Thank you for visiting nature.com. You are using a browser version with limited support for CSS. To obtain the best experience, we recommend you use a more up to date browser (or turn off compatibility mode in Internet Explorer). In the meantime, to ensure continued support, we are displaying the site without styles and JavaScript.

# Understanding how socioeconomic inequalities drive inequalities in COVID-19 infections

## Abstract

Across the world, the COVID-19 pandemic has disproportionately affected economically disadvantaged groups. This differential impact has numerous possible explanations, each with significantly different policy implications. We examine, for the first time in a low- or middle-income country, which mechanisms best explain the disproportionate impact of the virus on the poor. Combining an epidemiological model with rich data from Bogotá, Colombia, we show that total infections and inequalities in infections are largely driven by inequalities in the ability to work remotely and in within-home secondary attack rates. Inequalities in isolation behavior are less important but non-negligible, while access to testing and contract-tracing plays practically no role because it is too slow to contain the virus. Interventions that mitigate transmission are often more effective when targeted on socioeconomically disadvantaged groups.

## Introduction

With around 464 million confirmed cases around the world as of March 2022, the COVID-19 pandemic has disproportionately affected disadvantaged groups. Evidence from multiple settings suggests that within each country, poor and minority groups are more likely to contract the disease1,2,3,4. In Bogotá, Colombia, we estimate in a companion paper that individuals in the lowest socioeconomic strata (SES) are 3.7 times more likely to have been infected with COVID-19 than those in the highest strata as of March 3rd 20215. Addressing inequalities has been widely recommended to tackle the pandemic6,7,8. Some studies have documented factors that are likely to affect COVID-19 transmission patterns, including access to testing and contact tracing services9,10,11,12,13,14,15, biological factors related to susceptibility and infectiousness16,17, levels of exposure at work18,19, circumstances within the household18,19,20,21, lockdown and social distancing10,14,22, along with self-isolation behavior and compliance with regulations14,23,24. These factors are likely to differ by socioeconomic status, thereby driving inequality in COVID-19 infection rates across socioeconomic groups15,18,25,26. Optimal policy design will vary significantly depending on which of these factors is key: targeted policies that focus on high-risk groups will reduce both inequalities and overall transmission more effectively if they concentrate on the most important dimensions of inequality. However, few studies have so far been able to examine which factors are associated with socioeconomic inequality in COVID-19 infections and compare how important each one is for explaining overall inequality. We thereby provide the first study of a low- or middle-income country (LMIC) setting that (i) estimates multiple differences between socioeconomic groups in characteristics that are associated with inequality in COVID-19 infections, and then (ii) incorporates these differences into an epidemiological model to tease out their implied impact on the spread of the pandemic.

## Methods

### Differences in characteristics between socioeconomic strata

We use primary data from the CoVIDA project led by the University of Los Andes. This includes the results of 59,770 RT-PCR tests in Bogotá, targeted on a mostly asymptomatic adult population from the beginning of June 2020 to March 3rd, 2021. We combine this with administrative data from the Health Secretary of Bogotá (HSB) that covers all reported cases in Bogotá (Supplementary Materials S1.1 includes a detailed data description). Both datasets include information on individuals’ socioeconomic stratum, a classification commonly used as a proxy for economic welfare in Colombia. We use this six-level measure to create four SES groups for analysis, ranging from poorest to richest: 1&2, 3, 4, and 5&6.

Together, these data allow us to estimate a set of characteristics that are associated with infection rates, and to do so separately for each of the four socioeconomic groups. Table 1 displays the values of these characteristics for each group (and Table S7 includes standard errors, 95% CIs, source and estimation method). Throughout the paper, we classify the characteristics into 4 dimensions: (i) contacts outside of the household, (ii) contacts within the household, (iii) isolation behavior, and (iv) testing and tracing.

The number of non-work-related contacts outside the home does not differ significantly across SES ($$p= 0.06$$). Secondary attack rate (SAR) for contacts outside the home also exhibits no significant differences across SES ($$p=0.20$$), with the overall average estimated at 13%. This result is consistent with our finding that self-declared protection practices are not systematically better among higher SES; lower SES even appear to compensate for their inability to remain at home by wearing masks and using antibacterials more frequently (Table S1). By contrast, there is a large and significant difference in the number of days working outside of home during the 14 days prior to the survey, which varies from 2.5 days for SES 5&6 to 6.4 days for SES 1&2 ($$p<0.001$$). This substantial difference is likely to reflect the well-documented variation in the ability to work remotely27.

Characteristics related to infections inside of homes also reveal differences. First, mean household size shows modest variation, from approximately 2.5 in wealthier households to 3 in poorer households ($$p<0.001$$ for the difference). Lower SES individuals therefore have more contacts within the household, which is known to be a particularly important setting for transmission20. There is a substantial difference in the SAR within household ($$p=0.02$$), ranging from only 10% in SES 5&6 to around 27% in SES 1–3. Corroborating this result, the positive correlation between household size and infection probability is stronger for lower SES (Figure S9b), which may partly be explained by more crowded housing conditions, since the poor have fewer rooms per household (Figure S9c).

Self-reported isolation for individuals who have been tested positive is high (86%) and does not vary significantly by SES ($$p=0.61$$). Other high-risk circumstances that require isolation, such as experiencing symptoms, lead to a substantial reduction in days worked outside of home for all groups. But richer groups are able to restrict their working activity in these circumstances significantly more than poorer groups.

Finally, access to testing and tracing could also affect infections if it leads to effective quarantine and isolation. We find differences in testing and tracing characteristics across SES. These differences may plausibly be explained by variation in health service quality that is correlated with income28. The likelihood of being detected conditional on being infected varies substantially, from 11.7% in the SES 1&2 to over 20% in SES 4, 5 and 6. There are moderate differences in the average delay in test consultations and results, which sum to 8.3 days for SES 5&6 and 9.5 days for SES 1&2. But average delays across all groups are very long. They clearly exceed recommendations for an effective Test, Trace and Isolate strategy, which suggest delays of no more than 5 days from onset of symptoms to the results of the test11.

### A theoretical model that emphasizes differences between socioeconomic groups

In order to quantify how the differences shown in Table 1 translate into differences in COVID-19 infection patterns, we use the results as inputs for a novel branching-process model of the spread of COVID-19. (See Tables S4 and S5 for a complete list and description of the model parameters.).

The model is stochastic and individual-based, building on early modelling work of the pandemic10,11,22. We structure the model by SES, allowing all parameter values found in Table 1 that are significantly different at the 5% level to be SES-specific.

An example of a transmission process is shown in Fig. 1, and a second more detailed example is found in Fig. S10. The figures demonstrate some of the realistic features of the model, which include a distinction between contacts within and outside the home, assortative mixing, symptoms, testing, contact tracing, isolation, immunity, along with realistic distributions for all stochastically-generated timings.

The key implicit assumption underlying our analyses is that each of the parameters has a direct causal effect on COVID transmission. The size of this direct effect for each parameter, and the interactions between each parameter, are determined by the specific modelling assumptions about the transmission mechanism (for example, exactly how isolation reduces transmission, the length of the incubation period, and the household structure). These assumptions are discussed in detail in Sect. S1.2. This limitation should be taken into account, particularly when interpreting counterfactual results based on varying parameter inputs.

Our baseline simulation of the epidemic uses the parameters as described in Tables S4 and S5. Figure 2 shows the infection patterns in each SES, both using data from Bogotá (panels (a) and (d)) and comparing to the results of our model in the baseline scenario. We use two variations of the model. In the first (panels (b) and (e)), the average number of out-of-home contacts for each group stays constant over the course of the epidemic, leading to a one-wave pattern. In the second (panels (c) and (f)), we account for changes in mobility over time by scaling the number of out-of-home contacts by a time-varying constant, calibrated to match total confirmed incidence (see Supplementary Materials Section S1.3.4). This constant is the same for all groups, implying that all predictions of inequality result from the differences in characteristics described in Section 2.1, rather than the calibration process.

Panel (a) displays the per capita incidence over the preceding 2 weeks for each group based on data on confirmed cases from the Health Secretary of Bogotá. We see evidence of inequality between groups, particularly in the first wave, where SES 1&2 reaches a peak incidence rate of 0.72%, around double the level of SES 5&6. The model predictions of confirmed cases, seen in panels (b) and (c), match this observed pattern relatively well, with more detection of cases in the lower SES. The observed pattern of detected cases lies within the range of the confidence intervals of both models for all groups in first wave (Fig. S11).

We also show that the model predictions match estimated inequalities in true infections, estimated with the CoVIDA data, which is a sample of mostly asymptomatic individuals and is thus less likely to be biased due to differential propensity to be tested. Panel (d) shows that the inequality in estimated true infections is even starker than that of confirmed cases: cumulative per capita incidence (during the entire period of the study) varies from 69% in strata 1&2 to only 19% in strata 5&6. Because the model with no mobility change (panel (e)) only captures the first wave of the epidemic, it underestimates the cumulative incidence rate in all groups, but it yields a prediction of the proportion of total cases that come from each SES (a proxy of inequality) that matches the CoVIDA data well (Fig. S12). When accounting for mobility changes (panel (f)), the model gives very similar predictions to the CoVIDA data estimations, despite mobility being calibrated on only aggregated confirmed cases, with a difference of 41 percentage points in cumulative per capita incidence between the lowest and highest SES. Broadly, we are able to predict the macro-level differences in infections well by introducing micro-level inequality into our epidemiological model.

### Ethics statement

All methods were carried out in accordance with relevant guidelines and regulations. All experimental protocols were approved by Universidad de Los Andes and the Secretaría de Salud de Bogotá. Ethics approval was obtained from the ethics committee of Universidad de los Andes (Act number 1278 of 2020). The ethics committee approved that the participants would receive the information via telephone and give their informed verbal consent, in order to comply with physical distancing and limit the restriction for a study is part of a public health surveillance strategy implemented jointly with the Health Secretary of Bogotá.

## Results

### Effects of reducing inequality on virus incidence

In order to identify the key channels that are associated with overall inequality in infections between SES, we examine the effects of simulating a reduction in inequality along 4 dimensions: (i) contacts outside of the household, (ii) contacts within the household, (iii) isolation behavior, and (iv) testing and tracing. For each of these dimensions, we first simulate a “100% upward adjustment” scenario, in which the characteristics of all SES are set at the level of the highest SES (5&6), and then a “50% upward adjustment” scenario (Fig. 3), in which the differences with respect to SES 5&6 are reduced by half.

First, we find that in a simulation in which all SES have as few contacts outside the home as SES 5&6, the epidemic collapses, with a median cumulative incidence of less than 1% across all SES (column 2). Under the assumptions of the model, if every SES had the ability to work on average only 2.5 days every 2 weeks outside of home, then the $${\mathcal {R}}_t$$ would fall below one, leading to a containment of the virus from early stages. When we adjust by only 50%, there is still a marked reduction both in infections across all groups and in the inequality between groups. The lower bound on these simulations always includes a value less than 1%, since there are simulations in which the epidemic dies out in the early stages. This indicates that the differences in out-of-home contacts are strongly associated with overall inequality in COVID-19 infections between groups.

Second, attributing the within-home characteristics of strata 5&6 to all strata (column 3) leads to a reduction in infections that is as strong as the effect in the out-of-home scenario. Further analysis demonstrates that this effect is mostly driven by differences in the within-household SAR, while inequality in household size plays a significant but smaller role (Fig. S9). If the assumptions of the model are correct, there would be large potential benefits to policies that reduce within-home transmission for groups with crowded housing conditions, for example through recommending mask usage and social distancing within the home.

Third, we consider scenarios in which lower SES are just as able to isolate as SES 5&6 in high-risk circumstances (being in contact with or in the same household as a known case, or when presenting symptoms) (column 4). This leads to moderate simulated reductions in infections and inequalities: for example, cumulative incidence among SES 1&2 is reduced by 8 percentage points in the 100% scenario. Differences in isolation behavior are thus important, but are less associated with overall inequality in infections than the two previous channels. Reducing inequalities in isolation may nevertheless be more tractable for policy than changing characteristics like housing conditions or job-types. For example, using financial compensation to enable lower-income individuals to stay at home when symptomatic may be an effective strategy.

Finally, the simulated effect of improving access to testing and contact tracing among low SES to the level of SES 5&6 has an effect on infections that is not significantly different from 0 (column 5). This is true despite the substantial inequality in access (see in Table 1). The absence of effect can be explained by the fact that, on average in Bogotá, delays in accessing testing, receiving results, and being contact traced are so severe across all groups that testing and tracing has little effect at all in mitigating the spread of the virus in this context.

To further examine the role of socioeconomic inequality, we simulate the effect of a pure reduction in inequalities. Specifically, we reduce the dispersion of all the characteristics that were found to be significantly different across SES while preserving the mean of each variable (Fig. 4). We find that if these inequalities are fully collapsed, total infections would be reduced from 38.2% to 35.9% of the population. The simulated effect is moderate but statistically significant ($$p<0.001$$). Inequality in and of itself is associated with more widespread infection, even when holding constant the average characteristics of the population.

### Alternative policy scenarios

We next examine simulations of policy-style scenarios that indicate (i) what types of policies may be effective in combating the epidemic, and (ii) whether targeting policies on low socioeconomic groups can reduce the spread of the virus. Figure 5 shows the main results, while Fig. S17 shows the results when varying the intensity of each policy. The first set of simulations describes the results of (i) 10% population immunity (e.g., due to vaccinations at an early stage), (ii) a reduction of 1 in outside-home contacts (e.g., due to restricting economic activity, or a policy facilitating or enforcing remote work), and (iii) an increase in ability to isolate (e.g., due to financial support for those required to isolate). Increasing immunity and reducing out-of-home contacts lead to large reductions in simulated infections; increasing isolation leads to more modest effects. Even when holding constant the number of beneficiaries, reductions are between 28% and 49% larger when targeted on the lowest SES ($$p<0.001$$ for all differences). If the assumptions of the model hold, policies that target socioeconomically disadvantaged populations are likely to be substantially more effective.

By contrast, increased access to testing for symptomatic individuals has only a small impact on total infections, and this impact does not differ significantly if targeted or not. Access to tests is not associated with lower infection rates in Bogotá, likely because of severe delays across all SES, and because initial testing coverage is limited.

In keeping with this claim, even a complete removal of testing and tracing (the “No testing” scenario) only increases the overall simulated incidence rate by a mean of 1.0 percentage points ($$p < 0.001$$), while increasing the speed of testing to an average of 2 days from symptoms to detection results in a modest 4 percentage point reduction in incidence rate from the baseline scenario. Combining this “Fast testing” scenario with an increase in the probability of being tested leads to more substantial reductions in transmission, with the overall incidence rate reducing to below 30%. Targeting the improvements in self-testing on SES 1&2 leads to an additional mean reduction of 2.7 percentage points compared to the non-targeted scenario (p-value of the difference$$=0.04$$).

## Discussion

This paper documents differences across SES along multiple dimensions that are relevant to the spread of COVID-19. We provide one of the first studies that estimates the relative importance of each form of inequality for explaining disparities in COVID-19 infections. Such estimates are important to improve policy design. Disparities in types of jobs and the ability to work from home are shown to be a key factor associated with inequalities. Important channels that have seen less emphasis in existing research include within-home SAR and the ability to isolate when required (in particular when one has symptoms, a detected individual in the household, or a recent contact with a person tested positive). Finally, while poor individuals do have substantially less access to tests, and a lower chance of being detected and traced, this does is not associated with inequalities in incidence rates, probably because testing and tracing in Bogotá is too slow to contribute to the mitigation of the virus.

We find that simulating an improvement in the conditions of lower SES translates to incidence reductions that are approximately proportional across all groups. Our model allows for assortative mixing, in which each group is somewhat more likely to contact their own group. Despite this, even the highest SES benefit from improvements of the conditions faced by the lowest SES. Our results suggest that worse conditions for disadvantaged groups are associated with worse outcomes for all.

Even when maintaining the same mean characteristics across the whole population, a reduction in inequality reduces the simulated spread of the virus. In addition, simulated policies are more effective at reducing transmission when they target disadvantaged socioeconomic groups. There may be similar benefits of targeting for real-world policies. More generally, some factors in the model are more strongly associated with transmission reductions. Our results suggest placing a particular emphasis on (i) maximising the ability to work remotely for lower socioeconomic groups where possible, or temporary and targeted economic shutdown measures in the absence of other alternatives, and (ii) raising awareness that within-home infections are a major source of transmission, but that this transmission may be avoidable, and may be mitigated using within-house preventative measures such as mask use. Immediate financial compensation for individuals required to isolate, including close contacts, housemates of infected persons, and anyone with COVID-19-related symptoms, may also be a tractable policy lever. Our model shows a negligible relationship between inequality in infections and test-and-trace access. This suggests that although testing systems provide valuable information about the spread of the virus, they may be so slow that they have little mitigating effect on the transmission. In such cases, policymakers must be transparent about testing delays, and consider the possible benefits of a dramatic speed-up of the testing system if this is feasible.

There are a number of important limitations of this study. First, there may be other unobserved variables that are correlated with both the model inputs and the outcome variable (virus incidence). Our model makes assumptions to infer how virus transmission would change if the model inputs were to change. But if these assumptions are wrong, or fail to take into account other correlated factors, then these inferred causal relationships will be biased.

For example, our model does not incorporate differences in the age distribution between SES. As shown in Table S6, the lower SES have a younger population and, on average, younger individuals tend to have higher COVID-19 positivity rates. Since we do not explicitly incorporate these age differences between SES in our model, we may be overestimating the role of the model factors in explaining inequality in infections between groups. There may be a number of other unobserved factors that cause a similar bias, such as individual preventative measures (e.g., mask use, handwashing), use of public transport, transmission network structure, and the role of schools and children. Nevertheless, Fig. S12 indicates that the model predicts overall inequality well.

Second, the correlations between unobserved and observed variables may vary over time. This implies that any bias in our results may not be constant over the study period. It also limits our ability to generalise our results to the phase of the pandemic after our study period, particularly as we focus on the early period of the pandemic when lockdown restrictions were particularly severe. For example, our model does not account for differential effects across SES of subsequent mobility changes.

Third, the counterfactual scenarios we examine do not account for endogenous mobility reactions, leading to overly extreme results in which the virus is completely contained. Since the model starts in the conditions of generalised lockdown in Bogotá, with $$R_q = 1.22$$, any scenario sufficient to reduce this number to below 1 will lead to a total containment of the epidemic. In reality, mobility restrictions may have been loosened sooner if this was the case, leading to more infections than predicted by the model.

Fourth, the CoVIDA data we use is not fully representative of the Bogotá population, although it is one of the most comprehensive datasets available in Latin America, including both PCR and survey results.

Fifth, we do not examine infection mortality, which may differ by SES group. Inequality in mortality may be more extreme than inequality in infections (e.g., because of better quality hospital treatment in higher SES) or less extreme (e.g., because higher SES are older on average). Developing a model that combines differences between SES with age variation and mortality is a promising avenue for future research.

Sixth, our baseline model only accounts for a one wave epidemic. This permits a clear-cut examination of the drivers of inequality, but does not account for immunity effects that become important in later waves. When using a model that allows for mobility change, the upward adjustment scenarios generate two epidemic waves (Fig. S16), in which any reductions in infections in the first wave are offset by lower immunity levels that lead to much larger second waves. Such a model is not necessarily more realistic, since it assumes that mobility would not have been restricted even in cases of extremely high incidence rates in the second wave. Future research should therefore prioritise models that incorporate both inequality and endogenous mobility reactions. Nevertheless, the results imply that in the absence of widespread vaccination, measures that reduce infections may only delay an epidemic rather than prevent it.

Our findings provide new evidence on the importance of different channels that are associated with inequalities in COVID-19 infections. Our model simulations suggest that improving the circumstances of the most disadvantaged groups, including by targeting interventions on the poor, may have benefits for all. Socioeconomic inequalities should be taken into account in order to better handle both the COVID-19 crisis and potential future epidemics.

## Data availibility

All the code and data used for this study is available at the following GitHub repository: https://github.com/dmbwebb/covid_model.

## References

1. Kim, S. J. & Bostwick, W. Social vulnerability and racial inequality in COVID-19 deaths in Chicago. Health Edu. Behav. 47, 509–513 (2020).

2. Figueroa, J. F., Wadhera, R. K., Lee, D., Yeh, R. W. & Sommers, B. D. Community-Level Factors Associated With Racial And Ethnic Disparities In COVID-19 Rates In Massachusetts. Health Affairs (2020).

3. Millett, G. A. et al. Assessing differential impacts of COVID-19 on black communities. Ann. Epidemiol. 47, 37–44 (2020).

4. Mena, G. E. et al. Socioeconomic status determines COVID-19 incidence and related mortality in Santiago, Chile. Science 372 (2021).

5. Laajaj, R. et al. COVID-19 spread, detection, and dynamics in Bogota. Colombia. Nat. Commun. 12, 1–8 (2021).

6. The Associated Press. Biden adviser Marcella Nunez-Smith says race central to virus fight. https://www.nbcnews.com/news/nbcblk/biden-adviser-marcella-nunez-smith-says-race-central-virus-fight-n1249710.

7. Ahmed, F., Ahmed, N., Pissarides, C. & Stiglitz, J. Why inequality could spread COVID-19. Lancet Public Health 5, e240 (2020).

8. Stiglitz, J. Conquering the great divide. International Monetary Fund (2021). https://www.imf.org/external/pubs/ft/fandd/2020/09/COVID19-and-global-inequality-joseph-stiglitz.htm.

9. Klinkenberg, D., Fraser, C. & Heesterbeek, H. The effectiveness of contact tracing in emerging epidemics. PloS ONE 1, e12 (2006).

10. Kucharski, A. J. et al. Effectiveness of isolation, testing, contact tracing, and physical distancing on reducing transmission of SARS-CoV-2 in different settings: a mathematical modelling study. Lancet Inf. Diseases 20, 1151–1160 (2020).

11. Kretzschmar, M. E. et al. Impact of delays on effectiveness of contact tracing strategies for COVID-19: A modelling study. Lancet Public Health 5, e452–e459 (2020).

12. Aleta, A. et al. Modelling the impact of testing, contact tracing and household quarantine on second waves of COVID-19. Nat. Human Behav. 4, 964–971 (2020).

13. Grassly, N. C. et al. Comparison of molecular testing strategies for COVID-19 control: A mathematical modelling study. Lancet Inf. Diseases S1473309920306307 (2020).

14. Flaxman, S. et al. Estimating the effects of non-pharmaceutical interventions on COVID-19 in Europe. Nature 584, 257–261 (2020).

15. Schmitt-Grohé, S., Teoh, K. & Uribe, M. COVID-19: Testing Inequality in New York City. NBER Working Paper (2020).

16. Byrne, A. W. et al. Inferred duration of infectious period of SARS-CoV-2: Rapid scoping review and analysis of available evidence for asymptomatic and symptomatic COVID-19 cases. BMJ Open 10, e039856 (2020).

17. McAloon, C. et al. Incubation period of COVID-19: A rapid systematic review and meta-analysis of observational research. BMJ Open 10, e039652 (2020).

18. Brandily, P., Brébion, C., Briole, S. & Khoury, L. A poorly understood disease? the impact of COVID-19 on the income gradient in mortality over the course of the pandemic. Eur. Econ. Rev. 140, 103923 (2021).

19. Almagro, M., Coven, J., Gupta, A. & Orane-Hutchinson, A. Racial disparities in frontline workers and housing crowding during COVID-19: Evidence from geolocation data. Federal Reserve Bank of Minneapolis Working Paper (2020).

20. Lee, E. C., Wada, N. I., Grabowski, M. K., Gurley, E. S. & Lessler, J. The engines of SARS-CoV-2 spread. Science 370, 406–407 (2020).

21. Jing, Q.-L. et al. Household secondary attack rate of COVID-19 and associated determinants in Guangzhou, China: A retrospective cohort study. Lancet Infectious. Diseases 20, 1141–1150 (2020).

22. Hellewell, J. et al. Feasibility of controlling COVID-19 outbreaks by isolation of cases and contacts. Lancet Global Health 8, e488–e496 (2020).

23. Smith, L. E. et al. Factors associated with adherence to self-isolation and lockdown measures in the UK: A cross-sectional survey. Public Health 187, 41–52 (2020).

24. Williams, S. N., Armitage, C. J., Tampe, T. & Dienes, K. Public perceptions and experiences of social distancing and social isolation during the COVID-19 pandemic: A UK-based focus group study. BMJ Open 10, e039334 (2020).

25. Eslava, M., Becerra, O., Cárdenas, J.-C., Isaacs, M. & Mejía, D. The socioeconomic patterns of COVID outside advanced economies: the case of Bogotá. Documento CEDE 2020-45 (2020).

26. Rocha, R. et al. Effect of socioeconomic inequalities and vulnerabilities on health-system preparedness and response to COVID-19 in Brazil: A comprehensive analysis. Lancet Global Health (2021).

27. Dingel, J. I. & Neiman, B. How many jobs can be done at home?. J. Public Econ. 189, 104235 (2020).

28. Cifuentes, M. P., Rodriguez-Villamizar, L. A., Rojas-Botero, M. L., Alvarez-Moreno, C. A. & Fernández-Niño, J. A. Socioeconomic inequalities associated with mortality for COVID-19 in Colombia: a cohort nationwide study. J Epidemiol Community Health (2021).

## Acknowledgements

The authors acknowledge generous support from the Interamerican Development Bank, the Development Bank of Latin America (CAF), the University of Los Andes and the Universidad Nacional de Colombia. Duncan Webb gratefully acknowledges support from the ED 465 at University Paris 1, and the EUR project ANR-17-EURE-0001. The authors declare no conflicts of interest. We thank members of the CoVIDA working group, Yessica Daniela Campaz Landazabal, Marylin Hidalgo, Juliana Quintero, Paola Betancourt, Leonardo Leon, Pablo Rodríguez, Diana Sofia Rios Oliveros, Andrés Felipe Patiño, Jose David Pinzon Ortiz, Maribel Rincón, Alejandro Segura, Leonardo Salas Zapata, Philip Keefer, Diane Moyano, Elkin Osorio, Rodrigo Rodriguez.

## Author information

Authors

### Contributions

Conceptualization: R.L. and D.W. Formal analysis: R.L. and D.W. Project administration and Supervision: R.L., E.B., R.B., G.B., A.G., L.J.H., A.R.V., S.R. and M.V. Data Curation: D.A. and C.DLR Writing - original draft: R.L. and D.W. Writing - review & editing: all authors critically revised the various versions of the manuscript.

### Corresponding author

Correspondence to Rachid Laajaj.

## Ethics declarations

### Competing interests

The authors declare no competing interests.

### Publisher's note

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

## Rights and permissions

Reprints and Permissions

Laajaj, R., Webb, D., Aristizabal, D. et al. Understanding how socioeconomic inequalities drive inequalities in COVID-19 infections. Sci Rep 12, 8269 (2022). https://doi.org/10.1038/s41598-022-11706-7

• Accepted:

• Published:

• DOI: https://doi.org/10.1038/s41598-022-11706-7