Culture and contagion: Individualism and compliance with COVID-19 policy

Abstract

In the first wave of the pandemic, places where geographic mobility declined more rapidly saw fewer cases of COVID-19. And yet, there is significant variation in people’s compliance with the lockdown measures introduced by governments in order to curb the spread of the virus. In this paper, we show that much of this variation can be explained by different cultural traits. Specifically, we advance the hypothesis that individualism, which puts greater value on personal freedom, makes government intervention harder, whereas collectivism, which emphasises the wellbeing of the group, makes collective action easier. We find support for these ideas across 111 countries, but also when exploiting within country variation in the two largest economies in the world: China and the United States. Across a host of specifications, people were less abiding by the lockdown rules in places with greater prevalence of individualistic cultural traits. We conclude that cultural factors play a critical role in successful policy implementation.

1. Introduction

Populations around the world vary significantly in their psychological traits (Henrich, Boyd, Bowles, Camerer, Fehr, Gintis, McElreath, 2001, Nisbett, 2004, Nisbett, Peng, Choi, Norenzayan, 2001). For example, people from Western Europe and their cultural descendants in North America stand out as being more individualistic, less conformist and obedient, while exhibiting a higher degree of in-group loyalty and nepotism (Bond, Smith, 1996, Gächter, Schulz, 2016, Henrich, Boyd, Bowles, Camerer, Fehr, Gintis, McElreath, 2001, Herrmann, Thöni, Gächter, 2008, Schulz, Bahrami-Rad, Beauchamp, Henrich, 2019, Talhelm, Zhang, Oishi, Shimin, Duan, Lan, Kitayama, 2014). Among the many dimensions of cultural variation, cross-cultural psychologists see the individualism-collectivism distinction as the main divider (Heine, 2010). An important source of this variation relates to societies historical exposure to disease-causing pathogens.1 Indeed, geographies with greater pathogen prevalence have developed cultural norms that serve as buffers against pathogen transmission (Gangestad and Buss, 1993, Quinlan, 2007, Schaller, Murray, 2008), making them more collectivist, since deviance from those norms poses a contagion risk to the community (Fincher et al., 2008). Conformity, in contrast, reduces the risk of spreading the disease.2

Countries around the world have fared very differently during the COVID-19 pandemic. As is well-known, places where geographic mobility declined more rapidly have seen fewer cases (Glaeser et al., 2020). Yet despite this fact, there is wide cross-country heterogeneity in people’s compliance with the lockdown measures introduced by their governments. For instance, at the peak of the first wave of the pandemic in April 2020, when most national governments imposed strict stay-at-home requirements, time spent at home increased by forty percent in some countries, but just by around ten percent in others.3

Against this background, we examine the role of cross-cultural differences in explaining this variation. Our hypothesis goes as follows. Collectivism, which encourages conformity and sanctions deviant behaviour, makes collective action easier since individuals emphasize the wellbeing of the group to a greater degree. Individualism, on the other hand, which emphasises personal freedom and achievement, provides greater incentives for standing out as it rewards deviance (Gorodnichenko, Roland, 2011a, Gorodnichenko and Roland, 2017, Gorodnichenko, Roland, 2020). This makes collective action harder, such as coordinating a coherent response to a pandemic. Indeed, studies show that people are much less supportive of government interventions in more individualistic societies (Pitlik and Rode, 2016).

Consistent with our hypothesis, we find that places with a greater prevalence of individualistic cultural traits saw less compliance with the stay-at-home requirements their governments implemented. Our results are robust across a host of specifications across countries, but also when exploiting within-country variation in culture, policy and mobility in the two largest economies: the United States and China.4 Crucially, our analysis based on peoples self-reported ancestry in the United States allows us to exclude the impact of all factors except those operating through some form of intergenerational transmission. Consistent with our cross-country results, we find that the historic prevalence of individualistic cultural traits predicts lower levels of compliance with the lockdown measures taken across US counties today. Finally, because a society’s cultural traits could be influenced by sociopolitical and geopolitical factors, we further explore the relationship between culture and mobility in China. Despite China’s unique economic and political model, we find that our proxy for collectivism predicts more compliance with stay-at-home requirements across Chinese cities, reinforcing the external validity of our findings.

Our paper adds to a growing literature on compliance with public health policy during the COVID-19 pandemic. For example, Wright et al. (2020) argue that economic factors have played an important role in shaping peoples compliance with the lockdown measures introduced across the United States, showing that people in low-income areas were less abiding by those measures relative to their counterparts in high-income cities. Meanwhile, other papers have gone beyond economic factors and examined the role of culture and values in peoples compliance with COVID-19-related public health recommendations. In particular, Bazzi et al. (2020) argue that a higher incidence of individualistic cultural traits undermines collective action and hampered the response to the COVID-19 pandemic. However, their analysis is confined to the United States, which Hofstede (2001) shows is the most individualistic country in the world, making it a potential outlier. Finally, Bargain and Aminjonov (2020) provide evidence that within Europe, people complied more with mobility restrictions in regions characterised by higher trust, while Barrios et al. (2020) focus on the role of civic capital in determining compliance with COVID-19 policy. In Frey et al. (2020), we show that trust and civic capital are both closely linked to the prevalence of democratic institutions.5 In this paper, in contrast, we document that collectivism, which is prevalent in autocratic countries like China, but also in democratic countries like Taiwan and South Korea, has a strong independent relationship with people’s compliance with COVID-19 policy.

The remainder of this paper is structured as follows. Section 2 discusses our data and provides some descriptive statistics. Section 3 discusses our empirical methodology and presents the key findings from our country-level analysis. Section 4 zooms in on the United States, exploring how individualism shaped compliance with Covid-19 policy across counties. Section 5 turns to China, examining the link between policy compliance and cultural tightness. Finally, in Section 6, we provide some conclusions.

2. Data and descriptive statistics

For our country-level analysis, we use data on lockdown orders matched with daily data on geographic mobility across 111 countries, between February 15, 2020 and June 15, 2020.6 Our dependent variables are taken from Google’s Mobility Reports and captures daily changes in both time spent at home and patterns of geographic mobility, compared to the median value for the corresponding day of the week during the period between the 3rd of January and the 6th of February 2020 (Google, 2020). Our baseline policy variable—taken from the Oxford COVID-19 Government Response Tracker (OxCGRT)—is a dummy taking the value 1 if a government imposes a stay-at-home order by law (Hale et al., 2020).7 To capture differences in cultural traits across countries, we employ the widely used individualism-collectivism measure from Hofstede (2001). The individualism measure captures the extent to which people in a society believe that individuals are responsible for taking care of themselves as opposed to being loyal to a cohesive group. Countries ranking higher on the individualism scale value personal freedom, whereas people in lower-ranked countries value conformity. The measure is based on a broad array of survey questions and has been validated in numerous studies (see Gorodnichenko and Roland, 2011b).8

To shed some light on the relationship between individualism and compliance with COVID-19 policy, we calculate cross-country changes in time spent at home as well as in patterns of geographic mobility, before and after the implementation of any stay-at-home measures. We then estimate the differential changes in these variables in individualistic societies relative to others. Consistent with our hypothesis, we find people’s compliance with the stay-at-home requirements introduced by their governments to be lower in countries with individualistic cultural traits. In Figure B2 of the online appendix, we show that our estimated elasticities of time spent at home and the individualism-collectivism measure from Hofstede (2001) are negatively correlated (Panel A). Conversely, we note that our mobility elasticities are positively correlated with the prevalence of individualism (Panel B). While these relationships are by no means conclusive, they provide suggestive evidence that individualistic societies are less likely to abide by the lockdown rules. In Section 3, we further probe this relationship.

We next zoom in on the United States. For this part of our analysis, we exploit differences in past patterns of immigration to construct measures of individualism for 722 commuting zones across the mainland of the United States (see section 4.1). Using micro-level Census data on self-reported ancestry (Ruggles et al., 2020), we are able to characterise the cultural origins of US communities. Matched with county-level data from Google’s Mobility Reports and newly collected data on stay-at-home requirements across US states, we follow a large literature on cultural evolution (Heine, 2007, Henrich, 2017, Henrich, Boyd, Bowles, Camerer, Fehr, Gintis, McElreath, 2001, Schulz, Bahrami-Rad, Beauchamp, Henrich, 2019) in documenting how the historical prevelance of different cultural traits shape contemporary behaviour. Specifically, for each commuting zone, we calculate the composition of the population in terms of ancestry.9 Based on a commuting zone’s population composition, we obtain an estimate of the incidence of individualism by aggregating Hofstede’s measure from the ancestors’ country of origin.10 Fig. 1 shows a high degree of heterogeneity in the incidence of individualism across US counties (see section 4.1 for details on figure construction). In our analysis, we exploit this variation as well as the fact that different states imposed stay-at-home measures at different times. In Fig. B3 of the online appendix, we document a strong negative relationship between the stay-at-home elasticity to the lockdown measures taken and the prevalence of individualism across US commuting zones (Panel A). As expected, we also find a negative correlation between the elasticity of mobility and the prevalence of individualistic cultural traits (Panel B), which we investigate further in Section 4.11

Fig. 1.

Prevalence of individualistic cultural traits across US counties. This figure shows how each county is positioned in the total distribution of the individualism measure (which varies at the commuting zone-level). Darker shades correspond to a higher incidence of individualistic cultural traits (see section 4.1 for details on the construction of the figure). Sources: authors’ own calculations based on Hofstede (2001); Ruggles et al. (2020); David Dorn’s data page: https://www.ddorn.net/data.htm.

It must be noted, however, that China is not included in Google’s Mobility Reports, and thus excluded from the analysis above. To assess whether our findings are also applicable to the Chinese context, we use data on daily mobility provided by Baidu (China Data Lab, 2020).12 In addition, because lockdown measures across Chinese cities are missing in the OxCGRT database, we create a novel dataset on local lockdown orders across China, as outlined in the appendix. Moreover, since Hofstede’s measure of individualism is unavailable for China, we use cultural tightness from Gelfand et al. (2011) as an inverse proxy of individualism (i.e., collectivism).13 Cultural tightness is defined as “the degree to which a society is characterised by rules and norms and the extent to which people are punished or sanctioned when they deviate from these rules and norms” (Chua et al., 2019). Despite being conceptually different from collectivism, we find that across the countries for which data are available, the tightness measure is negatively correlated to Hofstede’s individualism measure.14 Fig. B4 of the online appendix shows that mobility was more elastic to lockdown measures at the local level in culturally tight Chinese provinces, as our hypothesis predicts.15

Thus, the descriptive evidence provided so far is consistent with our hypothesis that differences in cultural traits across and within countries help explain the heterogeneity in compliance with the lockdown measures introduced by national and local governments around the world. We next turn to exploring the robustness of the above described results in greater detail.

3. Individualism and policy compliance: a country-level analysis

To formally study the relationship between the mobility restrictions introduced by governments around the world and changes in actual mobility, we estimate OLS regressions at different levels of aggregation. We first turn to our cross-country analysis to assess whether our hypothesis broadly holds. To mitigate concerns about these estimates being driven by other country-specific factors, we next move on to studying within-country variation in culture, policy and mobility in the United States and China. By zooming in on these particular countries, we are able to explore whether our hypothesis are relevant across very different political systems, which might reward different types of behaviour. For example, the US Constitution, which emphasises individual freedom and liberty, would make a Chinese-style lockdown—involving involuntary isolation at quarantine centres—impossible.

3.1. Empirical strategy

In this part of our analysis, we exploit the variation in policy and mobility across countries and over time. Our regressions are based on the following specification:

$$M_{c,t,m}={\beta }_0+{\beta }_1{\Phi }_{c,t-1}+{\beta }_2[{\Phi }_{c,t-1}\times X_c]+{\beta }_3{\Theta }_{c,t}+u_{c,m}+u_t+{\epsilon }_{c,t,m}$$ (1)

where, $M_{c,t,m}$ is our mobility index in country $c$, at date $t$, in mobility category $m$. The variable ${\Phi }_{c,t-1}$ is our policy index, which is lagged by one day in order to ensure that changes in policy are known to the population.16 In addition, as discussed in greater detail in the appendix, we take advantage of the fact that Google provides mobility indexes for a host of categories:

$$m=\left\{\text{workplace,}\text{grocery,}\text{transit,}\text{retail}\text{and}\text{entertainment,}\text{residential,}\text{park}\right\}$$

For example, one concern is that mobility in parks could be systematically higher in countries with a larger number of parks, or a temperate climate. Thus, in Eq. (1) we include country-mobility category fixed effects, $u_{c,m}$, which allows us to purge our estimates from constant unobserved characteristics of each particular mobility category in a given country. Country fixed effects also allows us to exclude any bias arising from country-specific characteristics, including differences in the coverage and quality of the mobility data.

In Eq. (1), the coefficient of interest is ${\beta }_2$, which measures the differential impact of lockdown measures for countries characterised by different levels of individualism. We measure individualism based on either Hofstede’ scale or a measure of obedience (Inglehart et al., 2014), which should be highly correlated to collectivism, as it similarly encourages tradition and conformity.17 To ease interpretation, we normalise the two variables to have zero mean and unitary standard deviation.

Our specifications include the interaction between policy stringency ($X_c$) and i) the logarithm of real GDP per capita; ii) the number of cellphone subscriptions in the population; iii) a measure of the strength of democratic institutions; iv) a dummy for whether a country had experience in dealing with previous epidemics; and v) the log-number of daily confirmed cases. Such controls further mitigate concerns that systematic differences in economic development and institutions (Wright et al., 2020), as well as differential trajectories of the spread of the virus between individualistic and collectivistic countries, are driving our results.18 Finally, to account for the potential impact of complementary policies, we include measures of testing intensity and the extent of contact tracing in the vector ${\Theta }_{c,t}$.19

3.2. Country-level results

Before turning to our OLS results based on Eq. (1), we study whether the probability of implementing lockdown measures depend on cultural traits. To do so, we regress a dummy equal to 1 if stay-at-home requirements are imposed, on the log of confirmed COVID-19 cases and its interaction with our individualism or obedience measures. We use a probit model in which we control for country and date fixed effects, and we cluster errors at the country-level. The results are shown in Table B1 of the online appendix. As expected, the probability of implementing lockdowns depends on the number of infections in each country. However, we do not find evidence that governments in countries with strong individualistic cultural traits are less likely to lock down (columns 1–2). We also examine whether countries with high obedience scores are more likely to implement stay-at-home requirements (columns 3–4). Again, we find no evidence in support of this hypothesis.

Table 1 presents the results from estimating Eq. (1) on the time spent in residential places. On average, stay-at-home measures increased the time spent at home by roughly 7.5% over the sample period (column 1). We note that the increase in time spent at home due to restrictions on movement was muted in countries with stronger individualistic cultural traits, where it was approximately thirty percent lower (column 2). Concerns over omitted variables that might be correlated with cultural traits and mobility are alleviated by the fact that our results are robust to controlling for different levels of economic development, institutional differences, experience with past epidemics, and the number of confirmed COVID-19 cases (column 3). Regardless of the specification, the coefficients of interest are statistically significant.

Table 1.

Individualism and compliance across countries.

	(1)	(2)	(3)
	Residential	Residential	Residential
Stay at home index	7.590***	7.125***	2.648
	(0.775)	(0.910)	(3.683)
Stay at home index $\times$ individualism		-1.824***	-2.115*
		(0.630)	(1.111)
Observations	15,274	10,882	9127
R-squared	0.791	0.804	0.799
Country-mobility category FE	yes	yes	yes
Date FE	yes	yes	yes
Policy x controls	no	no	yes

This table presents OLS estimates from regressing time spent at home at the country-level on (the one day-lag of) a dummy taking value 1 if on a given day the government imposed restrictions to stay at home. Column 3 includes interactions between the dummy and: i) the logarithm of real GDP per capita; ii) the number of cellphone subscriptions in the total population; iii) a measure of democracy from the Polity Project, iv) a dummy for experience with previous epidemics, and; v) the log-number of confirmed COVID-19 cases. Errors are clustered at the country-level. The coefficients with ${}^{★}{}^{★}{}^{★}$ are significant at the 1% level, with ${}^{★}{}^{★}$ are significant at the 5% level, and with ${}^{★}$ are significant at the 10% level.

Table B2 of the online appendix presents the results from regressing mobility indexes, capturing the frequency of visits to a given place, on stay-at-home requirements. On average, stay-at-home measures reduced mobility by roughly 17% over the sample period (column 1). We note that the reduction in mobility was relatively muted in countries with stronger individualistic cultural traits, where declines in mobility were approximately twenty-five percent lower (column 2). Adding a full set of controls leaves our results substantially unchanged (column 3).

That said, the average effects reported in columns 1–3 hide a great deal of heterogeneity, not least since people move around for different reasons. To capture this, we break down mobility into essential (e.g. grocery shopping) and non-essential travel (e.g. walking in parks), based on the categories provided in Google's Mobility Reports. Unsurprisingly, the results of columns 4–9 of the table suggest that restrictions reduced mobility less for essential than for non-essential travel. In other words, cross-cultural differences in mobility are not primarily driven by people in individualistic countries fulfilling more basic needs, but their greater willingness to deviate from stay-at-home measures for non-essential purposes.

We also perform a battery of robustness checks. First, we replace the individualism measure from Hofstede with our obedience variable (see Table B4 of the online appendix). Consistent with our hypothesis, lockdown measures increased the time spent at home more in countries at one standard deviation above the average obedience score (column 1). The coefficient of the interaction term is again statistically significant across all specifications. Columns 2–4 in Table B4 also show that the elasticity of mobility to stay-at-home requirements is lower for essential than for non-essential travel. We note that the reduction in mobility is even larger in countries with higher levels of self-reported obedience.

Second, instead of coding the policy indicator as a dummy equal to 1 if a government requires people to stay at home, we use the original value of the index from OxCGRT.20 The results are very similar, suggesting that our parsimonious policy indicator, which is easier to interpret than a categorical variable, appears to capture all the relevant information. As a final robustness check, we also use a broader policy stringency index. This index, which is also taken from the OxCGRT data, is an average of eight sub-indicators. Besides stay-at-home requirements, which we have focused on, it also includes school, workplace and public transport closures, as well as bans on public events and size restrictions on gatherings (see Hale et al., 2020 for further details). Again, we obtain similar results, implying a high correlation among the sub-indicators.21

3.3. Pre-trends

To check for pre-trends, Fig. B7 of the online appendix presents the results of an event study regression. We find no evidence of significant differences in mobility before the announcement of the lockdown measures. At the time of implementation, however, the increase in time spent at home was significantly lower in individualistic countries relative to others, consistent with the regression results in Table 1.

Figure B7 shows that mobility in individualistic countries starts diverging from the others a few days before official implementation. The underlying reason is that in most countries, authorities recommended staying at home several days in advance of any mobility restrictions being put into place. Data provided by Cheng et al. (2020) allows us to measure the number of days between the recommendation and the implementation of the lockdown measures taken. We calculate that the lag between announcement and official implementation of the stay-at-home requirements variy between 1 and 4 days in our sample, which is consistent with the patterns presented in Fig. B7.

To make sure that this anticipation effect does not alter our results, in Table B3 of the online appendix, we include up to 4 days-lead values of the policy indicator. We note that these results are very similar to our baseline specification. Reassuringly, the presence of rather mild anticipation effects do not seem to substantially bias our baseline regressions.

4. Individualism and policy compliance in the United States

To further alleviate the concern that our country-level estimates are driven by country-specific factors, such as different institutions, we next exploit within-country variation in culture, policy and mobility across the United States. In the below, we discuss our approach and empirical findings.

4.1. Empirical strategy

For this part of our analysis, we exploit county-level mobility data as well as regional differences in the demographic composition of the population to construct a measure of individualism for 722 commuting zones (CZs) across the mainland United States. For simplicity, we refer to CZs as “cities”, whose boundaries are based on commuting patterns (Tolbert and Sizer, 1996).

Following Berger and Engzell (2019), we rely on micro data on ancestors’ country of origin from the American Community Survey 2014–2018, and map the county-level information to cities using the crosswalk from Autor and Dorn (2013).22 We then compute the shares of peoples ancestors’ country of origin (for which Hofstede’s individualism measure is provided) for each city. This allows us to infer the historic exposure of a population to individualistic cultural traits and exclude the impact of all factors except those operating through some form of intergenerational transmission.23 Formally, the city-specific measure of individualism, $In{d}_z$, is given by:

$$In{d}_z=\sum _{j\in z}{\sigma }_{j}IN{D}_j$$

where $z$ indexes the city, ${\sigma }_j$ is the share of the population with ancestors from country $j$, and $IN{D}_j$ is the individualism measure of country $j$.24 For robustness, we also construct a city-level measure of obedience, $Ob{e}_z$, following the same approach. Figure B5 of the online appendix provides a glance of the geography of the incidence of obedience in the United States, revealing a wide variation within and across states.

Our approach allows us exploit detailed lockdown and re-opening dates at the state-level to study how restrictions on movement impacts mobility across counties within states. Thus, in this setting, we are much less concerned about the possibility that economic, institutional or geographical differences might be driving our results. Specifically, our econometric analysis is based on the following specification:

$$M_{y,t,m}={\beta }_0+{\beta }_1{\Phi }_{s,t-1}+{\beta }_2[{\Phi }_{s,t-1}\times X_z]+u_{y,m}+u_t+{\epsilon }_{y,t,m}$$ (2)

where, $M_{y,t,m}$ is our mobility index in county $y$ (in city $z$ and state $s$), at date $t$, and in mobility category $m$. In equation (2), ${\Phi }_{s,t-1}$ represents the lockdown dates for each state $s$. The vector $X_z$ includes $In{d}_z$, and in some specifications, the log of real wage income, to control for the potential systematic correlation between income and cultural traits.25 The variables $u_t$ and $u_{y,m}$ represent year- and county-mobility category fixed effects, respectively.26

4.2. US regional results

Table 2 presents our results from estimating Eq. (2). On average, stay-at-home requirements increased time spent at home by 1.3 percent (column 1). The increase in time spent at home is around seventy percent lower in counties at one standard deviation above the mean individualism score (column 2). However, when we control for wage income in column 3, which is strongly correlated to the incidence of individualism (Gorodnichenko and Roland, 2011a), the interaction coefficient becomes insignificant at conventional levels but remains negative.

Table 2.

Individualism and compliance across the United States.

	(1)	(2)	(3)	(4)	(5)
	Residential	Residential	Residential	Residential	Residential
Stay at home dates (state-level)	1.295***	1.103***	0.773***
	(0.199)	(0.208)	(0.181)
Stay at home dates $\times$ individualism		-0.741***	-0.228	-0.746***
		(0.231)	(0.138)	(0.154)
Stay at home dates $\times$ obedience					0.477**
					(0.227)
Observations	139,009	139,009	139,009	138,847	138,847
R-squared	0.916	0.918	0.922	0.956	0.956
County FE	yes	yes	yes	yes	yes
Date FE	yes	yes	yes	yes	yes
State-date FE	no	no	no	yes	yes
Policy x log wages	no	no	yes	yes	yes

The table presents OLS estimates from regressing time spent at home at the county-level on (the one day-lag of) a dummy taking value 1 if on a given day, the government required people to stay at home. Columns 3, 4 and 5 include interactions between the dummy and the logarithm of wage income per capita. Columns 4 and 5 include state-date fixed effects. The individualism measure is obtained by summing up country-level measures weighted by the share of ancestors’ country of origin. The obedience measure is derived analogously. Errors are clustered at the state-level. The coefficients with ${}^{★}{}^{★}{}^{★}$ are significant at the 1% level, with ${}^{★}{}^{★}$ are significant at the 5% level, and with ${}^{★}$ are significant at the 10% level.

Column 4 includes state-date fixed effects, which absorb the potential confounding effect of state-specific unobserved events, such as changes in other types of regulation, the impact of weather conditions, and differences in the spread of COVID-19.27 Controlling for meteorological events might be important, as the mobility index takes the winter month of January as the reference period. As the weather improves over spring, this could lead to a decrease in time spent at home that is unrelated to mobility restrictions. Similarly, in states with a high incidence of individualistic cultural traits, fewer cases might have induced people to stay home less. Reassuringly, when including state-date fixed effects our results remain intact. Finally, column 5 shows that stay-at-home restrictions increased time spent at home even more in cities with a relatively high incidence of obedience. Thus, alternative measures of individualism (or in this case the inverse of it), point in the same direction: cultural traits are important in explaining why places have fared differentially during the pandemic.28

To further assess the robustness of our findings, Table B5 of the online appendix presents the results from estimating Eq. (2), using mobility changes as the dependent variable, rather than time spent at home. On average, stay-at-home requirements reduced mobility by 3 percent (column 1). Column 2 shows that the reduction was lower in cities with a higher incidence of individualistic cultural traits. We note that this relationship remains statistically significant also when we control for wage income (column 2). We further note that within the US, the reduction in mobility for essential and non-essential travel is roughly the same. However, all interaction coefficients are positive and significant at the 99% level. Finally, when we rely on measures of individualism based on people’s country of origin (birthplace), rather then ancestry, this delivers very similar results as shown in Table B6 of the online appendix.

5. Cultural tightness and policy compliance in China

In the sections above, we have relied on data from Google’s Mobility Reports to calculate changes in time spent at home as well as in patterns of geographic mobility. This dataset, however, does not include China. In the below, we examine whether cultural traits also play an important role in shaping COVID-19 policy compliance in the context of China’s unique economic and political model. By compiling a novel dataset for China, we are able to further assess the external validity of our findings.

5.1. Empirical strategy

For our analysis, we exploit policy variation across Chinese cities as well as province-level differences in cultural traits.29 As we do not have regional data on the incidence of individualism across China, we proxy collectivism with cultural tightness, following Chua et al. (2019). In Figure B6 of the online appendix, we show that tightness and individualism are inversely related (correlation = -0.38) in a cross-section of countries.

Our econometric analysis is based on the following specification:

$$M_{i,t}={\beta }_0+{\beta }_1{\Phi }_{i,t-1}+{\beta }_2[{\Phi }_{i,t-1}\times X_p]+u_i+u_{p,t}+{\epsilon }_{i,t}$$ (3)

where, $M_{i,t}$ is our mobility index in city $i$ (in province $p$), at date $t$. We rely on data provided by Baidu, which unlike the Google Mobility Reports does not have any breakdown by mobility categories (see the online appendix A.1). Thus, we focus on aggregate trends in mobility. Specifically, model (3) exploits variation in measured cultural tightness across Chinese provinces, which is included in $X_p$, interacted with the city-level policy, ${\Phi }_{i,t-1}$. The vector $X_p$ includes the log of city-level GDP per capita in some specifications to control for the potential systematic correlation between income and cultural incidence (Gorodnichenko and Roland, 2011a). We also include province-date fixed effects, $u_{p,t}$, to control for residual factors correlated to mobility and culture that might bias our coefficients, such as differences in the spread of the virus across provinces. Given that there are only 29 provinces, we chose not to cluster errors at the province-level.

5.2. Chinese city-level results

Table 3 presents the results from estimating model (3) with OLS. We find that stay-at-home requirements reduced mobility much more in cities belonging to culturally tight provinces. Our estimates suggest that stay-at-home restrictions reduced mobility by twenty-five percent on average (column 1), while in provinces characterised by tighter cultural traits, mobility declined by forty percent. Thus, consistent with our hypothesis, mobility fell sixty percent more in tight provinces relative to others. Column 3 shows that our coefficients are unchanged when we control for city-level GDP per capita. The size of the estimated coefficients presented in Table 3 suggest that compliance with mobility restrictions was higher in China overall than in the United States (see Table 2). This speaks to the intuition that Chinese citizens are generally more collectivist than their American counterparts, though it might also partly reflect more severe sanctions being imposed on violations of stay-at-home requirements by the Chinese authorities.

Table 3.

Cultural tightness and compliance across China.

	(1)	(2)	(3)
	Mobility	Mobility	Mobility
Stay at home dates (city-level)	-24.301***	-24.589***	-23.726***
	(3.323)	(0.897)	(0.443)
Stay at home dates $\times$ cultural tightness		-14.731***	-13.927***
		(1.416)	(0.324)
Observations	38,430	38,430	11,956
R-squared	0.931	0.931	0.961
City FE	yes	yes	yes
Province-date FE	yes	yes	yes
Policy x log GDP	no	no	yes

The table presents OLS estimates from regressing changes in mobility at the city-level on city-specific lockdown dates. The tightness measure for each province is taken from Chua et al. (2019) . Column 3 includes the interaction between stay at home dates and the log of city-level GDP per capital. Errors are clustered at the city-level. The coefficients with ★★★ are significant at the 1% level, with ★★ are significant at the 5% level, and with ★ are significant at the 10% level.

6. Discussion and implications

Scholars have documented striking deviations from what textbook representations of Homo economicus would predict (Camerer, Malmendier, 2007, Fehr, Gächter, 2000, Roth, Prasnikar, Okuno-Fujiwara, Zamir, 1991). As observed in Henrich et al. (2001), many people “are willing to reward those who act in a cooperative manner while punishing those who do not even when these actions are costly to the individual.” How cooperation is rewarded, however, depends on cultural traits, which vary significantly across countries (Gelfand, Raver, Nishii, Leslie, Lun, Lim, Duan, Almaliach, Ang, Arnadottir, et al., 2011, Heine, 2010, Henrich, Heine, Norenzayan, 2010, Schulz, Bahrami-Rad, Beauchamp, Henrich, 2019). These differences have been traced back to numerous historical events including the rise of agriculture (Nisbett, Peng, Choi, Norenzayan, 2001, Talhelm, Zhang, Oishi, Shimin, Duan, Lan, Kitayama, 2014), the family policies of the Latin Church (Schulz et al., 2019), and the differential exposure to disease-causing pathogens (Fincher et al., 2008). Whatever the source of cultural variation today, a key message from this research is that culture is highly persistent even at very long horizons, as people socially transmit helpful behaviour-shaping knowledge and information across generations (Boyd, Richerson, 1988, Henrich, 2017).

In this article we have shown how cultural factors have shaped contemporary public policy outcomes, in terms of peoples compliance with the lockdown rules introduced by governments around the world to contain the spread of COVID-19. Taking our findings seriously implies acknowledging the limits policymakers face in altering peoples behaviour when history casts a long shadow. That said, our findings should not be taken to suggest that the policies themselves do not matter. This research has shown that lockdown measures have significantly contributed to reduce mobility and enact social distancing. Nonetheless, it is easy to arrive at misleading conclusions about the effectiveness of different policies without accounting for the role of culture. Our study suggest that a realistic understanding of the role of cultural factors is likely to be critical for successful policy implementation.

Declaration of Competing Interest

We state that there are no conflicts of interest

Appendix A. Supplementary materials

Supplementary Data S1

Supplementary Raw Research Data. This is open data under the CC BY license http://creativecommons.org/licenses/by/4.0/