The unequal effects of the healtheconomy trade-off during the … – Nature.com

We build a data-driven, granular ABM of the New YorkNewarkJersey City, NYNJPA metro area. The main agents of the model are the 416,442 individuals of a synthetic population that is representative of the real population across multiple socioeconomic characteristics, including household composition, age, income, occupation and possibility to work from home (WFH) (for a schematic representation, see Fig. 1, and for a detailed description of the model, see Methods and Supplementary Information).

The economic module depicts the flow of goods and services between industries and from industries to final consumers (inputoutput network), while the epidemic module tracks pathogen exposure at workplaces, community/consumption venues, schools and households (contact network). Agents display high heterogeneity across various socioeconomic characteristics (see box). The economic and epidemic modules are closely linked: the epidemic module impacts economic outcomes by reducing consumption due to infection fear, while the economic model influences epidemic spread by altering workplace and community contacts through employment changes in different industries.

The epidemic module of the ABM is built on the contact network that connects synthetic individuals. This network has multiple layers, where each layer captures interactions occurring (1) in the household, (2) in school, (3) in the workplace and (4) in the community (during on-site consumption, such as in shops, restaurants or movie theatres). Epidemic propagation occurs on these networks, built via anonymized, privacy-enhanced mobility data from opted-in users, which inform workplace and community interactions. These data, collected through a General Data Protection Regulation (GDPR)-compliant framework by Cuebiq, provide daily workplace visitation patterns and estimates for colocation probabilities in community spaces, based on a Foursquare dataset. The ABM employs a stochastic, discrete-time disease transmission model on the contact network and synthetic population, with individuals transitioning between epidemic states based on key time-to-event intervals (for example, incubation period, generation time and so on) derived from SARS-CoV-2 transmission data.

From an economic point of view, individuals play a role both as workers and consumers. They work in one of multiple industries, producing goods and services that are either sold to other industries as intermediate products or sold to final consumers as consumption products. The economic module specifically emphasizes employment and consumption. In particular, hiring and firing decisions are driven by industry workforce requirements, closures of economic activities and possibility of remote work. Consumption patterns vary among agents based on age and income, and dynamically adjust in response to the evolving state of the pandemic. Specifically, households tend to curtail their demand for services from customer-facing industries because of the fear of the disease (customer-facing industries are entertainment, accommodationfood, other services, retail, transportation, health and education; Supplementary Section 3.1.3). The model also considers the inputoutput network of intermediates that industries use to produce final goods and services18, leading to the propagation of COVID-19 shocks to the entire economy.

We calibrate the models key parameters, including the parameter regulating the behavioural changes, dubbed fear of infection, to fit crucial epidemic and economic statistics from the first wave in the NY metro area (Supplementary Section 4). Epidemiological parameters are adjusted to fit ancestral SARS-CoV-2 lineages (Supplementary Table 5). Simulations start on 12 February 2020, protective measures are imposed on 16 March and relaxed on 15 May 15, and simulations end on 30 June. As protective measures, we close schools, mandate WFH and shut down all non-essential economic activities, such as entertainment and most of the accommodationfood industry, but also large parts of manufacturing and construction. We use the official NY regulations to estimate the degree to which a given industry is essential (Supplementary Section 3.2.1) and assume that workers who can WFH are not directly affected by these closures2. We name this set of assumptions the empirical scenario.

Our model accurately matches the six official economic statistics we calibrated it on (Fig. 2a). It correctly reproduces the fact that employment declined more strongly than gross domestic product (GDP) (this is because industries most affected by shutdown orders produce less output per worker). It also correctly reproduces the fact that consumption of goods and services produced by customer-facing industries declined more strongly than consumption of goods and services produced by industries that are not customer facing, ranging from manufacturing products to utilities and financial services (Supplementary Fig. 15).

a,b, Statistics that were directly targeted in the parameter calibration. a, The percentage change from OctoberDecember 2019 (2019Q4) to AprilMay 2020 (2020Q2) across six official economic statistics, in the model and in the data (Supplementary Section 4). Here, and throughout the paper, we report the mean and error bars (2.597.5 percentile range) across simulation runs that differ by stochastic factors (Methods). b, A comparison between model and data for the number of COVID-19 deaths, which is the key epidemic statistic that we targeted. ce, Validation results for statistics that were not directly targeted. c, Employment in April 2020 as a percentage of employment in February 2020, across the main two-digit NAICS industries, in the model and in the data. The circle size is proportional to employment in February 2020. d, Employment and consumption, in the model and in the data3, among low-income and high-income households (low income <$27,000 and high income >$60,000; these bands are chosen for comparison to real data; Supplementary Section 5.1.1). e, The ratio between community contacts with infectious individuals (Supplementary Section 5.2) before and after the imposition of protective measures, in the model and in the data, for the seven customer-facing industries (Supplementary Section 3.1.3). The circle size is proportional to the share of pre-pandemic contacts.

Our model can also be validated against empirical properties that were not directly targeted in the parameter calibration procedure (Fig. 2c,d). First, thanks to our estimate of pandemic shocks2 and shock propagation model17, we are able to recover industry-specific changes in employment induced by the pandemic (Fig. 2c), with a Pearson correlation coefficient of 0.82 (P value 2104) between model and data. This accuracy is in large part due to our models ability to take industry-specific estimates as inputs, but also to the propagation mechanisms embedded in the model. Indeed, the estimates by supply shocks alone have only a Pearson correlation of 0.69 (P value 4103, Supplementary Fig. 14). Second, thanks to our granular and data-driven characterization of employment and consumption patterns (Supplementary Figs. 8 and 12), we reproduce a key fact: low-income individuals were more likely to become unemployed but reduced consumption less than high-income individuals3,19 (Fig. 2d). This happens because low-income individuals are more likely to work in the occupations most affected by closures, such as food preparation and serving, building and grounds cleaning and personal care and service (Supplementary Fig. 16), but they spend a larger share of their income on essential goods and services such as housing and utilities (we do not consider here the effect of the stimulus programme, which would further increase the spending of low-income individuals).

On the epidemic side, our model correctly matches the death count data on which it has been calibrated, correctly replicating the spike in the number of reported deaths in April 2020 and the strong reduction in June (Fig. 2d and Supplementary Fig. 18). It also correctly estimates the changes in contact patterns that occurred after protective measures were implemented, although these data were not used for parameter calibration (Fig. 2e). Both in the model and in the data, community contacts substantially reduced (Pearson 0.75 and P value 0.05), more in mostly non-essential industries such as entertainment and restaurants than in mostly essential industries such as retail and health. We also accurately estimate the reduction in workplace contacts across industries (Supplementary Fig. 20; Pearson 0.88 and P value 5106), the temporal profile of reduction in contacts (Supplementary Fig. 19) and the increase in prevalence over time (Supplementary Fig. 18). Finally, the model makes a number of estimates about how many infections happen across each layer and industry over time, as well as which occupation, income and age groups are most affected (Supplementary Figs. 2123). While we are not able to find data to quantitatively evaluate these estimates, our literature review provides some support to these findings (Supplementary Section 5.2.1).

In our analysis, we quantitatively explore the effects of three key factors that shaped the behavioural and policy response to the first COVID-19 wave. In the following, we use baseline to refer to estimated parameters calibrated with empirical data.

As a first set of counterfactual scenarios, we explore the effect of adjusting the magnitude of the fear of infection parameter that regulates behaviour change. Generally, we treat this parameter as uniform across individuals, per survey evidence20. However, we also consider an age-specific fear counterfactual. Our baseline calibration yields a fear of infection parameter distribution (Supplementary Fig. 13), implying a 14% consumption demand reduction in customer-facing industries due to infection fear at the epidemic peak. This calibrated value, which merges NPI effects with behaviour change, cannot be causally interpreted. Absence of NPIs would necessitate a steeper consumption drop for the model to explain observed behavioural changes, thus estimating higher infection fear. But the lack of real-world data without NPIs hampers this estimate. Instead, we explore two counterfactuals: low (0.1 times baseline) and high (10 times baseline), translating to a 1% and 77% consumption reduction due to fear at the peak, respectively. These scenarios facilitate comparing stronger fear of infection effects with stricter NPIs.

Next, we vary two policy-related factors. First, we experiment with different economic activity closures. Besides the baseline scenario with all non-essential industries closed, we consider two milder closure scenarios: (1) only non-essential customer-facing industries are closed and (2) no closures, with all economic activities open. Second, we simulate protective measures starting either 4weeks earlier (17 February 2020) or 2weeks later (30 March 2020). Additional counterfactuals, including partial closure of customer-facing industries and no WFH or school closures, are explored in Supplementary Information (Supplementary Figs. 24 and 25).

Aggregate economic and epidemic results are shown in Fig. 3, while results disaggregated by income, geography and industry are shown in Fig. 4 (Supplementary Figs. 2733). Figure 3a conveys our first main result: stricter closure of economic activities and higher fear of infection both lead to increased unemployment and fewer COVID-19 deaths. To illustrate this, consider a scenario with baseline fear of infection and all economic activities open, represented by the light-coloured circle. If we maintain the fear of infection at the baseline level, but close all non-essential economic activities (as in the baseline scenario), unemployment surges by 64%, while the number of deaths drops by 35%. Likewise, if we instead keep the closure level at the empirical baseline but increase fear of infection (represented by the dark triangle), we see a 40% rise in unemployment and a 50% decrease in deaths relative to the empirical scenario. Similar trends are observed in other scenarios. Although the total death count and average unemployment can vary substantially across simulation runs, the relative impacts of different policies remain robust (Supplementary Fig. 26).

a,b, Deaths and unemployment across scenarios. For each scenario, we show the aggregate unemployment rate and the cumulative number of deaths, as averaged throughout the simulation period and the simulation runs (Supplementary Fig. 26 shows the variability across simulation runs and discusses its interpretation). The empirical scenario is highlighted to serve as a benchmark. Scenarios are distinguished by the strength of behaviour change, as exemplified by the fear of infection parameter (square: low; circle: baseline; triangle: high). a, Scenarios are further distinguished by the specific closure of economic activities (all non-essential industries, as occurred empirically, only customer-facing industries and no closures), keeping the start of protective measures fixed at the baseline, empirically observed date. b, Scenarios are further distinguished by the start of protective measures (baseline: 16 March 2020, as empirically; early: 17 February 2020 and late: 30 March 2020), keeping closures fixed at all non-essential industries. c, For the specific combination of high fear of infection and three different starts of protective measures, a time series of unemployment and deaths corresponding to the three scenarios is shown.

a, Workplace infections and unemployment across income classes. Top: we vary the level of closures, keeping fear of infection and the start of protective measures to their baseline values (so the case with all non-essential activities closed is the empirical scenario). Bottom: we vary fear of infection keeping all economic activities open and starting protective measures on the baseline date. b, Maps of unemployment across census tracts in New York City, corresponding to two scenarios in a, including the empirical scenario (asterisk) and the counterfactual with no closures and low fear (hash). c, Infections and reduction in consumption across five selected industries and three levels of closures, for baseline fear and start of protective measures.

Both higher fear of infection and stricter closures lead to saving lives at the expense of jobs, for low and high income workers alike (Fig. 4a). However, for low income workers, higher fear of infection or stricter closures have a larger effect, leading to more lives saved and more jobs lost, compared with high-income workers. As we will show later, outside the household setting, most infections occur in customer-facing industries, where most low-income workers are concentrated. Thus, mandated closure or spontaneous avoidance of these industries leads to both more unemployment and fewer workplace infections among low-income workers.

The unequal economic outcomes of the empirical scenario also lead to geographical disparities. Figure 4b shows two maps of unemployment in Manhattan in the empirical scenario (asterisk) and in a counterfactual with low fear and no closures (hash). We see that in the counterfactual, the unemployment rate is very evenly spatially distributed, while in the empirical scenario, low-income areas such as the Queens and the Bronx have a high unemployment rate of more than 20%, compared with high-income areas such as Manhattan, with unemployment rates around 15%.

Overall, these results contribute to the ongoing debate on the relative effectiveness of behavioural change versus NPIs in preserving both public health and the economy. While it is intuitive to expect stricter mandated NPIs to increase unemployment and decrease COVID-19 deaths, it is less apparent that heightened behavioural adaptation would yield similar results (Discussion). Our findings highlight a qualitative parallel between substantial behavioural change and stringent economic activity closures. Spontaneous avoidance of services offered by customer-facing industries, akin to their mandated closure, results in increased unemployment but fewer fatalities. This trend is particularly pronounced among low-income individuals.

Our model also evaluates the efficacy of closing all non-essential economic activities, including large segments of manufacturing and construction, compared with exclusively closing customer-facing industries. We find that the mandated closure of all non-essential activities only marginally decreases deaths compared with solely closing customer-facing industries, but it drastically increases unemployment. In comparison with the baseline scenario, a counterfactual that only closes customer-facing industries results in a slightly higher death rate (4% higher), but substantially mitigates unemployment, reducing it by 36%. To explain these results, consider Fig. 4c. Most infections occur in customer-facing sectors such as entertainment and accommodationfood. Their closure curbs infections considerably but also consumption and employment. Conversely, closing manufacturing or construction marginally impacts infections but drastically reduces consumption. Professional services remains largely unaffected also because of WFH adaptability.

Methodologically, these industry-specific results were obtained because we associated each consumption venue from mobility data to an economic activity, allowing the quantification of industry-specific contacts. This granular, data-driven approach provides insights that more aggregate, qualitative models might overlook.

Another counterfactual exploration concerns the effectiveness of starting epidemic mitigation and control earlier (4weeks before) or later (2weeks later) than in the empirical scenario. As we show in Fig. 3b, delaying these measures marginally reduces unemployment by 2% but causes a notable 50% rise in deaths. In a high fear-of-infection scenario, late measures result in both a 46% increase in deaths and a 12% rise in unemployment. The mechanism for these results is suggested in Fig. 3c, where we show time series across the three counterfactuals with high fear of infection. We see that an early start of mitigation measures prevents an epidemic wave, leading to no further increase in unemployment due to fear of infection. Conversely, with a baseline or late start, substantial behaviour change leads to reduced consumption, and this, in turn, leads industries to fire their employees, increasing unemployment. Thus, starting mitigation measures early is crucial to improve epidemic outcomes, and possibly economic outcomes too. Our preliminary investigation (Supplementary Section 6.4) also shows that with an early start of protective measures, it is possible to avoid an excessive burden on the healthcare system, as measured by a usage of more than 50% of the nominal capacity of intensive care units.

In the empirical scenario, the parameterization of the fear of infection is uniformly applied across individuals, as suggested by survey evidence20. However, we also examined a counterfactual in which young individuals adopt less behavioural changes (low fear of the disease) than older individuals (high fear of the disease), considering this might be a more optimal situation for pandemic control through behavioural change. Here, at-risk older individuals would internalize the infection risk more, while younger individuals, less likely to suffer severe consequences, could maintain higher consumption and contribute to herd immunity. We explore how this scenario plays out quantitatively in the data-driven, granular agent-based model.

To explore the effects of heterogeneous behavioural changes, we group all households into three classes based on the age of their household head (034, 3564 and 65+years). We assume that fear of infection in each class is proportional to the risk of death in that class (Supplementary Section 6.5). We also normalize the fear of infection parameter across age bracket so that the mean takes the same baseline value that we considered in the empirical scenario. This normalization ensures that results are driven by a different distribution of fear across age groups, rather than by changes in overall fear. At the end of this procedure, the fear level among households aged 034years is 0.02 times the baseline, households aged 3564years have a fear level of 0.48 times the baseline and households aged 65+years have a considerably higher fear level of 4.91 times the baseline. We compare the age-specific scenario with the scenario in which all agents have uniform baseline fear (as in the empirical scenario).

We report aggregate results in Fig. 5a, which reproduces the same scenarios as Fig. 3 for uniform fear, next to the new results for age-specific fear. Adjusting for age-specific fear, while keeping other factors constant, marginally reduces both unemployment and deaths compared with uniform fear. In the all open scenario, where the effect is most pronounced, age-specific fear reduces deaths by 6% and unemployment by 5%. For comparison, closing customer-facing industries cuts deaths by 28% but increases unemployment by 22%.

a, Aggregate deaths and unemployment across scenarios. The general interpretation is the same as Fig. 3; here, uniform fear is represented by circles and age-specific fear is represented by plus symbols. b, For the scenario all openearly start, time series of the level of workplace and community contacts and consumption demand of customer-facing industries, disaggregated by type of fear (solid lines, age-specific fear and dashed lines, uniform fear) and by age groups with heterogeneous fear. These time series show how fear of infection reduces contacts and consumption demand. c, For the same scenario as b, the time series of infections per 1,000 individuals disaggregated by age groups and by whether they occurred outside or within households, and actual consumption demand (relative to the pre-pandemic level) disaggregated by industry and by whether industries are customer facing or not.

In Fig. 5b,c, we examine industry- and age-specific effects, focusing on ages 034years and 65+years. First, in Fig. 5b, we show how fear of infection reduces contacts and consumption demand. As expected, uniform fear leads to equal reductions across all ages by construction, while age-specific fear instead leads to the least reduction in young agents and the most in older agents. Total consumption decreases less than contacts as it may not require direct contact, such as ordering takeaway food (Methods).

In Fig. 5c, we first consider infections (left plots). We distinguish infections occurring inside households from those happening outside (community or workplace contacts). Outside the household, infections among older agents decrease with age-specific fear compared with uniform fear, especially around the epidemic peak where they are 30% lower. However, in the waning phase of the epidemic, infections are comparable in both scenarios. In contrast, we see an opposite trend among young households, where a large number of infections happen later due to their very low fear of infection. Within households, the differences between age-specific and uniform scenarios are less pronounced.

The relatively small decrease in deaths with age-specific fear can be explained in two ways. On the one hand, the time series of reductions in contacts and infections show that older individuals drastically reduce their contacts only after the epidemic peak. This delay results from the lag between infection and death reporting; behaviour change intensifies when individuals become aware of the number of COVID-19 deaths. On the other hand, older individuals cannot avoid infections within their own households.

As we can also see in Fig. 5c (right plots), age-dependent fear of infection alters consumption demand across industries. Consumption demand decreases more in health, a customer-facing industry on which old agents spend a disproportionate amount of income, and less in accommodationfood, on which young agents spend a higher share of their income. At the same time, consumption demand increases towards industries that are not customer facing because households reallocate part of their consumption budget to those industries. As individuals in older age groups decrease consumption more and thus have more budget to reallocate because of higher fear of infection, this results in higher consumption demand towards industries that have high consumption share among old individuals, such as finance. By contrast, because younger individuals spend a large fraction of their income on real estate and, with low fear, they do not reallocate much, the increase in demand for real estate services is lower than if fear of infection was uniform across ages.

In summary, these findings show that even when individuals adjust their behaviours in response to their personal risk levels during a pandemic, it only modestly affects health and economic outcomes. Moreover, our results quantify the complex ripple effects across various sociodemographic groups.

Original post:

The unequal effects of the healtheconomy trade-off during the ... - Nature.com