Professional Documents
Culture Documents
Peterson Institute - pb20-14
Peterson Institute - pb20-14
In March 2020, most countries implemented stringent measures—closing schools Julien Maire was a
research analyst at the
and workplaces, limiting public gatherings, and curbing travel—to reduce the Peterson Institute for
spread of the SARS-CoV-2 virus, which causes COVID-19. Thanks to smartphone International Economics
from November 2019
data from Google, it is possible to measure the impact of pandemic-related
through October 2020.
lockdowns on individual mobility. Understanding the impact of lockdown He worked with C.
measures on individual mobility and how it differed across countries is important, Fred Bergsten Senior
Fellow Olivier Blanchard
because it is the first step in determining the effectiveness of lockdown measures on issues related to
on health outcomes and their impact on the economy. macroeconomic policy.
Previously, he worked
This Policy Brief examines the effects of the stringent measures implemented as a research assistant
in March–May 2020 on individual mobility. The results suggest that stringent at the World Bank on
the World Development
measures were more effective at reducing individual mobility in higher-income
Report 2020. He thanks
countries than in lower-income countries and that the differences reflect Olivier Blanchard,
factors such as extreme poverty, perception of risk, the share of vulnerable Pinelopi K. Goldberg,
and Adam Posen for
employment, number of hospital beds, age distribution of the population, and their invaluable guidance
population density. and support and Egor
Gornostay, Cullen S.
The Policy Brief uses the Oxford Stringency Index (Hale et al. 2020b), which
Hendrix, Edwin M.
is based on nine measures: school closures, workplace closures, cancellation of Truman, and Eva Zhang
public events, restrictions on gatherings, closures of public transportation, stay-at- for their comments.
1. Available at www.google.com/covid19/mobility.
Figure 1
Oxford Stringency Index, by country income group, February 12–May 6, 2020
80
60
40
20
0
February 12 March 4 March 25 April 15 May 6
2 Although smartphone ownership varies widely across countries, mobility data from smart-
phones are robust to potential biases that can arise from different geographic and socioeco-
nomic groups. See Laura Silver, “Smartphone Ownership Is Growing Rapidly around the World,
but Not Always Equally,” Pew Research Center, February 5, 2019, https://www.pewresearch.
org/global/2019/02/05/smartphone-ownership-is-growing-rapidly-around-the-world-but-not-
always-equally/. See also Wesolowski (2013).
PB 20-14 | NOVEMBER 2020 3
Figure 2
Percentage change in mobility relative to January 2020 baselines, by country
income group
a. Unsmoothed data
percentage change relative to baseline
−20
−40
−60
b. Smoothed data
percentage change relative to baseline
−20
−40
−60
February 12 March 4 March 25 April 15 May 6
One could argue that the direction of causality could go either way: The
decrease in mobility could have been a response to more stringent measures,
or stringent measures could have been taken after an increase in individual
mobility. A Granger causality test suggests that stringency predicts mobility,
however, mobility also Granger-causes stringency (see table A.1 in the appendix).
An instrumental variable (IV) approach is adopted to deal with the issue of
reverse causality. The IV used for the Stringency Index is an average of the
index values of a country’s region excluding the country being measured.3 This
variable is highly correlated with the Stringency Index of the country (correlation
coefficient of 0.89) and uncorrelated with the error term of the first specification
(correlation coefficient of 0.02).
Columns 1 and 3 of table 1 indicate that overall stringency measures are
associated with a decrease in individual mobility: A 1 percentage point increase
3 Regions correspond to World Bank regions: East Asia and Pacific, Europe and Central Asia,
Latin America and the Caribbean, Middle East and North Africa, North America, South Asia,
and Sub-Saharan Africa.
Table 1
Regression results on effect of Stringency Index, number of new COVID-19 cases and deaths on individual mobility
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11)
Mobility Mobility Mobility Mobility Mobility Mobility Mobility Mobility Mobility Mobility Mobility
Item (OLS) (OLS) (IV) (IV) (OLS) (OLS) (OLS) (IV) (IV) (IV) (IV)
–0.726*** –0.750*** –0.808*** –0.814*** –0.682*** –0.690*** –0.673*** –0.762*** –0.769*** –0.759*** –0.765***
Stringency Index
(–102.69) (–125.70) (–96.01) (–126.31) (–93.79) (–95.65) (–95.60) (–85.97) (–88.04) (–95.64) (–103.77)
–0.215*** –0.276*** –0.172*** –0.175***
New COVID-19 cases in region per million people
(–18.10) (–19.19) (–14.00) (–11.62)
PB 20-14 | NOVEMBER 2020
14.83*** 30.52*** 20.34*** 21.86*** 14.53*** 14.44*** 29.76*** 19.32*** 19.37*** 17.31*** 17.73***
Constant
(28.70) (14.62) (33.82) (12.31) (28.94) (28.64) (14.73) (32.82) (32.78) (9.78) (10.11)
N 5,442 5,442 5,442 5,442 5,442 5,442 5,442 5,442 5,442 5,442 5,442
R-squared 0.660 0.814 0.651 0.810 0.679 0.678 0.826 0.672 0.670 0.821 0.822
Figure 3
Average number of COVID-19 cases and deaths, by country income group
30
20
10
0
February 12 March 4 March 25 April 15 May 6
0
February 12 March 4 March 25 April 15 May 6
5 The number of new cases and the number of new deaths are highly correlated.
PB 20-14 | NOVEMBER 2020 7
Table 2
Regression results on effect of Stringency Index on individual mobility, by country income
group
• Most people in poor countries have no savings, and few safety nets exist; few
people can therefore afford not to work.
• The consequences of COVID-19 are less severe for young people, who may
therefore be willing to take more chances than older people. If a country’s
population is young, the effect of lockdown measures on mobility may
therefore be weaker.
PB 20-14 | NOVEMBER 2020 8
• If population density favors the spread of the virus, denser countries may
have a stronger mobility response than less dense ones.
• A country with a strong health system can more easily face a health crisis and
avoid overloading its capacity. If people in such a country are therefore more
likely to survive, they may be less willing to restrict their mobility. All else
equal, one might therefore expect a lower decline in individual mobility in a
country with a larger number of hospital beds per inhabitants.
Table 3
Description of covariates (average) by country income group (percent except where
indicated otherwise)
a. The International Labor Organization defines vulnerable workers as the sum of own-account workers and contributing
family workers.
Note: Data are for last year available for each country.
Source: World Bank.
Table 4
Regression results on effects of Stringency Index, new COVID-19 cases and deaths, and country
covariates on individual mobility
F-statistic (first-
13,764.1 14,300.6 13,806.2 13,752.6 9,458.6 10,325.9 9,241.9 9,308.06
stage regression)
Note: t-statistics in parentheses.
* p < 0.05, ** p < 0.01, *** p < 0.001.
Source: Author’s calculations.
6 The variable “extreme poverty” is not included, because the share of vulnerable employment
and the share of extreme poverty are highly correlated (coefficient of correlation of 0.77).
PB 20-14 | NOVEMBER 2020 10
Table 5
Regression results on effect of Stringency Index, new COVID-19 cases, and
country covariates on individual mobility, by country income group
(1) (2)
Mobility Mobility
Item (IV) (IV)
–0.844*** –0.777***
Stringency Index x high income
(–86.99) (–66.74)
–0.839*** –0.816***
Stringency Index x upper-middle income
(–96.28) (–83.43)
–0.773*** –0.771***
Stringency Index x lower-middle income
(–79.86) (–65.95)
–0.694*** –0.766***
Stringency Index x low income
(–49.09) (–41.02)
0.212***
Extreme poverty
(9.19)
–0.129***
New COVID-19 cases in region per million people
(–8.27)
2.071***
Hospital beds per thousand people
(19.57)
–0.398***
Share of population 65 and older
(–6.74)
–0.00561***
Population density
(–4.87)
20.48*** 17.85***
Constant
(34.41) (18.37)
N 5,204 5,204
not possible to determine whether the different responses reflect country fixed
effects or interactions with the country income group. Table A.5 does not show
a significant reduction in differences between coefficients associated with the
Stringency Index and the country income group. Thus, although covariates such
as extreme poverty, vulnerable employment, a larger share of young people, and
population density decrease the effect of stringency measures on mobility, it is
not clear how much they explain the different responses of individual mobility to
stringent measures across country income groups.
These data are reported by percentile rank term, ranging from 0 (lowest
percentile) to 100 (highest percentile).
Table 6 shows the effects of these variables on individual mobility. Without the
additional variables presented in table 4, all of the variables measuring the quality
of institutions have a significant negative effect on individual mobility (columns
1–3). When other variables are added, however, only political stability has a robust
significant effect (columns 4–6). If the index measuring political stability increases
by 1 percentage point, mobility decreases by 0.07 percentage points. The rule of
law and to some extent government effectiveness are not as important as other
variables, such as the share of extreme poverty, the perception of risk, the number
of hospital beds, the share of young people in the population, and population
density, in understanding differences in individual mobility across countries. These
results are robust when both periods are considered (see table A.6).
One might posit that the quality of institutions operates through the
Stringency Index. However, regressions of the Stringency Index on the three
governance variables show no significant effects. Hale et al. (2020a) find
the same results.
PB 20-14 | NOVEMBER 2020 12
Table 6
Regression results on the effects of the quality of institutions on individual mobility
CONCLUSIONS
The stringent measures to reduce the spread of COVID-19 that were adopted
across countries in the spring of 2020 decreased individual mobility. But
the response varied across countries, with lower-income countries revealing
weaker responses. This finding can be explained by factors such as extreme
poverty, perception of risk measured by the number of new COVID-19 cases and
deaths, the share of vulnerable employment, the number of hospital beds, the
share of young people in the population, and population density. The results
highlight the importance of considering these factors, regardless of the level of
stringency. Variables measuring the quality of governance, such as rule of law
and to some extent government effectiveness, do not explain the decrease in
individual mobility.
PB 20-14 | NOVEMBER 2020 13
REFERENCES
Hale, Thomas, Noam Angrist, Beatriz Kira, Rafael G. Goldszmidt, Anna Petherick, and
Toby Phillips. 2020a. Pandemic Governance Requires Understanding Socioeconomic
Variation in Government and Citizen Responses to COVID-19. Available at https://ssrn.
com/abstract=3641927.
Hale, Thomas, Noam Angrist, Anna Petherick, Toby Phillips, and Samuel Webster. 2020b.
Variation in Government Responses to COVID-19. Working Paper 31. Oxford: Blavatnik
School of Government.
Wesolowski, Amy, Nathan Eagle, Abdisalan M. Noor, Robert W. Snow, and Caroline O.
Buckee. 2013. The Impact of Biases in Mobile Phone Ownership on Estimates of Human
Mobility. Journal of the Royal Society Interface. Available at https://doi.org/10.1098/
rsif.2012.0986.
World Bank. 2019. Worldwide Governance Indicators. Washington.
PB 20-14 | NOVEMBER 2020 14
APPENDIX A
Figure A.1
Changes in individual mobility from January 2020 baseline, by purpose of
mobility and country income group (percentage change from the baseline
value, which is the median mobility in January 2020 for each country)
−20
−40
−60
−10
−20
−30
−40
−20
−40
−60
d. Work
−20
−40
−60
February 12 March 4 March 25 April 15 May 6
Table A.1
Granger causality test for individual mobility and the
Stringency Index
(1) (2)
Stringency
Item Mobility Index
–0.0352*** 1.005***
Lagged Stringency Index
(–31.17) (427.31)
0.955*** 0.0317***
Lagged mobility
(665.70) (10.62)
0.310*** 1.357***
Constant
(7.37) (15.51)
N 9,834 9,813
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11)
Mobility Mobility Mobility Mobility Mobility Mobility Mobility Mobility Mobility Mobility Mobility
Item (OLS) (OLS) (IV) (IV) (OLS) (OLS) (OLS) (IV) (IV) (IV) (IV)
–0.757*** –0.773*** –1.289*** –1.140*** –0.704*** –0.709*** –0.681*** –1.216*** –1.222*** –1.048*** –1.054***
Stringency Index (March 12–April 1)
(–54.25) (–60.84) (–42.55) (–59.34) (–49.80) (–50.38) (–54.19) (–38.62) (–38.94) (–52.65) (–53.77)
–0.870*** –0.890*** –1.325*** –1.197*** –0.846*** –0.847*** –0.826*** –1.265*** –1.270*** –1.124*** –1.125***
Stringency Index (April 2–May 9)
(–71.91) (–80.07) (–50.31) (–72.46) (–71.10) (–71.04) (–75.59) (–47.83) (–47.86) (–66.98) (–67.76)
PB 20-14 | NOVEMBER 2020
23.60*** 37.88*** 60.62*** 53.19*** 23.65*** 23.28*** 37.23*** 57.48*** 57.47*** 46.83*** 46.93***
Constant
(24.05) (19.97) (28.44) (24.87) (24.81) (24.33) (20.89) (27.19) (27.05) (22.46) (22.59)
N 4,489 4,489 4489 4,489 4,489 4,489 4,489 4,489 4,489 4,489 4,489
R-squared 0.556 0.811 0.405 0.775 0.583 0.580 0.833 0.454 0.449 0.801 0.800
Table A.3
Regression results on effect of Stringency Index on individual mobility, by country income
group and period
Table A.4
Regression results on effect of Stringency Index, new COVID-19 cases and deaths, and country
covariates on individual mobility, by period
F-statistic (first-
25,509.1 25,504.7 25,518 24,539.5 14,093.6 15,785.2 14,083.9 14,121.6
stage regression)
Note: t-statistics in parentheses.
* p < 0.05, ** p < 0.01, *** p < 0.001.
Source: Author’s calculations.
PB 20-14 | NOVEMBER 2020 20
Table A.5
Regression results on effect of Stringency Index, new COVID-19 cases, and
country covariates on individual mobility, by country income group and period
(1) (2)
Mobility Mobility
Item (IV) (IV)
–1.368*** –1.237***
Stringency Index x high income (March 12–April 1)
(–41.64) (–33.30)
–1.394*** –1.307***
Stringency Index x high income (April 2–May 9)
(–48.02) (–41.77)
–1.288*** –1.192***
Stringency Index x upper-middle income (March 12–April 1)
(–42.01) (–36.76)
–1.345*** –1.273***
Stringency Index x upper-middle income (April 2–May 9)
(–51.50) (–46.49)
–1.302*** –1.208***
Stringency Index x lower-middle income (March 12–April 1)
(–36.69) (–31.97)
–1.308*** –1.238***
Stringency Index x lower-middle income (April 2–May 9)
(–46.45) (–41.29)
–1.243*** –1.179***
Stringency Index x lower-middle income (March 12–April 1)
(–27.56) (–25.61)
–1.314*** –1.259***
Stringency Index x lower-middle income (April 2–May 9)
(–37.95) (–34.68)
0.133***
Extreme poverty
(4.84)
–0.172***
New COVID-19 cases in region per million people (March 12–April 1)
(–4.71)
–0.0307
New COVID-19 cases in region per million people (April 2–May 9)
(–1.43)
1.814***
Hospital beds per thousand people
(13.44)
–0.241**
Share of population 65 and older
(–2.83)
–0.000574
Population density
(–0.38)
62.35*** 53.01***
Constant
(27.96) (20.95)
N 4,261 4,261
Table A.6
Regression results on effect of Stringency Index and the quality of institutions on individual
mobility, by period
This publication has been subjected to a prepublication peer review intended to ensure
analytical quality. The views expressed are those of the author. This publication is part of the
overall program of the Peterson Institute for International Economics, as endorsed by its
Board of Directors, but it does not necessarily reflect the views of individual members of
the Board or of the Institute’s staff or management.