Global evidence on the economic effects of disease suppression during COVID-19

In early 2020, governments implemented unprecedented non-pharmaceutical interventions (NPIs) to mitigate COVID-19 transmission, including stay-at-home orders, travel and gathering restrictions, and closure of schools and workplaces. Prior work links such measures to reduced mobility and economic activity, though some argue substantial economic disruption would have occurred absent formal restrictions. While NPIs likely reduced deaths, evidence on their causal effects is mixed; meanwhile, other measures (masking, contact tracing) can reduce infections without directly limiting economic activity. The pandemic triggered a historic global contraction and increased poverty, with disproportionate economic harm among low-SES households. This study asks: (1) How prevalent was pandemic-related economic harm globally, and how was it related to subjective well-being and financial security? (2) What is the association between economic harm and the stringency of restrictions? (3) How did harm vary by socioeconomic status within and across countries? (4) How do stringency effects compare with alternative NPIs? Using Gallup World Poll microdata from 117 countries (July 2020–March 2021), matched to policy and disease data, the paper provides quasi-global evidence on individual-level economic harm and its association with policy regimes.

The paper situates its contribution within several strands: (a) Economic activity and mobility fell in response to COVID-19 and NPIs, evidenced in global and country-specific studies (e.g., Hale et al. 2021; Deb et al. 2020; Boone and Ladreit 2021; Carvalho et al. 2021). Some U.S.-focused analyses suggest voluntary behavior changes drove large parts of the decline (Goolsbee and Syverson 2020; Forsythe et al. 2020; Gupta et al. 2020). (b) NPIs likely reduced mortality, though findings are not uniform (Liu et al. 2021; Chernozhukov et al. 2021; Violato et al. 2021; Berry et al. 2021; Herby et al. 2022; Spiegel and Tookes 2022). Other health policies (contact tracing, masking) reduce infections without directly restricting economic activity (Fetzer and Graeber 2021; Abaluck et al. 2022). (c) The pandemic caused a 3.4% global output contraction (World Bank 2022), increased poverty (Mahler et al. 2021; Kim et al. 2021), and widespread income losses in low-income countries (Egger et al. 2021; Josephson et al. 2021). Economic effects were regressive within countries by income and education (Rothwell and Smith 2021; Narayan et al. 2022; Bundervoet et al. 2021; Kugler et al. 2021), consistent with historical pandemics (Furceri et al. 2022). The paper advances this literature with globally harmonized microdata and multilevel models linking policy stringency to individual-level harm and distributional impacts.

Data: The primary source is Gallup World Poll microdata (July 9, 2020–March 3, 2021): 321,386 adults (15+) in 117 countries/territories, nationally representative with survey weights. Key individual outcomes explicitly attributed to the coronavirus situation include: (1) whether life was affected a lot; among those working pre-pandemic: (2) temporary stoppage; (3) permanent job/business loss; (4) reduced hours; (5) reduced pay; and (6) a composite harm index (standardized mean of five items). Additional individual covariates: age groups, gender, foreign-born status, education, within-country household income quintile, urbanicity, marital status, presence of children under 15, indicator for being out of the labor force at survey time. Policy and context measures: Country-level, time-varying cumulative-to-date averages up to the interview month are constructed from (a) Oxford COVID-19 Government Response Tracker stringency index (containment and health-related public information component), standardized across 184 countries; (b) Oxford economic support index; (c) disease burden measured by reported COVID-19 deaths per capita and IHME model-based estimates of actual deaths; (d) behavioral proxies: Google Community Mobility Reports (visits to retail and restaurants) and additional validation data including FluNet influenza positives (2016–2021) and University of Maryland/Facebook survey on social contacts. For countries with subnational policy data (Brazil, Canada, UK, US), policies are population-weighted to national levels. Modeling: Primary analyses use multilevel mixed-effects models (Stata 17 mixed), with individuals nested in country-by-month-year clusters. Random intercepts vary by country and by month-year, allowing intragroup correlation. The dependent variable is individual-level economic harm (harm index or specific outcomes). Key regressors include cumulative-to-date stringency and economic support indices, COVID-19 deaths per capita (reported or IHME estimates), and individual covariates. Google mobility is added in robustness specifications to proxy voluntary behavior. Stringency is modeled continuously (standardized) and via quintiles to allow nonlinearity. Interaction models multiply stringency by within-country income quintiles to estimate heterogeneous associations by socioeconomic status, with predictions plotted across stringency centiles. Standard errors are robust and adjusted for clustering at country and time levels. Additional decomposition examines specific policies (school closures, stay-at-home orders, workplace closures, internal/international travel restrictions, public transport closures, mask mandates, public information, restrictions on gatherings, vaccination policy, testing policy, contact tracing, protections for the elderly) in separate models controlling for deaths and economic support. Validation and robustness: The harm index’s construct validity is evaluated via associations with subjective well-being outcomes (life evaluation, worsening living standards, worry, food insecurity). Alternative harm indices (factor-based, labor-only) yield similar results. World Poll job loss aligns with administrative unemployment changes (r≈0.52 across 52 countries). Stringency validity is probed via associations with mobility declines, reduced flu transmission, and lower self-reported social contact in more stringent settings. An out-of-sample U.S. state analysis relates cumulative stringency to COVID-related job loss using OLS and IV (party control) as a supportive, non-causal check. Weighting/coverage: Gallup weights adjust for selection probabilities and nonresponse; 2020–2021 mode shifts to CATI are accounted for in weighting. Analyses are limited to data through March 2021 to align measures temporally with outcomes.

- Prevalence of harm: Across 117 countries (July 2020–March 2021), 42% of adults reported being affected a lot by the pandemic. Among those working pre-pandemic: 51% were temporarily laid off, 50% lost hours, 49% lost income, and 27% permanently lost jobs. - Stringency associations: A one standard deviation increase in cumulative-to-date stringency predicts a 0.37 standard deviation increase in the harm index and a 14.2 percentage point increase in permanent job loss (95% CI ~8.3–20.1 ppt). Results are similar when comparing top vs. bottom stringency quintiles. Quintile model effects are monotonic: less stringency predicts less harm. - Economic support: Economic support policies are associated with reduced job loss and some mitigation of adverse outcomes (e.g., hours lost, temporary layoff), though effects on the composite harm index are mixed across specifications. - Disease burden: Reported or IHME-estimated COVID-19 deaths per capita are not significantly associated with the harm index or job loss in preferred models (95% CI includes zero), after controls. - Behavioral controls: Conditional on policies, Google mobility to restaurants sometimes positively correlates with harm or job loss; however, main stringency results are robust and stronger when using error-corrected death estimates. - Distributional impacts: Lower-SES individuals (bottom income quintiles; elementary education) experienced substantially greater harm. Interaction models show the harm–stringency slope is much steeper for low-income groups; in the highest stringency centile, permanent job loss was ~46.9% in the bottom income quintile vs. ~18.8% in the top quintile, while differences are minimal under low stringency. - Cross-country patterns: Lower-income countries had relatively low COVID mortality but high economic burden. GDP per capita correlates negatively with COVID-related job loss (r = -0.74) and positively with deaths per capita (reported r = 0.35; estimated r = 0.18). GDP per capita correlates with economic support (r = 0.55) but not with stringency (r = 0.02). - Mobility vs. survey indicators: Google mobility signals differ from survey-based harm; in richer countries, mobility fell despite job/income preservation, likely due to digital substitution. - Policy granularity: School closures, stay-at-home orders, internal travel restrictions, public transport closures, canceling public events, mask orders, workplace closures, and international travel restrictions are significantly associated with more harm/job loss. Vaccination policy, public information, testing policy, contact tracing, and protections for the elderly are not significantly associated with increased harm (some have negative but insignificant coefficients). - Validation: Seasonal flu positives dropped to ~18% of pre-pandemic baseline in 74 countries, consistent with substantial social distancing; stringency is linked to larger flu declines and reduced social contact. - U.S. out-of-sample check: Across U.S. states, cumulative stringency predicts ~0.6 SD (≈3 ppt) higher COVID-related job loss in OLS with controls; IV estimates using party control are stronger, though causality remains uncertain.

The study addresses whether and for whom stringent disease-suppression policies are associated with economic harm. Using harmonized global microdata and multilevel models that combine individual demographics with time-varying country policies and disease burden, the findings indicate that stricter cumulative restrictions are consistently linked to higher rates of job loss, income loss, and subjective disruption. Crucially, the economic burden is concentrated among lower-SES groups, with inequality widening as stringency rises: under low stringency, income-group differences in harm are negligible; under high stringency, gaps become large. The analysis also shows that several targeted public health measures (testing, tracing, protections for the elderly, vaccination policy intensity) are not associated with increased economic harm, whereas broad economic and social restrictions are. This suggests that policy design can influence the distribution and magnitude of economic costs. While the data cannot determine causality, the cross-national patterns are robust across specifications and validated against behavioral and epidemiological proxies, implying that policy stringency likely contributed to the observed economic harms beyond voluntary behavior changes. These findings are relevant for balancing public health benefits with economic costs, particularly in low-income settings and for vulnerable workers.

This paper provides the first quasi-global, harmonized microdata evidence linking COVID-19 policy stringency to individual-level economic harm and subjective well-being. It documents widespread job and income losses, especially among lower-SES groups, and shows that more stringent restrictions are associated with larger harms, while certain alternative public health measures are not linked to increased harm. The results underscore the distributional consequences of broad restrictions and highlight the potential of targeted measures to mitigate health risks without exacerbating economic hardship. Future research should: (1) develop causal identification strategies (e.g., natural experiments, better instruments) to isolate policy effects; (2) evaluate the cost-effectiveness and distributional impacts of alternative, more targeted NPIs and support policies; (3) examine long-run labor market and inequality consequences; and (4) integrate health benefits with economic outcomes to inform optimal policy mixes for future pandemics.

- Non-random policy assignment: Disease-suppression policies are endogenous to unobserved, time-varying country factors; omitted variable bias cannot be ruled out. Instrumentation is challenging due to limited predictors of disease burden; U.S. IV results using party control are supportive but not definitive. - Mode effects: Gallup World Poll shifted from face-to-face to telephone interviewing in many countries; although weights adjust for mode, residual bias may remain. - Attribution and measurement: Outcomes rely on self-reported attribution to COVID-19; while salient, recall and reporting differences across contexts may persist. Differences in death reporting motivate use of IHME modeled deaths, but measurement error remains possible. - External validity over time: Analysis covers July 2020–March 2021; later pandemic phases (vaccination rollout, variants, policy adaptations) are not analyzed. - No cost-benefit analysis: The study estimates associations, not causal effects or welfare balances; benefits of lives saved versus economic costs are beyond scope.