1.54142.55.22020-11-03
Newfoundland and Labrador-12.46047.60.62021-03-25

When combined with our previous findings of minimal overall differences between governments’ vote in pandemic elections and their last election result, the fact that the pandemic bounce mostly remained in place up until election day seems to suggest that incumbents who went to re-election during the pandemic were for some reason on track to do worse than their previous election result. I initially thought that this might be because governments who went to re-election earlier gained a little on the pandemic bounce in April 2020, while governments who went to re-election later saw their pandemic bounce fade. However, comparing the swing from April 2020 to election day against when each government ran for re-election, the relationship isn’t very strong:

Date of election vs change in incumbent govt polling since Apr 2020
Scatterplot of changes in the pandemic bounce in polling for incumbent governments who went to re-election during the pandemic, along with a fitted linear regression using least-squares. The relationship between the two is not statistically significant, p > 0.1.

While there is some relationship between when a government went to re-election and the change in their pandemic bounce, it seems to mostly be driven by the three governments in the bottom-right corner who went to re-election in 2021 and who also saw a big drop-off in their polling since April 2020 (Netherlands, Newfoundland and Labrador, and Bulgaria).Interestingly, most of the governments who improved on their early pandemic polling by election day mostly went to re-election in late 2020, which would run counter to the idea that rally-round-the-flag bounces are short-lived. This may be because voters were more interested in government management of COVID-19 for governments who had to go to re-election in 2020, while by 2021, voters became more interested in their government’s management of vaccine procurement and rollout (and/or other issues), or it might just be an artifact of the relatively small number of re-elections we have to work off (n = 18).

In addition, there was very little correlation between a country’s caseload and the pandemic bounce enjoyed by the incumbent government. As each government went to re-election at different times, we can’t simply compare the number of COVID-19 cases per population to the change in voting intention; while a government with 30 000 cases in May 2020 might be considered to have failed in containing the virus, one with 30 000 cases in May 2021 would likely be considered to have mostly succeeded in limiting the spread of COVID-19.

Hence, I’ve calculated the ratio between each country’s per-capita COVID-19 caseload (the number of cases per population) and the global per-capita COVID-19 caseload (how many COVID-19 cases have there been per population around the world) as a measure of how each country is performing relative to others in controlling COVID-19 caseload . As an example, assuming a population of 10 million people, the first government (30 000 cases in May 2020) would have a ratio of about 6.9 (30 000 cases per 10 million, divided by 432.1 cases per million worldwide) while the latter (30 000 in May 2021) would have a ratio of 0.15 (30 000 cases per 10 million, divided by 19 525 cases per million worldwide), demonstrating how this measurement quantifies how a government is doing “relative” to the rest of the world in terms of COVID-19 control.

Comparing this ratio to the size of each government’s pandemic bounce by election day, we can see that the relative quality of a government’s COVID-19 control had little impact on its pandemic bounce: Due to the near-exponential growth of COVID-19 in many countries around the world, it’s basically not feasible at all to plot COVID-19 caseload against anything, when the caseloads come from significantly different time periods.

Any attempt to analyse the data that way will run headfirst into the fact that some of those countries have > 50 000 COVID-19 cases per 1m population, while others have just a couple hundred cases per 1m population, due to the fact that the measurements come from different points in time. As a result any findings are likely to be badly confounded (or for that matter, due to) massive outliers on the COVID-19 caseload front.

Scatterplot of COVID-19 caseload ratio vs change in incumbent govt polling
Scatterplot of the pandemic bounce for incumbent governments who went to re-election during the pandemic versus the ratio of each country’s COVID-19 per-capita caseload to the global COVID-19 per-capita caseload, along with a fitted linear regression using least-squares. The relationship between the two is not statistically significant, p > 0.5.

Meanwhile, in countries which did not go to an election during the pandemic, the pandemic bounce initially enjoyed by incumbents has pretty much completely faded. Comparing the polling for incumbents in April 2021 with their polling before the COVID-19 pandemic, we can see that voting-intention for incumbents has broadly returned to where it was at the end of 2019:

The pandemic bounce has mostly faded for incumbents

(if you’re on a mobile device, scroll right for full data or turn your device landscape. Click the Previous and Next buttons to view all data.)

Nation/subdivisionSwingEnd 2019Apr 2021Region
Canada3.531.535North America
Brazil-0.53231.5South America
United Kingdom-0.74443.3Europe
Czech Republic-7.731.623.9Europe

While there is quite a bit of variation in how each government is travelling compared to before the pandemic, on average the pandemic bounce has completely faded:

Histogram of changes in incumbent government voting intention 1 year into the pandemic
Histogram of changes in incumbent government voting intention 1 year into the pandemic (April 2021), along with a Gosset’s t-distribution fitted using maximum-likelihood estimation.

Another way of thinking about this is to measure how many countries’ incumbents saw a move in the opposite direction to what happened in the first month in the pandemic (i.e. reversion). For example, in Sweden, the governing Swedish Social Democrats were polling at about 25.6% at the end of 2019, a figure which rose to 31.9% in the first month of the pandemic, then dropped to 26.9% by April 2021. This would be counted as a incumbent whose polling reverted; while incumbents who either saw sustained increase/decline or stagnation throughout the periods we test would be considered to not have reverted. Of the 22 countries we have adequate polling data for, just two (Australia and France, where the incumbents are doing worse now than they were at the end of 2019) did not see any form of reversion; strongly suggesting that incumbent voting intention has at least partially reverted to where it was pre-pandemic.

Part of this reversion seems to be due to sharp declines in incumbent polling in countries where COVID-19 has been mismanaged. While it isn’t statistically significant due to a small sample size, there appears to be a weakly negative correlation between the COVID-19 caseload of a country and its incumbent government’s polling: Since all the COVID-19 caseload data in this case came from the same time period (Apr 2021), we can make direct comparisons without having to calculate the ratio figures we used earlier (doing so would not change the relationship as it would be dividing all numbers by the same global per-capita caseload).

Scatterplot of COVID-19 caseload vs change in incumbent govt polling (Apr 2021)
Scatterplot of COVID-19 caseload versus change in incumbent government voting intention as of April 2021, along with a linear regression (in red) fitted using least-squares. The relationship is weakly significant (p = 0.08, R2 = 0.103) but does not meet the cutoff of p < 0.05 for statistical significance.

Some of the correlation between COVID-19 caseload and the shift in voting-intention seems to be driven by the outlier to the bottom right (Czech Republic), which had both a high COVID-19 caseload and a big drop-off in the incumbent ANO 2011 government’s polling. There’s an argument to be made that this point should actually be excluded from the analysis, as the drop in voting intention for ANO 2011 there seems to be partly due to the sudden rise of the new Piráti a STAN and SPOLU coalitions in late 2020.

However, a cursory look at the data seems to suggest that there was no sharp drop-off in voting intention for ANO 2011 upon the founding of both coalitions. Instead, it looks more like a steady drop-off in voting intention for ANO 2011 ever since early 2020; hence I’m fairly comfortable with including it in the analysis anyhow.

After all, if people weren’t already dissatisfied with the government in some way, rearranging a few chairs on the deck of the Opposition isn’t exactly going to produce a drop-off in government support that large.
However, a more likely explanation is that governments are indeed being punished for high COVID-19 caseloads, at least in polls if not necessarily in elections but the small sample size and the high degree of noise (after all, we are analysing the politics of many different nations, each with their own local factors plus the various sources of error inherent to polling) renders statistical significance hard to achieve. If this is the case, we would expect that governments going to re-election within the year would experience swings to or against them based on how well they have managed the COVID-19 situation in their country.

However, an alternative explanation might be that unless the crisis is fairly close to the election (as in those incumbents who went to re-election in 2020), any bounce/drop-off in polling is completely illusory. You can actually run further with this, and argue that in some of the countries which sustained a polling bounce to election day in our above analysis, the incumbent was due for a boost anyway.

For example, in Queensland, the state Labor government was likely to see some improvement in their voting-intention figures anyway as state governments run by a different party to the Australian federal government usually do fairly well. Similarly, in New Zealand, where the Labour government won a majority for the first time since the introduction of proportional-representation, the Prime Minister’s party historically tends to gain a swing to them at their first run, meaning that at least part of the monster swing to Labour there should have been expected regardless of COVID-19.

It can be easy to overemphasise this point though, as such “fundamentals” models can easily be overfit to past data without adequate consideration to whether they actually predict future electoral performances.
For example, a respondent who thinks the government has mismanaged the COVID-19 situation may tell pollsters that they intend to vote for someone else, but change their mind closer to election day for any number of reasons (e.g. new set of issues to consider, govt success on some other issue, negative partisanship a la “I can’t stand the other lot getting into power”). If that is the case, then it’s unlikely that future elections will be in any way correlated with COVID-19 caseload – while there might be the odd country where a government manages to sell itself on its COVID-19 management or where the opposition successfully beats the government with the country’s high COVID-19 toll, the overall correlation will be weak and mostly dominated by country-specific factors. Only time, and the voters of each country, will tell.

As the data above demonstrates, there was indeed a rally-round-the-flag effect early in the pandemic, with voters reporting that they were more likely to vote for the governing party in response to the crisis. While governments who went to re-election during the pandemic have mostly benefited from this, the polling bounce for all other governments seem to have mostly faded. In part, this is probably due to simple reversion to pre-pandemic politics (even in countries with a low COVID-19 caseload, the average difference from pre-pandemic voting-intention is just 2.4%, just half of the initial pandemic bounce), but some of the drop-off also likely stems from voter dissatisfaction with high COVID-19 caseloads. With the pandemic bounce having effectively disappeared, governments around the world should reconsider attempts to push for early elections; those who do so anyway may find that they don’t receive the big swings they were hoping for.

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)