First, we examine whether the level of globalization of the country is correlated with the timing of international travel restrictions. With a simple correlation analysis, we find that the Pearson’s correlation between the first policy implementation-first case gap and globalization index is significantly positive ρ=0.35 (p<0.001; 95%CI=[0.210, 0.475]; n=170)[1], demonstrating that more globalized countries exhibited a delay in imposing travel restrictions compared with less globalized countries (Figure 3a). Figure 3a also indicates that countries that reacted before the first local COVID-19 case tended to adopt screening on arrivals or quarantine rules as the first precautionary measures. We find that more globalized countries tend to have a higher number of confirmed local cases of COVID-19 at the time of implementing travel restrictions (Pearson’s correlation between the log of confirmed cases and KOF index: ρ=0.408; p<0.001; 95%CI=[0.276, 0.525]; n=173), Figure 3b)[2]. One noteworthy case is the United Kingdom, which only enforced quarantine on travelers from high-risk regions on the 08 June 2020, 129 days after COVID-19 was first confirmed in the country.
These correlations persist and remain significant when each level of travel restriction is evaluated (see Figure S3 in SI Appendix). This shows that more globalized countries are more likely to impose international travel restrictions later, relative to the first confirmed case in the country, regardless of policy strictness. Interestingly, the two least strict policies (i.e., screening and quarantine) have slightly higher correlation coefficients meaning that it took more globalized countries longer to impose these policies relative to the first local COVID-19 case. One would think that the least strict policies would represent a lower barrier to continued globalization and hence, be the more likely route for a COVID-19 response measure for more globalized countries. Regardless, it appears that the implementation of any level of travel/border restriction is delayed by more globalized countries.
Do more globalized countries take longer to implement travel restriction policies in general?
We present the results from the survival analysis in Table 2, which shows the hazard ratios (HRs) for each factor. For binary explanatory variables, HRs can be interpreted as the ratio of the likelihood of adopting travel restrictions between the two levels, while for continuous variables, it represents the same ratio for unit difference.
Table 2. Time-to-event analysis (marginal risk set model) predicting implementation of international travel restrictions.
Model
|
(1)
|
(2)
|
(3)
|
(4)
|
KOF Globalization Index
|
1.08
|
1.17†
|
1.76**
|
1.80*
|
|
(0.0739)
|
(0.0958)
|
(0.363)
|
(0.535)
|
Neighbor restriction adoption
|
|
1.30*
|
1.43***
|
1.45***
|
|
|
(0.141)
|
(0.146)
|
(0.158)
|
Neighbor COVID-19 case (7-day total, log)
|
|
1.09**
|
1.15***
|
1.15***
|
|
|
(0.0312)
|
(0.0429)
|
(0.0438)
|
Domestic COVID-19 case (7-day total, log)
|
|
1.02
|
1.02
|
1.02
|
|
|
(0.0426)
|
(0.0470)
|
(0.0493)
|
Less restrictive travel policy adopted
|
|
3.28***
|
3.11***
|
3.14***
|
|
|
(0.577)
|
(0.582)
|
(0.601)
|
Weekends
|
|
0.49*
|
0.39**
|
0.39**
|
|
|
(0.172)
|
(0.133)
|
(0.134)
|
Government Effectiveness (WGI)
|
|
|
0.83
|
0.83
|
|
|
|
(0.241)
|
(0.245)
|
Electoral democracy index
|
|
|
1.05
|
1.04
|
|
|
|
(0.109)
|
(0.111)
|
GDP per capita (log)
|
|
|
1.10
|
1.08
|
|
|
|
(0.193)
|
(0.196)
|
Unemployment (%)
|
|
|
1.04**
|
1.04**
|
|
|
|
(0.0152)
|
(0.0154)
|
GINI index
|
|
|
0.99
|
0.99
|
|
|
|
(0.0104)
|
(0.0106)
|
Hospital beds (per 1k people)
|
|
|
1.10*
|
1.09*
|
|
|
|
(0.0399)
|
(0.0403)
|
Population ages 65+ (%)
|
|
|
0.93**
|
0.94**
|
|
|
|
(0.0216)
|
(0.0219)
|
Urban population (%)
|
|
|
1.00
|
1.00
|
|
|
|
(0.00470)
|
(0.00490)
|
Population density (log)
|
|
|
0.99
|
0.99
|
|
|
|
(0.0556)
|
(0.0563)
|
MERS or SARS experience
|
|
|
0.78
|
0.78
|
|
|
|
(0.269)
|
(0.270)
|
Continent
|
|
|
|
|
Africa
|
|
|
0.94
|
0.95
|
|
|
|
(0.305)
|
(0.316)
|
Asia
|
|
|
1.36
|
1.38
|
|
|
|
(0.407)
|
(0.421)
|
Central America
|
|
|
0.52
|
0.54
|
|
|
|
(0.235)
|
(0.243)
|
Europe
|
|
|
(ref.)
|
(ref.)
|
North America
|
|
|
1.00
|
0.99
|
|
|
|
(0.481)
|
(0.482)
|
Oceania
|
|
|
2.32**
|
2.27*
|
|
|
|
(0.737)
|
(0.726)
|
South America
|
|
|
1.35
|
1.40
|
|
|
|
(0.448)
|
(0.462)
|
KOF*Neighbor restriction adoption
|
|
|
|
0.93
|
|
|
|
|
(0.0720)
|
KOF*Neighbor COVID-19 case (7-day total, log)
|
|
|
|
1.02
|
|
|
|
|
(0.0312)
|
Num. obs.
|
55163
|
45484
|
35418
|
35418
|
Num. countries
|
173
|
158
|
121
|
121
|
Num. failures
|
655
|
594
|
455
|
455
|
Pseudo R2
|
0.001
|
0.037
|
0.068
|
0.068
|
Log likelihood
|
-2140.460
|
-1813.963
|
-1227.150
|
-1226.585
|
Note: Hazard ratios. Standard errors (clustered at country level) in parentheses. † p < .10; * p < .05; ** p < .01; *** p < .001.
Despite the strong positive correlation observed in the bivariate analysis between globalization and the time difference between first local confirmed case and implementation of travel restriction, we did not find substantial evidence suggesting that more globalized countries are more reluctant to adopt travel restriction policies relative to their first local confirmed case. In fact, after adjusting for the date that COVID-19 was first locally contracted (through observation stratification), more globalized countries are more likely to adopt travel restriction policies. Specifically, as the KOF globalization index increases by one standard deviation (e.g., Paraguay to New Zealand), the likelihood of adopting travel restrictions increases by 80% (p=0.007; 95%CI=[1.163, 2.617], Table 2 model 3).
We find strong evidence of travel restriction policy diffusion between countries that are heavily interdependent in the tourism sector; that is, a country is more likely to adopt a travel restriction if neighboring countries (in terms of share of non-resident visitor arrivals) have done so. As expected, an increase in COVID-19 prevalence in regions comprising the majority of inbound international tourist arrivals increases the likelihood of enforcing travel restrictions. Specifically, for every 1% increase in COVID-19 cases in neighboring countries, the chance of adopting a travel policy increases by about 15% (p<0.001, 95%CI=[1.075, 1.237]). On the other hand, increases in domestic COVID-19 cases do not appear to influence travel policy adoptions[3] suggesting that travel restriction policy decisions may be driven more by ‘keeping the disease out’ than containing the disease locally for the greater global good. The likelihood of adopting a restrictive travel policy (e.g., arrivals ban) is about three times higher if the country has already implemented a less strict policy, suggesting there may be decreased difficulty in implementing more restrictive policies over time or an increased preference to do so. Moreover, policy change is 60% less likely to occur during weekends (p=0.005, 95%CI=[0.199, 0.757]), perhaps because government officials are less likely to be working on weekends and hence, less active in the political decision-making process.
The effect of the electoral democracy index is not statistically significant, and our results are contrary to the findings of [38], where OECD countries with higher electoral democracy have lower rates of domestic policy adoption[4]. Perhaps decisions to implement international travel restrictions are less controversial to voters than domestic policies as the former primarily aims at limiting mobility from outside country borders rather than restricting the freedom and mobility within country borders as the latter do. In addition, we find that countries with a higher unemployment rate are more likely to implement travel restrictions. Surprisingly, countries with a larger share of older population are less likely to implement travel restrictions, while no statistically significant effect was observed for the share of urban population and population density. Contrary to our expectation, countries with greater healthcare capacity tend to be more likely to adopt a travel restriction policy[5].
Government capacity as a relevant mediator. Government effectiveness as a proxy of state capacity can act as a mediator as evidence is available that countries with higher effectiveness took longer to implement COVID-19 related responses [36, 46]. Countries with higher levels of health care confidence exhibit slower mobility responses among its citizens [48]. Those results may indicate that there is a stronger perception that a well-functioning state is able to cope with such a crisis. More globalized countries may therefore take advantage of a better functioning state. They weight advantages and disadvantages of policies and, consequently, slow down the policy restriction process to benefit longer from international activities. When including the interaction term between the globalization index and measures of state capacity in the model, we find strong evidence suggesting that more globalized countries with higher government effectiveness are slower to adopt travel restrictions. Each regression includes the same set of control variables as those used in Table 2 model 5. As shown in Figure 4, the hazard ratios of the interaction terms between globalization index and government effectiveness are statistically less than one (p=0.001), as well as the interaction term with an alternative measure of state capacity, namely ICRG quality of government (p=0.006) and tax capacity (p=0.018). Moreover, we also find a similar effect with the interaction terms between globalization and health capacity (as measured by number of hospital beds (p=0.075), physicians (p<0.001), or nurses and midwives per 1,000 (p<0.001), and current expenditure on heath (log) (p<0.001)). This evidence supports the notion that countries with higher state or health capacity and globalization were less likely to limit international travel, even when the stakes might be comparatively higher, i.e., when the country is more globalized and hence, more susceptible to infectious disease outbreaks.
Which aspect of globalization can primarily account for these responses?
We assess which aspects of globalization are more important when predicting travel restriction policy adoption by examining the influence of each (sub)dimension of the globalization index. Overall, we find the likelihood of implementing travel restriction policies among countries with high state capacity is robustly estimated for all subcomponents (Figure 5A), with HRs ranging from 0.72 (political globalization) to 0.79 (financial globalization). A closer inspection distinguishing between de facto (actual flows and activities, Figure 5B) and de jure (policies, resources, conditions and institutions, Figure 5C) measures [26] leads to interesting insights.
First, we find that de jure trade (trade regulation, trade taxes, tariff rates and free trade agreements) and de jure political (number of treaties and memberships in international organizations) globalization, have the largest effect out of all other sub-dimensions of globalization. The former implies that policies with the intent to facilitate and promote trade flows between countries actually became barriers to the implementation of travel controls in response to the pandemic. The latter is a highly surprising result given the call for international cooperation and coordination by many international organizations (e.g., WHO[6], World Economic Forum[7], United Nations[8]). We find that those countries with high government effectiveness and engagement in international political cooperation are less likely to implement travel restriction policies and hence, slower to do so. On the other hand, de facto trade globalization, which measures actual trade activities (such as exchange of and goods and services) over long distances, is not as strongly related to the timing of travel policy adoption for countries with high government effectiveness. De facto interpersonal globalization, which largely comprises international tourism, international students, and migration, has the largest effect among other de facto globalization dimensions. These results suggest that a nation with high government effectiveness and more global social, interpersonal, and cultural flows is less likely to implement travel restriction policies in pandemic crises and hence, will delay doing so. Countries with higher government effectiveness, policies, and conditions that tend to facilitate or favor globalization (e.g., trade policy, political connectedness and engagement in international political cooperation) are also less likely to implement travel restrictions.
Placebo analysis with domestic COVID-19 responses
To assess whether the observed delay in travel restriction adoption is better explained by globalization and its interplay with state capacity, we conduct a placebo analysis using COVID-19 policy responses that, at least in theory, cannot be explained by the same mechanism. Specifically, we employ the same event history analysis on domestic non-pharmaceutical interventions (NPIs) imposed to mitigate COVID-19 transmission. While previous studies have argued for [49] and found a substantial negative effect of government effectiveness on the timeliness of enacting school closure policies [36] and other NPIs across Europe [46], there is no obvious reason why the delayed responses to implement domestic NPIs would be related to globalization. Thus, we would expect that the interaction term between globalization and government effectiveness to be zero. If our expectation is correct, then we are more comfortable interpreting our previous results as truly reflective of the effect of globalization on travel restrictions, rather than as the effect of globalization on the propensity to implement all types of NPIs.
Data on domestic NPIs adoption are derived from the same source we obtained records on international travel restriction (i.e., the OxCGRT database). Domestic containment and closure policies include closing of schools, workplace, and public transport, restriction on gatherings and internal movement, cancellation of public events, and shelter-in-place order. We follow the approach of [38], who focus only on mandatory nationwide policies adopted[9]. We again utilize the marginal risk set model in analyzing the timing of adoption of the seven domestic policies, that is, we stratified the seven different policies and their variation in strictness. Similarly, adoption of a stricter version of the policy (e.g., restrictions on gatherings between 11-100 people or 10 people or less) implies the adoption of the less strict version.
The results of the placebo analysis are presented in Table S3, showing the hazard ratios of each factor predicting the adoption of any COVID-19-related NPIs. Comparison of the estimates of several key variables to previous studies, while subject to a larger set of countries and more complete time frame, suggests that our modelling approach is reasonable[10]. Similar to the adoption of international travel restrictions, more globalized countries are quicker to implement domestic NPIs than their less globalized counterparts. Notably, the estimates of HRs are larger in magnitude and with higher statistical significance compared to the set in Table 2 for the case of international travel restrictions. This shows that the relative speed of more globalized countries in adopting travel restrictions is slower than domestic NPIs, compared to less globalized countries, suggesting the former takes relatively more time to impose international travel restrictions, where one would expect international travel policy to be adopted relatively earlier. Secondly and more importantly, we did not find any statistical evidence suggesting the effect of state capacity varies across countries with different levels of globalization as the interaction effect between KOF and government effectiveness is not significant. This result holds for the alternative measures of state capacity as well as using measures of health system capacity. Finally, we also show that the results of the placebo analysis are not sensitive to the type of domestic policy adopted (see Table S4) nor when different dimensions of globalization were considered, as none of the HRs of their interaction terms are statistically significantly smaller than one[11].
Nevertheless, while the results from the placebo analysis suggests that the results we see in Table 2 are less likely to arise from, e.g., confounding effects due to other unobserved variables, given the difference in nature of domestic and international NPIs[12], we cannot conclusively claim that this is in fact the case. For example, an alternative explanation for why more globalized countries respond relatively faster with domestic policies than do less globalized countries might be found in the fact that most of the domestic policies were implemented at a later stage of the pandemic (compared to travel restrictions which were typically adopted early on). Hence, globalized countries may be better at learning how to coordinate resources and implement social distancing policies.
COVID-19 case severity at the introduction to international travel restriction policies
We conduct an analysis using the Ordinary Least Squares model predicting the number of confirmed COVID-19 cases at the time each travel restrictions were implemented[13]. In each regression, we control for the date when the country has the first confirmed COVID-19 case. For countries that have no confirmed cases when the travel restriction was implemented (i.e., date of the first confirmed is later than date of the policy adoption), we recode this variable to the date when the policy was adopted.
In Figure 6, we present the estimates of KOF globalization index on COVID-19 prevalence (total number of cases in log (6A) and case per capita in log (6B)) at the time the travel restriction was implemented. We report the estimates obtained from the models without controlling for other factors except for the date of the first confirmed case, as well as models in which we include a full set of control variables (full regression results are presented in Table S5 and Table S6). This includes government effectiveness, electoral democracy, GDP per capita, unemployment rate, GINI coefficient, number of hospital bed per 1,000 people, urban population, population density, whether the country experienced SARS or MERS, and region dummies. Additionally, we also control for containment policies implemented before the introduction of the travel restrictions of interest. We proxy this variable by the average value of the stringency index from the beginning of the time period to the day before the travel policy was adopted[14].
We find strong positive associations between the globalization index and the number of confirmed COVID-19 cases (and per capita cases) at the time the travel restriction policy was first introduced, when we only account for when the country was first exposed to COVID-19. In particular, with a one standard deviation increase in globalization index, the predicted number of COVID-19 cases increases by about 1.9 times when screening (or more strict policies) was first adopted, while cases per capita are 7.7 times higher. The globalization multiplier in COVID-19 cases (or cases per capita) is higher when considering firmer travel restrictions (i.e., adoption of quarantine and banning entry from high-risk regions) except for total lockdown. However, the coefficient estimates for globalization predicting COVID-19 cases at the time of total border closure is likely to be underestimated, as a number of highly globalized countries, such as the USA, Japan, South Korea, and a large group of European countries (with the exception of Germany) did not totally close their borders at any point.
However, except for the adoption of screening and quarantine, the effect of globalization became statistically insignificant when other control variables are added to the model. The reduction in the effect size is not unexpected as globalization index is highly correlated with several control variables, such as GDP per capita (ρ=0.631), government effectiveness (ρ=0.751), and share of population over 65 (ρ=0.775).
Additionally, we find further evidence supporting the mediating role of state capacity to the effect of globalization as suggested by the statistically significant interaction effect between globalization and government effectiveness (Table 3). That is, among globalized countries, those with higher state capacity are more likely to have more COVID-19 cases when the government impose travel restrictions. This echoes the findings from the time-to-event analysis.
Table 3. State capacity mediating effect on globalization.
|
Screening
|
Quarantine
|
Ban high-risk
|
Total lockdown
|
KOF Globalization Index
|
0.92*
|
1.18**
|
0.85†
|
-0.24
|
|
(0.458)
|
(0.415)
|
(0.436)
|
(0.487)
|
Government Effectiveness (WGI)
|
0.023
|
-0.25
|
0.14
|
0.24
|
|
(0.377)
|
(0.413)
|
(0.467)
|
(0.467)
|
KOF*WGI
|
0.37†
|
0.47*
|
0.44*
|
0.53**
|
|
(0.204)
|
(0.202)
|
(0.196)
|
(0.185)
|
Controls
|
Yes
|
Yes
|
Yes
|
Yes
|
Number of countries
|
118
|
118
|
118
|
118
|
Prob. > F
|
0.000
|
0.000
|
0.000
|
0.000
|
R2
|
0.759
|
0.718
|
0.709
|
0.835
|
Notes: OLS estimates. Dependent variable: Number of confirmed COVID-19 cases (log) at time of travel policy adoption. Standard errors (heteroskedasticity-robust) in parentheses. † p < .10; * p < .05; ** p < .01; *** p < .001.
[1] Since the effect of travel restrictions might delay an outbreak of the virus, which itself might be more salient for more globalized countries, we check the correlation by censoring negative gaps (travel policy implementation before first confirmed COVID-19 case) to zero. The correlation is highly statistically significant, while the effect size is smaller (ρ=0.248; p=0.0011; n=170). Four countries were excluded from the calculation as they have zero COVID-19 cases during the entire sample period. The correlation increases to ρ=0.366 (p<0.001) when the end of the sample period date is used to calculate the first policy implementation-first case gap for these countries.
[2] We obtain very similar results when confirmed cases are adjusted for population size, i.e., log confirmed COVID-19 cases per million people (ρ=0.397; p<0.001; n=173).
[3] In a separate model, we control for death rate instead of number of new confirmed cases in the last seven days; the effect of either variable is statistically insignificant when added separately in the model or together.
[4] The results are highly robust when we substitute other measures of democracy for electoral democracy, such as the Boix-Miller-Rosato (BMR) dichotomous coding of democracy [55], (revised) polity score and institutionalized democracy score from Polity V.
[5] In addition, the effect is more pronounced if health capacity is measured with number of physicians per 1,000 people. However, using the number of nurses and midwives per 1,000 has no effect.
[6] See https://www.who.int/nmh/resource_centre/strategic_objective6/en/
[9] In unreported analysis, we included policy recommendation of closure and containment. This does not alter the findings.
[10] For example, we find some evidence of policy diffusion beyond the OECD context [38], while timing of domestic NPIs adoption is not sensitive to foreign COVID-19 case. We also find robust evidence that countries with a large state capacity delay implementation of domestic COVID-19 policies [36, 46]. Interestingly, we also find that countries with relevant past experience (SARS and MERS) intervened relatively early [49].
[11] Notably, the HRs for gathering and internal movement restrictions are statistically (at 10% level) larger than one.
[12] Where the former are purposed to prevent and control mass transmission of the virus within the country and the latter aims to avoid the virus from coming in to the country. In particular, countries adopt travel restrictions at an earlier stage compared to domestic policies (between mid-March to April).
[13] If the country did not adopt the travel restriction, we take the COVID-19 case statistics at the end of the sample period (n=4 for entry ban and n=37 for total border closure). Since we use cumulative case statistics, the resulting coefficients are likely to be underestimated. This is because the sample of countries that did not implement travel bans has a higher level of globalisation than the mean, including the UK and the USA.
[14] This measure captures the adoption of seven domestic containment policies and public information campaign, as well as the implementation of less restrictive travel restrictions. See https://github.com/OxCGRT/covid-policy-tracker/blob/master/documentation/index_methodology.md for the construction of the stringency index.