Was COVID-19 a Game Changer for the Tokyo and Beijing Olympics?

Abstract

No, it was not. We do not find evidence that the COVID-19 pandemic affected the distribution of medals in a systematic way. Countries that suffered a higher share of deaths due to the pandemic, or that imposed stricter lockdowns, performed just as prior performances and standard economic and demographic variables alone would have predicted. Hence, while COVID-19 has surely affected the performance of some individual athletes, it did not have a perceptible effect on the distribution of medals across countries. The only impact of the pandemic on the outcome of the Games was to reduce the advantage that host countries traditionally have. This effect, probably caused by the lack of local spectators in most competitions, was particularly strong for teams’ events, where the “host effect” vanished.

Keywords

Olympics COVID-19 pandemic host effect

Introduction

In the months before the 2020 Summer Tokyo (postponed to 2021) and the 2022 Winter Beijing Games, numerous anecdotal accounts circulated in the press about athletes who could not prepare themselves appropriately because of COVID-19.¹ Indeed, some athletes were not allowed to participate because they contracted the virus just before the Games started; others could participate, but were affected by long-Covid symptoms. Furthermore, some athletes who never got the virus were also affected because they could not practice properly or participate in pre-Olympic competitions due to lockdown restrictions. In this paper, we provide a systematic analysis of whether the COVID-19 pandemic affected the aggregate performance of countries in those games.

Despite the presumption of a negative effect of the pandemic on athletes’ performance, the Olympics is a zero-sum game, which means some athletes’ or countries’ loss would become others’ gain. But did the countries most affected by COVID-19 perform worse than expected, benefiting countries where the consequences of the pandemic were lighter? We find that neither country-level direct measures of COVID-19 incidence nor proxies for lockdown stringency are systematically associated with less Olympic success. We investigate how countries’ COVID-19 incidence, proxied by a COVID-19 death ratio and countries’ lockdown stringency, affect country-level medal shares (top 3 and top 8 shares). Our results show that COVID-19 was not a game changer for the 2020 Tokyo and 2022 Beijing Games. COVID-19 incidence and lockdown stringency have no statistically significant impact on medal shares at the aggregate level or in the subcategories for individual, team, indoor, or outdoor games.

We do find, however, an impact on the host advantage during the Olympics. Historically, the country hosting the Games wins considerably more medals than predicted by standard economic and demographic variables. However, for the two Olympic Games during COVID-19, the host effect becomes significantly weaker. This happened primarily in teams’ events, for which the host effect virtually vanished, but is also observed in individual events. Presumably, this reflects the lack of local supporters, since most competitions happened either without any audience or with very restricted attendance.

Literature

The COVID-19 pandemic has generated a large body of research studying its various effects. Here we discuss the limited number of studies on the impact of COVID-19 on sports performance. See Singleton et al. (2022) and Szymanski (2003) for a review of the literature on commercial sports through the lens of economic theory.

Proxying productivity by the number of passes of a football player, Fischer et al. (2021) find that COVID-19 infection leads to a 6% temporary productivity drop and to a 5% drop after half a year. Even if the individual effects that they identify are also present in our context in some form, they may be too small to affect the distribution of medals at the country level.

Our paper is intrinsically connected to the literature that seeks to explain countries’ performance in Olympic Games. Existing studies have investigated the factors behind Olympic performance at the country level. Lowen et al. (2016) find that women’s empowerment campaign enhances a country’s female participation in Olympic Games. Noland and Stahler (2017) find that host advantage is more significant in judged contests, while country size and gross domestic product (GDP) per capita have smaller impacts on Olympic performance over time. Rewilak (2021), using data for Summer Olympic Games during 1996–2016, also finds evidence for a host advantage effect. Potts (2022) finds that athletes’ respect of rules and a lower corruption level lead to smaller numbers of disqualifications in Olympic Games. At the athlete level, Greenleaf et al. (2001), based on the interviews with past Olympians, identify the departure from normal routine as the most important factor that negatively affected their performance.

Among Olympic studies, a prominent example is Bernard and Busse (2004), who consider the role of population and economic resources in determining medal totals from 1960 to 1996 Summer Olympics Games. They show that both a large population and high per capita GDP generate high medal totals, and being the host country is a medal booster. We build on their setup to examine how COVID-related variables affected countries’ performance once we control for those factors. We depart from their approach by (a) considering both Summer and Winter Games; and (b) by considering only Games since 1992, to avoid dealing with the classification complications caused by transition economies before 1992.

On how COVID-19 affected the home advantage and host effect, two papers obtained results related to ours using data from soccer matches. Cueva (2020) finds that the home advantage is halved during pandemic and that referee bias against away teams disappears during lockdown. Bryson et al. (2021) find negative effects on the number of yellow cards issued to away teams. Considering indoor basketball games, De Angelis and Reade (2023) also find a smaller home advantage during the pandemic. We find that their results extend to the aggregate of sports included in Olympic Games.

Data

This section briefly describes the data sources and variable definitions. Please refer to Appendix I for details.

Based on data from the Olympic Games official website, we construct the following Olympic performance measures by country: the top 3 medal share (gold, silver, and bronze) and the top 8 share for each Olympic event. In our baseline regressions, we count the number of medals at the country level, assigning an equal weight to all medals. We use the share measure, instead of medal count, because it is not sensitive to the changing number of total medals over time. We include only countries that sent out at least one athlete to that specific Olympic Game. We use data since 1992, so we do not have the Soviet or planned economy dummies in the analysis. We explain in Appendix II how we deal with some of the complications related to the breakup or consolidation of countries, and so on.

Our COVID-19 death and lockdown stringency data come from the Oxford COVID database. Following the approach of much of the growing literature on the consequences of the pandemic (see, e.g., Liu et al., 2022), we construct the COVID-19 incidence variable based on the cumulative number of COVID-related deaths divided by population. The lockdown stringency index covers nine subindexes, including four lockdown measures (Shutdown) and another four mobility restriction measures (RestMob), as well as a measure for public information campaigns. We use either the combined index or split it into two subindexes (Shutdown and RestMob).

We consider both “long” and “short” definitions of COVID-19 cumulative incidence and lockdown stringency. The former considers the period from the onset of the pandemic until just before the Games started; the latter considers the 6 months before the Games. The original COVID data from the Oxford Covid database are at a daily frequency. We aggregate them to the monthly level by taking the average of the daily COVID variables over the relevant period of time (short period or long period).

Empirical Strategy

Our data is a panel by country over time, so a natural choice for our empirical strategy is a panel data estimation using the following specifications:

\begin{aligned} T o p 3_s h a r e_{i t} = & β_{0} + β_{1} C u m D C s h a r e_{i t} + β_{2} S t r i n g e n c y_{i t} \\ + β_{3} H o s t_{i t} + β_{4} H o s t_{i t} \times P a n d e m i c_{t} + Z_{γ} + a_{i} + a_{t} + e_{i t}, \end{aligned}

where the dependent variable Top3_share is a country’s share of gold, silver, and bronze medal count among all Olympics participating countries, measured in percentage (0–100); COVID-19 variables include its incidence (cumulative number of deaths normalized by population, CumCDshare) and the Stringency index (or the subindexes: Shutdown and RestMob); Host is the Olympics host country dummy; Pandemic is a dummy variable, equal to 1 for years 2020–2022 and 0 for earlier years;

Z

is a vector all other control variables such as population and GDP per capita;

a_{i}

and

a_{t}

are, respectively, country fixed effects and year dummies;

e_{i t}

is the error term. The interaction

Host \times Pandemic

assesses whether the host effect is still present when spectators’ access to the events is restricted during the pandemic.

The country fixed effects absorb much of the economic, cultural, and demographic factors that do not change or change slowly over time, and are associated with a country’s share of medals. A Hausman test rejects the null, indicating that the random effects estimator is inconsistent. Therefore, we adopt a specification with country fixed effects.²

Because many participating countries end up with no medals, the dependent variable ( $t o p 3_s h a r e$ ) has a large number of zero values. Although the zeros are not censored, Tobit has been used in the literature, as in Bernard and Busse (2004). However, since Tobit estimation does not allow for country fixed effects, in our baseline, we use a linear panel data estimation with country-fixed-effects and year dummies, reporting a Tobit random effects estimation as a robustness check.

Results

In Table 1, we show our baseline results. Observe, first, that population has the expected positive effect on Olympic performance. The impact of GDP per capita is estimated to be positive but not statistically different from zero, probably due to insufficient variation when we include country fixed effects. Since the effects of those variables are well known, we do not report them in the subsequent tables, although they are included in every regression.

Table 1.

OLS, all Games (Summer and Winter Games Combined).

	(1)	(2)	(3)	(4)
	All games	All games	All games	All games
	(long)	(long)	(long)	(long)
Log(Pop)	0.346**	0.332**	0.330*	0.319*
	(0.161)	(0.162)	(0.176)	(0.174)
Log(GDP/Pop)	0.075	0.088	0.088	0.095
	(0.117)	(0.113)	(0.114)	(0.115)
Host	2.847***	3.125***	3.123***	3.121***
	(0.611)	(0.642)	(0.644)	(0.644)
Host $\times$ Pandemic		$-$ 2.254*	$-$ 2.271*	$-$ 2.339*
		(1.336)	(1.340)	(1.282)
CumCDshare			$-$ 0.019	0.000
			(0.062)	(0.062)
Stringency			$-$ 0.001
			(0.004)
Shutdown				$-$ 0.058
				(0.043)
RestMob				0.051
				(0.043)
$P$ -value for $H_{0}$ :
Host + Host * Pandemic = 0		0.462	0.473	0.485
Country FEs	Yes	Yes	Yes	Yes
Year dummies	Yes	Yes	Yes	Yes
Observations	2,210	2,210	2,182	2,182
R-squared	0.777	0.778	0.778	0.778

Note. Dependent variable is $t o p 3_s h a r e$ , measured in percentage. Robust standard errors are in parentheses. OLS = ordinary least squares; GDP = gross domestic product; FE = fixed effect; Pop = population.

*** $p < .01$ , ** $p < .05$ , * $p < .1$ .

The host effect is positive, as usual. However, during the pandemic, its effect is reduced by more than 70%.³ Presumably, this may be caused by the absence of local fans in the competitions. The p-value reported in the table shows that the host effect during the pandemic is weak and statistically not significant at the 10% level. Interestingly, all COVID-related variables are highly insignificant in the regressions. It is not simply that standard errors are too high to allow for precise inference; the point estimates are very close to 0.

In Table 2, we split the sample between Summer and Winter Games. Columns 1 and 2 are the analogs of columns 3 and 4 of Table 1 for Summer Games, whereas columns 3 and 4 are their analogs for Winter Games. The null results from the baseline analysis are unchanged. In fact, even the $host \times pandemic$ interaction loses significance with the split in both subsamples. Considering that only two Olympic Games were held during the pandemic, splitting them into Summer and Winter Games means that the estimation of the coefficient relies on only one Olympics. As a result, the estimated coefficients may not carry as much statistical significance or predictive power on their own.⁴

Table 2.

OLS, Summer Versus Winter Games.

	(1)	(2)	(3)	(4)
	Summer	Summer	Winter	Winter
	(long)	(long)	(long)	(long)
Host	1.587***	1.588***	2.228***	2.215***
	(0.397)	(0.397)	(0.825)	(0.824)
Host $\times$ Pandemic	0.903	0.890	$-$ 0.975	$-$ 0.753
	(0.576)	(0.579)	(0.912)	(0.941)
CumCDshare	0.038	0.040	$-$ 0.012	0.021
	(0.039)	(0.039)	(0.062)	(0.053)
Stringency	0.001		$-$ 0.010
	(0.002)		(0.007)
Shutdown		$-$ 0.002		$-$ 0.184
		(0.019)		(0.115)
RestMob		0.006		0.085
		(0.019)		0.074
Country FEs	Yes	Yes	Yes	Yes
Year dummies	Yes	Yes	Yes	Yes
Observations	1,501	1,501	681	681
R-squared	0.949	0.949	0.907	0.907

Note. Dependent variable is $t o p 3_s h a r e$ , measured in percentage. Log(Pop) and Log(GDP/Pop) are included in all regressions. Robust standard errors are in parentheses. OLS = ordinary least squares; FE = fixed effect; Pop = population; GDP = gross domestic product.

*** $p < .01$ , ** $p < .05$ , * $p < .1$ .

In Table 3, we split the sample between individual and team competitions. Its structure is analogous to Table 2. It is plausible that the pandemic may have a negative effect on individual sports, for which replacement of top athletes may be more consequential than for team sports. Nevertheless, the null results from the baseline analysis remain unchanged. We do observe, however, that the negative $host \times pandemic$ effect is driven mainly by competitions involving teams. For them, the net host effect is statistically indistinguishable from zero.

Table 3.

OLS,Individual Versus Team Events.

	(1)	(2)	(3)	(4)
	Individual	Individual	Team	Team
	(long)	(long)	(long)	(long)
Host	2.816***	2.815***	4.044***	4.034***
	(0.755)	(0.755)	(0.793)	(0.792)
Host $\times$ Pandemic	$-$ 1.621	$-$ 1.676	$-$ 3.666***	$-$ 3.802***
	(1.385)	(1.338)	(1.136)	(1.014)
CumCDshare	$-$ 0.034	$-$ 0.017	$-$ 0.011	0.022
	(0.070)	(0.072)	(0.086)	(0.086)
Stringency	$-$ 0.001		$-$ 0.004
	(0.004)		(0.009)
Shutdown		$-$ 0.047		$-$ 0.149
		(0.049)		(0.113)
RestMob		0.044		0.082
		(0.051)		(0.092)
P-value for $H_{0}$ :
Host+Host*Pandemic = 0	0.311	0.310	0.654	0.730
Country FEs	Yes	Yes	Yes	Yes
Year dummies	Yes	Yes	Yes	Yes
Observations	2,105	2,105	1,376	1,376
R-squared	0.747	0.747	0.736	0.736

*** $p < .01$ , ** $p < .05$ , * $p < .1$ .

In Table 4, we redefine the pandemic variables to consider only the 6-month period prior to the beginning of the Games (short period). Column 1 shows the analog of column 4 of Table 1, while columns 2 and 3 split the sample between individual and teams’ sports, respectively. We confirm that the mitigation of the host effect is due to the teams’ competition. We also find that the restrictions related to the shutdown of activities in the previous 6 months show a negative effect on countries’ aggregate performance in the Games. Based on the result from the first regression in Table 4 and the summary statistics reported in Appendix III, the top 3 share would drop by about 0.6 percentage point for a country with the highest shutdown index of the sample (11.95 for Honduras, based on the short COVID-19 period), compared to the pre-COVID period. The corresponding estimate is only 0.13% when $S h u t d o w n_s h o r t$ increases by a standard deviation (2.53). Overall, the magnitude is rather small, even when it is statistically significant. Moreover, that effect is restricted to team competitions. It is not entirely clear why that is the case, but the explanation may rest on the simple fact that there are more athletes in a team sport than in individual sports. Even if the probability of an athlete getting affected by COVID-19 is independent of each other in a team, the probability of at least one athlete getting COVID-19 is higher. Considering that COVID-19 is highly contagious, the risk would be even higher when members of a team practice and play together. This could help explain why there is a stronger COVID-19 effect in team sports than in individual sports.

Table 4.

OLS, COVID Variables Based on the Short COVID-19 Period.

	(1)	(2)	(3)
	All games	Individual	Team
	(short)	(short)	(short)
Host	3.118***	2.811***	4.029***
	(0.644)	(0.755)	(0.791)
Host $\times$ Pandemic	$-$ 2.239*	$-$ 1.579	$-$ 3.593***
	(1.248)	(1.306)	(0.951)
CumCDshare	0.029	$-$ 0.019	0.135
	(0.112)	(0.128)	(0.190)
Shutdown	$-$ 0.051*	$-$ 0.040	$-$ 0.159**
	(0.028)	(0.033)	(0.065)
RestMob	0.024	0.015	0.055
	(0.030)	(0.035)	(0.065)
P-value for $H_{0}$ :
Host+Host*Pandemic = 0	0.416	0.254	0.444
Country FEs	Yes	Yes	Yes
Year dummies	Yes	Yes	Yes
Observations	2,182	2,105	1,376
R-squared	0.778	0.747	0.737

*** $p < .01$ , ** $p < .05$ , * $p < .1$ .

In Table 5, we show results from a Tobit estimation with random effects. The first three columns use the “long” definition of the pandemic variables, considering all events in column 1 and then splitting between individual and team sports in columns 2 and 3, respectively. The last three columns are the analogs for the “short” definition of the pandemic variables. The estimated $host \times pandemic$ effect is consistent with the fixed effects results. For the pandemic effects, we find no statistically significant effect. Even the negative effect of shutdown measures in the last 6 months for team events, which we obtain in the fixed effects analysis, is not confirmed in the Tobit analysis, indicating a lack of robustness.

Table 5.

Tobit, all Games (Summer and Winter Games Combined).

	(1)	(2)	(3)	(4)	(5)	(6)
	All	Individual	Team	All	Individual	Team
	(long)	(long)	(long)	(short)	(short)	(short)
Log(Pop)	1.015***	1.049***	1.624***	1.013***	1.047***	1.625***
	(0.084)	(0.091)	(0.199)	(0.084)	(0.091)	(0.199)
Log(GDP/Pop)	1.118***	1.141***	2.285***	1.121***	1.146***	2.286***
	(0.109)	(0.117)	(0.290)	(0.110)	(0.117)	(0.290)
Host	3.328***	3.059***	4.541***	3.323***	3.054***	4.530***
	(0.477)	(0.516)	(0.810)	(0.476)	(0.515)	(0.809)
Host $\times$ Pandemic	$-$ 2.855**	$-$ 2.199	$-$ 5.046**	$-$ 2.706**	$-$ 2.084	$-$ 4.613**
	(1.359)	(1.471)	(2.337)	(1.352)	(1.463)	(2.311)
CumCDshare	0.074	$-$ 0.000	0.131	0.059	$-$ 0.112	0.342
	(0.153)	(0.173)	(0.305)	(0.302)	(0.343)	(0.621)
Shutdown	$-$ 0.143	$-$ 0.076	$-$ 0.277	$-$ 0.115	$-$ 0.066	$-$ 0.267
	(0.122)	(0.135)	(0.275)	(0.075)	(0.083)	(0.173)
RestMob	0.024	$-$ 0.051	0.035	$-$ 0.006	$-$ 0.062	$-$ 0.025
	(0.138)	(0.154)	(0.288)	(0.090)	(0.100)	(0.196)
P-value for $H_{0}$ :
Host+Host*Pandemic = 0	0.711	0.534	0.818	0.626	0.480	0.969
Country random effects	Yes	Yes	Yes	Yes	Yes	Yes
Year dummies	Yes	Yes	Yes	Yes	Yes	Yes
Observations	2,182	2,105	1,376	2,182	2,105	1,376

Note. Dependent variable is $t o p 3_s h a r e$ , measured in percentage. Pop = population; GDP = gross domestic product.

*** $p < .01$ , ** $p < .05$ , * $p < .1$ .

In Appendix III, we show additional results. First, we split the sample for Summer Games between indoor and outdoor events (Table A2). Second, we consider the share of results among the top 8, instead of the top 3 (Table A3). In both cases, the results change very little, with the only notable difference being that the $host \times pandemic$ negative effect is present also for individual competitions when we consider the top 8 definition of country performance. Third, we assign different weights to gold, silver, and bronze (3, 2, and 1, respectively) and report the results in Table A4 based on the identical specifications of Table 1. Our main findings of the weak COVID-19 effect remain robust. Although the coefficient of $host \times pandemic$ interaction term becomes statistically insignificant, it still bears the expected negative sign.

A potential concern is that most countries in most Games receive no medals, so their top 3 shares are mostly zeroes, making the data rather skewed. When we drop those zero observations, our sample shrinks by almost two-thirds. The Host dummy remains significant and positive in the regressions, in line with the literature, but its interaction with the Pandemic dummy is no longer statistically significant. The results can be found in Table A5. The p-value reported in the table shows again that the host effect during the pandemic is weak and statistically insignificant at the 10% level.

Finally, to deal with outliers, we also tried dropping the observations with the largest $t o p 3_s h a r e$ . For example, if we drop the top 1% and 2% of countries with the most medals, our results change little. Table A6 reports the results from regressions without the top 2% observations, using otherwise the identical specifications as in Table 1. The COVID variables remain as before and the point estimate of the host effect decreases as substantially as before, although the coefficient of the interaction is not statistically significant. The p-value shows again that the host effect during the pandemic is statistically insignificant. The results from regressions without the top 1% are similar.

Conclusion

Societies in most countries value Olympic Games, as evidenced by the significant resources spent on them. Large global health shocks could affect the competitiveness of the games in important ways, distorting its results. Yet we show that, despite its wide and heterogeneous economic and social effects across countries, the COVID-19 pandemic was virtually inconsequential for the distribution of medals in the last two Olympic Games. The results are robust to various measures of the key variables and estimation techniques. We do find, however, that the host effect identified in the literature (which may be a distorting effect) virtually vanishes during the pandemic.

Our results show that team sports are more likely to be affected by the lack of spectators during the pandemic. Countries had adopted various policies during the pandemic to address this issue, such as physical distancing, face covering, health screening and monitoring, and contactless transactions. By implementing these strategies and prioritizing the health and safety of spectators, sporting events can be conducted more safely during a pandemic while still allowing fans to enjoy the excitement of live sports.

Although our findings suggest that COVID-19 was not a game changer for the Tokyo 2021 and Beijing 2022 Olympic Games, they should be interpreted as a short-run assessment. In contrast, its long-run impact remains to be seen. For instance, lockdown policies affect young children’s participation in sports. Some youth athletes may lose interest in sports altogether due to prolonged disruptions caused by lockdowns. Financial strain on families, decreased access to facilities, and changes in priorities may also contribute to a decline in participation rates. In addition, countries and families may invest more in outdoor sports that are relatively less affected by future outbreaks of contagious diseases. This is the first analysis of the impact of the Covid-19 pandemic on countries’ performance in Olympic Games. Future research may investigate the precise mechanisms behind our findings.

Footnotes

Acknowledgments

We would like to thank two anonymous referees for their insightful comments and suggestions. We would like to thank Yixuan Zhang for her excellent RA work. We also thank Weihong Li and other participants at the 2023 CES North America Conference for their discussions and comments.

Declaration of Conflicting Interests

The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The authors disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This project is supported by a fund for building world-class universities (disciplines) at Renmin University of China. Project No. KYGJC2023003.

ORCID iD

Huimin Shi

Notes

Appendix I: Data

The Olympic Games Data. The data come from its official website: https://olympics.com/en/. First, we convert the International Olympic Committee representative team codes into ISO-3 country codes to make them consistent across time. Second, we construct Olympic performance measures by country. Third, we distinguish individual events from team events. Fourth, for Summer Olympics, we split the events into indoor and outdoor ones to see if the indoor games were more severely affected by the pandemic. For Summer Olympics, we use data from the 1992, 1996, 2000, 2004, 2008, 2012, 2016, and 2020 (delayed to 2021) Games. For Winter Olympics, we use data from the 1992, 1994, 1998, 2002, 2006, 2010, 2014, 2018, and 2022 Games. We include countries that have sent out at least one athlete to that specific Olympic Game. The participation in Summer Olympic Games (200 countries/teams on average) is much more widespread than that of Winter Olympics (only 80 on average). Over time, some countries broke up or dissolved while others consolidated into a new, single country. Sometimes, due to sanctions and disciplinary actions, some countries are not allowed to send players, but their players may be able to compete for themselves under a generic Olympic team, such as Russian Olympic Players in 2021. We explain later in Online Appendix II how we deal with these and other special cases.

Macroeconomic and Demographic Data. The main source for macroeconomic and demographic variables is the World Bank’s World Development Indicators (WDI).⁵ We use data for real GDP and population. The real GDP uses 2015 as the basis year (constant 2015 USD). For a few countries with GDP data in current USD but without GDP deflators (mainly Liechtenstein), we use the US GDP deflators instead to calculate their GDP in 2015 constant USD. For other missing GDP data from the WDI database, we use the Penn World Table and the Maddison Project . We only use the GDP data measured in constant USD from these sources. When the base year is not 2015, we adjust them using the US GDP deflators. Considering the possible differences between the US GDP deflators and each country’s own national GDP deflators, we also run regressions using only the data from the WDI to check the robustness of the results. Since we only used the data since 1992, we do not have the Soviet or planned economy dummies in the analysis, as in previous papers on the topic. When an Olympic team includes athletes from several countries, such as the “unified team” for the Commonwealth of Independent States (CIS) in the 1992 Summer Games (which is comprised of the independent countries from the Former Soviet Union), we calculate these countries’ (CIS) total real GDP and population and treating them together as a single country. The Tokyo 2020 Game took place in 2021. We use the 2020 population and real GDP data for the games. Since the GDP and population data for 2022 are not yet available for most countries, we use the 2021 population and real GDP data for the 2022 Beijing Games. When there is missing population and real GDP information, we use the most-up-to-date WDI data.⁶

COVID-19 data. The COVID-19-related variables are obtained from the Oxford COVID-19 Government Response Tracker. We construct the COVID-19 incidence variable based on the cumulative number of Covid-related deaths divided by population. For the lockdown stringency, it ranges from 0 to 100. The larger the number, the stricter the lockdown policies are. There are nine subindexes under the overall lockdown stringency, including four measures of lockdown (Shutdown: closings of schools and universities, closings of workplaces, cancelation of public events, and limits on gatherings), and another four measures of mobility restrictions (RestMob: closing of public transport, orders to “shelter-in-place” and otherwise confine to the home, restrictions on internal movement between cities/regions, restrictions on international travel), as well as the ninth measure for the health system policy (presence of public info campaigns). We use the combined index, but also split it into two subindexes, differentiating between measures of lockdown (Shutdown) and measures that limit the movement of people (RestMob). We consider both a “long” and a “short” definition of COVID-19 cumulative incidence and lockdown stringency. The former considers the whole period until just before the Games started; the latter considers approximately the previous 6 months before the Games. More precisely, we proceed as follows. The Tokyo Summer Olympic Games started on July 23, 2021. The corresponding COVID-related deaths based on the “long” period are calculated as the cumulative deaths until July 15, 2021. The corresponding lockdown stringency measures the average stringency index between January 1, 2020 and July 15, 2021. The “short” definition corresponds to the period between January 1, 2021 and July 15, 2021, and the COVID-related deaths and stringency variables are defined in a similar way, but considering only that sorter period. For Winter Games, the Beijing Olympic Games began on April 1, 2022. The long period covers January 31, 2020–December 31, 2022, while the short period covers January 8, 2021–January 31, 2022. For each Game’s data, we include in our data only the countries that send out at least one athlete to any event. Nonparticipating countries are not in our sample.

Appendix II: Special country notes

Independent Olympic Athletes (IOA) and Refugee Olympic Team (ROT): Since neither IOA nor ROT has GDP or population data, they are not included in our sample for regressions. The number of medals they received is small, so ignoring them will not affect significantly our results. IOA won only one silver medal and two bronze medals in 1992s Summer Games. IOA won only one gold medal and one bronze medal in 2016 Summer Games. ROT participated in 2016 and 2020 Olympic Games but won no medal.

The unified team for the CIS: CIS, including the following 13 independent states from the Former Soviet Union, participated in the 1992 Summer Olympics Games in Barcelona, Spain: Russia Federation (RUS), Armenia, Azerbaijan, Kazakhstan (KAZ), Georgia, Kyrgyzstan, Moldova, Belarus (BLR), Tajikistan, Turkmenistan, Ukraine (UKR), and Uzbekistan (UZB).⁷ We use the WDI GDP (2015 constant USD) and population data for these countries in 1992 to calculate their total GDP and population. In the 1992 Winter Olympic Games in Albertville, France, the CIS was comprised of the following five countries: KAZ, RUS, BLR, UKR, and UZB. We again use the WDI GDP (2015 constant USD) and population data to calculate their total GDP and population.

Athletes from Russia (OAR) & Russian Olympic Committee (ROC)

Russia, as a country, was banned from the 2018 Pyeongchang Olympics Games. Athletes of the OAR participated in the games, representing themselves rather than Russia. In Tokyo 2020 and Beijing 2022 Games, Russian players formed ROC to participate in the games.⁸ For these three events, we assign their medals to Russia and use Russian GDP and population data in the corresponding years. To be consistent with the data in other years, we continue to use the same country ISO code, “RUS,” for Russia. Note that this treatment is only for analytical convenience, and it does not represent our opinion regarding Russia’s status during the games.

Federal Republic of Yugoslavia, Serbia, and Montenegro. On April 27, 1992, the Federal Republic of Yugoslavia was established. Later, it was dissolved and divided into several countries. See the following chart for its transitions and participation in the Olympics over time.⁹ From 1992 to 2006, we used the country code YUG. The real GDP data for 1992 is from the Madison database. For 1996–2006, we use the total GDP of Serbia and Montenegro. From 2008 onwards, we treat each independent state separately (MKD, SRB, KOS, and MNE) and used their GDP and population data.

Other teams with missing real GDP and/or population data from the WDI. Democratic People’s Republic of Korea and Chinese Taipei’s real GDP and population data are missing from the WDI database. We fill in the missing data using the Madison database for the former and Penn World Table data for the latter.

Appendix III: Additional tables

Table A6.

OLS, Dependent Variable = top3_share, Drop Outliers (i.e., Observations With LARGEST 2% top3_share).

	(1)	(2)	(3)	(4)
	all_games	all_games	all_games	all_games
Host	1.800***	1.993***	1.994***	1.995***
	(0.492)	(0.547)	(0.548)	(0.547)
Host*Pandemic		$-$ 0.960	$-$ 1.000	$-$ 1.064
		(1.093)	(1.128)	(1.104)
accCDshare_long			$-$ 0.030	$-$ 0.016
			(0.049)	(0.048)
stringency_long			0.002
			(0.003)
ShutDown_long				$-$ 0.026
				(0.038)
RestMob_long				0.051
				(0.039)
P-value for $H_{0}$ :
Host+Host*Pandemic = 0		0.275	0.313	0.331
Country FEs	Yes	Yes	Yes	Yes
Year dummies	Yes	Yes	Yes	Yes
Observations	2,165	2,165	2,137	2,137
R-squared	0.741	0.741	0.740	0.741

Note. Dependent variable is $t o p 3_s h a r e$ , measured in percentage. Observations with the largest 2% of top3_share (based on the sample used in the first regression of Table A1) are dropped. Log(Pop) and Log(GDP/Pop) are included in all of the regressions, but their coefficients are not shown to save space. Robust standard errors are in parentheses. OLS = ordinary least squares; FE = fixed effect; Pop = population; GDP = gross domestic product.

*** $p < .01$ , ** $p < .05$ , * $p < .1$ .

References

Bernard

A. B.

Busse

M. R.

(2004). Who wins the Olympic Games: Economic resources and medal totals. Review of Economics and Statistics, 86(1), 413–417. https://doi.org/10.1162/003465304774201824

Bryson

Dolton

Reade

J. J.

Schreyer

Singleton

(2021). Causal effects of an absent crowd on performances and refereeing decisions during COVID-19. Economics Letters, 198, 109664. https://doi.org/10.1016/j.econlet.2020.109664

Cueva

(2020). Animal spirits in the beautiful game. Testing social pressure in professional football during the COVID-19 lockdown. OSF Preprints Web. https://doi.org/10.31219/osf.io/hczkj

De Angelis

Reade

J. J.

(2023). Home advantage and mispricing in indoor sports’ ghost games: The case of European basketball. Annals of Operations Research, 325(1), 391–418. https://doi.org/10.1007/s10479-022-04950-7

Fischer

Reade

J. J.

Schmal

W. B.

(2021). The long shadow of an infection: Covid-19 and performance at work. Technical report, DICE Discussion Paper. https://www.econstor.eu/handle/10419/237290/

Greenleaf

Gould

Dieffenbach

(2001). Factors influencing Olympic performance: Interviews with Atlanta and Negano US Olympians. Journal of Applied Sport Psychology, 13(2), 154–184. https://doi.org/10.1080/104132001753149874

Liu

Ornelas

Shi

(2022). The trade impact of the COVID-19 pandemic. The World Economy, 45(12), 3751–3779. https://doi.org/10.1111/twec.13279

Lowen

Deaner

R. O.

Schmitt

(2016). Guys and gals going for gold: The role of women’s empowerment in Olympic success. Journal of Sports Economics, 17(3), 260–285. https://doi.org/10.1177/1527002514531791

Mundlak

(1978). On the pooling of time series and cross section data. Econometrica, 46(1), 69–85. https://doi.org/10.2307/1913646

10.

Noland

Stahler

(2017). An old boys club no more: Pluralism in participation and performance at the olympic games. Journal of Sports Economics, 18(5), 506–536. https://doi.org/10.1177/1527002515588138

11.

Potts

T. B.

(2022). Does it pay to play by the rules? Respect for rule of law, control of corruption, and national success at the Summer Olympics. Journal of Sports Economics, 23(2), 222–245. https://doi.org/10.1177/15270025211049787

12.

Rewilak

(2021). The (non) determinants of Olympic success. Journal of Sports Economics, 22(5), 546–570. https://doi.org/10.1177/1527002521992833

13.

Singleton

Bryson

Dolton

Reade

Schreyer

(2022). Economics lessons from sports during the COVID-19 pandemic. In P. M., Pedersen (Ed.), Research handbook on sport and COVID-19 (chapter 2, pp. 9–18). Edward Elgar Publishing.

14.

Szymanski

(2003). The economic design of sporting contests. Journal of Economic Literature, 41(4), 1137–1187. https://doi.org/ 10.1257/002205103771800004