Poverty, inequality, and COVID-19 in the US
Recent research has pointed to multiple ways in which the pandemic might impact poverty and inequality (for example, Furceri et al. 2020, Kikuchi et al. 2020, Lakner et al. 2020, and Palomino et al. 2020). There is also the reverse effect: poverty and inequality may matter to viral responses and outcomes. Epidemiology is the obvious place to look for theory and evidence on this. However, the words ‘poverty’ and ‘inequality’ are surprisingly hard to find in the classic epidemiological texts or the standard models used in understanding and predicting the spread of infections in human populations. The literature in the social sciences does, however, suggest many ways in which socioeconomic variables might be expected to matter to human vulnerability to infections.
It is well documented that many of the risk factors associated with the severity of COVID-19 are correlated with income. Poor people often have a harder time isolating as a means of protecting themselves from infection. Papageorge et al. (2020), for example, show that income is strongly associated with self-protective behavioural responses, with poorer individuals less able to practice social distancing and less able to telework. Stabile et al. (2020) find that low-income gig workers, such as food-delivery bikers, are more likely to keep working during the pandemic.
Yet, it is also the case that places with higher average incomes tend to have a higher customary density of personal interactions both in production and consumption, including stronger links to external sources of infection through travel (for work and leisure) and by attracting visitors. It may be harder for those living in relatively high-income areas to adapt rapidly to the pandemic.
The relevant socioeconomic covariates are likely to be correlated with each other and with classic epidemiological factors. So, looking at only one covariate on its own can yield misleading conclusions. For example, higher population density, which results in more interpersonal contacts, is widely seen as an important predictor of the spread of infection. Yet, population density is correlated with other variables, including the incidence of poverty and racial/ethnic composition.
For the US, several researchers have pointed to poverty and race as covariates of COVID-19 incidence (for example, Chen and Krieger 2020 and Chin et al. 2020). However, race and poverty are also correlated with each other in the US; so, is it race or poverty or both? Is population density relevant when one controls for correlated socioeconomic factors? And to what extent do adverse health factors reflect poverty?
Disentangling the socioeconomic influences on COVID-19 behaviour and outcomes for the US
In a new paper, we explore the empirical relationships among these factors across the 3,000 counties of the US (Brown and Ravallion 2020). We merge recorded counts of cases and deaths at the county level with socioeconomic characteristics – average incomes, race, income inequality, and poverty – and data on other covariates as suggested by the epidemiological literature. We use these data to model social-distancing behaviour, the infection rate, and the death rate conditional on infection.
There are theoretical ambiguities in how these socioeconomic covariates affect the spread of infections, given the induced behavioural responses and also allowing for costs of adjusting behaviour to the threat of infections, in addition to potential lags and nonlinearities. In the absence of effective pharmaceutical interventions, social-distancing choices are a plausible channel linking socioeconomic factors to the spread of infections. Our analysis of the bivariate relationship in the data is consistent with the view that social distancing lowers the COVID-19 infection rate in the US (Figure 1).
Figure 1 COVID-19 cases across US counties plotted against performance in social distancing
Notes: The figure shows a nonparametric regression function, giving the conditional mean at each point, based on a locally smoothed scatter plot. Each point on the x-axis corresponds to Unacast’s social-distancing grade, ranging from F (x = 1) to A (x = 12). The infection rates are from the US Centers for Disease Control and Prevention, up to 18 June 2020.
However, the marginal cost of greater social distancing is likely to be higher for poorer families, whose work and living circumstances cannot be easily changed; this is what we dub the ‘protection effect’. Against this, the pre-epidemic levels of social and economic interactions are likely to be higher for wealthier people, and there are costs in adjusting quickly to a lower level of interactions during the epidemic – the ‘adjustment-cost effect’.
Similarly, there are a priori reasons why a more elderly population can yield lower infection rates (they are likely to have fewer interactions, particularly for those who are retired), but higher death rates conditional on infections. Our results support the view that both these opposing and (as it turns out) roughly equal channels of effect on an elderly population are present in the US county-level data.
We see signs of both the protection and adjustment-cost effects in the relationship between COVID-19 outcomes across US counties and incomes. Without controlling for the incidence of poverty, a higher median income tends to be associated with greater improvements in social distancing and lower infection rates. However, this is because counties with a higher median income tend to have a lower poverty rate: after controlling for the poverty rate, a higher median income tends to come with higher infection rates and death rates. A higher poverty rate does the same, reflecting how a less pro-poor distribution at a given median positively affects infection rates.
We interpret the median-income effect on social distancing and infection rates as indicative of the adjustment-cost effect, while the positive impact of a higher poverty rate is interpreted as reflecting the protection effect. We find that the protection effect outweighs the adjustment-cost effect. Independent data on the social-distancing response to the epidemic also support this view: richer counties, and also more unequal counties, see stronger social-distancing responses. Counties with higher overall income inequality tend to have higher infection rates, which is in part because higher inequality comes with higher poverty rates. Similarly, higher inequality between counties increases the overall (national) infection rate.
We also find a strong effect of race, separately from poverty and inequality: a higher population share of Black Americans and Hispanics is associated with higher infection rates at the county level. There is also a strong interaction effect of race and poverty on the infection rate. Here, we see the combined effect of the higher incidence of poverty among these racial/ethnic groups and the fact that there is a concentration of these groups among the designated ‘essential workers’ in healthcare, food preparation, and other services, who are more exposed to the virus through their work. Without controlling for the racial composition of counties, one substantially overestimates the viral impact of higher-population density.
In modelling death rates across US counties, we find that, as expected, deaths rise with the number of cases. Additionally, the ratio of fatalities to cases tends to rise with the number of cases. This is a strong (and statistically significant) pattern in our results. We use the famous ‘epi-curve’ for identification (given the endogeneity of cases), with controls for local healthcare capacity. We find that socioeconomic covariates tend to matter more to deaths via infection rates, rather than independently of the latter. Age, however, is a consistent significant covariate of deaths, with a higher share of the population over 65 yielding a higher number of deaths after controlling for cases.
Our interpretation is that poorer people are less able to protect themselves, which leads them to different choices – they face a steeper trade-off between their health and their economic welfare in the context of the threats posed by COVID-19. This points to a role for anti-poverty policies, such as cash transfers, to complement health policies in combating this infectious disease.
Alongside the poverty effect, our results are also consistent with the view that richer people tend to interact more (in both their income-earning and consumption choices). They reduce these interactions relative to pre-epidemic levels, but the costs of adjustment still leave richer counties with higher infection rates once one controls for the poverty rate and/or the share of Black Americans.
These socioeconomic effects on the spread of the virus do not fade over time from the first infection; rather, the effects tend to become even stronger. There is little to suggest that the mixture of different socioeconomic groups dulls the effects of the underlying inequalities. What we see in the data is more consistent with a model of socioeconomic segmentation as the virus spreads, probably reflecting a learning process in combination with the different economic constraints on social distancing.
An exception to this pattern is found in how infection rates vary with an elderly population, which tends to matter less over time, probably reflecting younger families adopting greater social distancing (which comes more naturally for many of the elderly).
Controlling for these socioeconomic characteristics, we still find signs of the effects documented in epidemiological and medical literature, though those effects become weaker. Population density remains a significant predictor of infection rates, but the coefficient is greatly attenuated once we control for socioeconomic characteristics, especially racial composition. The partial correlations with the incidence of pre-existing health conditions are generally weak when one controls for socioeconomic covariates.
These findings point to the importance of incorporating socioeconomic characteristics into standard epidemiological models that predict the spread of an infection such as COVID-19. Policy also plays a role in mitigating the adverse effects of the virus for lower-income communities.
Brown, Caitlin, and Martin Ravallion (2020), “Inequality and the coronavirus: Socioeconomic covariates of behavioral responses and viral outcomes across US counties’, NBER Working Paper 27549.
Chen, Jarvis T, and Nancy Krieger (2020[EN1] ), “Revealing the unequal burden of COVID-19 by income, race/ethnicity, and household crowding: US county vs ZIP code analyses”, Harvard Center for Population and Development Studies Working Paper Volume 19, Number 1.
Chin, Taylor, Rebecca Kahn, Ruoran Li, Jarvis T Chen, Nancy Krieger, Caroline O Buckee, Satchit Balsari and Mathew V Kiang (2020), “US county-level characteristics to inform equitable COVID-19 response”, medRxiv preprint.
Furceri, Davide, Prakash Loungani, Jonathan Ostry and Pietro Pizzuto (2020), “COVID-19 will raise inequality if past pandemics are a guide”, VoxEU.org, 8 May.
Kikuchi, Shinnosuke, Sagiri Kitao and Minamo Mikoshiba (2020), “Heterogeneous employment vulnerability and inequality in Japan”, VoxEU.org, 3 May.
Lakner, Christoph, Daniel Gerszon Mahler, Mario Negre and Espen Beer Prydz (2020), “How much does reducing inequality matter for global poverty?”, Global Poverty Technical Note 13, World Bank.
Palomino, Juan, Juan Gabriel Rodríguez and Raquel Sebastian (2020), “Inequality and poverty effects of the lockdown in Europe”, VoxEU.org, 16 June.
Papageorge, Nicholas, Matthew Zahn, Michèle Belot, Eline van den Broek-Altenburg, Syngjoo Choi, Julien Jamison and Egon Tripodi (2020), “Socio-demographic factors associated with self-protecting behavior during the Covid-19 pandemic”, NBER Working Paper 27378.
Stabile, Mark, Bénédicte Apouey and Isabelle Solal (2020), “COVID-19, inequality, and gig workers”, VoxEU.org, 1 April.