US Embassy air-quality tweets led to global health benefits

Significance

Over 4 million premature deaths per year are attributed to air pollution, most of which are in low- and middle-income countries where residents do not have access to reliable information on air quality. We evaluate a large-scale program that provided real-time air-quality updates at over 40 US diplomatic sites around the world with poor preexisting monitoring. We find that the embassy monitoring program led to substantial reductions in fine particulate concentration levels, resulting in substantial decreases in the premature mortality risk faced by the over 300 million people living in cities home to a US embassy monitor. Our research indicates that monitoring and information interventions that draw attention to poor air quality in developing countries can generate substantial benefits.

Abstract

The World Health Organization estimates that over 90% of the world’s population is exposed to hazardous levels of local air pollution. Air pollution is markedly worse in low- and middle-income countries, yet air-quality monitoring is typically sparse. In 2008, the US Embassy in Beijing began tweeting hourly air-quality information from a newly installed pollution monitor, dramatically improving the information on air quality available to Beijing residents. Since then, the United States has installed over 50 monitors around the world, tweeting real-time reports on air quality in those locations. Using spatially granular measurements of local air pollution from satellite data that span the globe, we employ variation in whether and when US embassies installed monitors to evaluate the impact of air-quality information on pollution. We estimate that embassy monitors led to reductions in fine particulate concentration levels in host countries of 2 to 4 µg/m³. Our central estimate of the annual monetized benefit of the decrease in premature mortality due to this reduction in pollution is $127 million for the median city in 2019. Our findings point to the substantial benefits of improving the availability and salience of air-quality information in low- and middle-income countries.

One of every nine deaths worldwide was attributable to air pollution in 2012 (1). Approximately 87% of these deaths were in low- and middle-income countries. Yet by 2011, less than 10% of developing countries monitored levels of fine particulate matter, the leading cause of premature mortality among air pollutants (2, 3).

We investigate the impacts of a large-scale third-party intervention that dramatically improved the availability of air-quality information in low- and middle-income countries: the installation of air-quality monitors along with the live tweeting of air-quality readings by US Embassies. Namely, in 2008, the US Embassy in Beijing began tweeting hourly air-quality readings from its newly installed monitor, providing real-time information on fine particulate matter to people in Beijing.* This drew international attention to poor air quality in China. By 2020, over 50 other US diplomatic facilities in 38 countries around the world had installed monitors and were live tweeting air-quality readings.^†

We evaluate the impacts of this natural experiment in monitoring using satellite-derived measurements of pollution at 466 cities in 136 countries, of which 50 cities in 36 countries received embassy monitors by 2020. Using panel variation on the existence and timing of air-quality tweets across cities, we estimate that air-quality monitoring by US Embassies significantly lowered fine particulate concentration levels in host cities. Using established relationships between pollution levels and premature mortality along with estimates of the value of a statistical life, we calculate the monetized benefit of monitor-induced reductions in pollution enjoyed by the over 300 million people living in cities home to a US Embassy monitor as of 2019. Evidence from this approach indicates that the benefits from the decreases in pollution due to the monitoring program were substantial, highlighting the importance of improving the availability and salience of air-quality information in low- and middle-income countries. Finally, we provide ancillary evidence that hardship pay, compensation received by US Embassy workers for pollution exposure and other conditions, declined after monitor installation; this means that the US Department of State actually saved (rather than lost) money from implementing this program.

This paper contributes to the small literature documenting the benefits from improvements in air-quality information. Namely, previous work has measured the health benefits of air-quality warnings (5–8). A related literature focuses on behavioral responses to improved information such as moving away from more-polluted areas (4) or individuals installing air purifiers in their home (9). We contribute to this literature by providing evidence that the provision of credible high-quality information on fine particulate concentration levels can directly induce reductions in pollution levels, generating substantial public health benefits.

The real-time information on air quality provided by embassy monitors may cause decreases in pollution in host cities for several reasons. First, monitoring networks in many developing countries are sparse. In addition, concentrations levels of fine particulate matter (particles with diameter up to 2.5 micrometers, known as PM${}_{2.5}$) are rarely measured by local monitors in developing countries, despite the fact that the bulk of the measured benefits from reductions in pollution stem from decreases in mortality risk from fine particulate exposure (10–12). The daily reporting of PM${}_{2.5}$ readings from embassy monitors may arm local and federal governments with the evidence necessary to implement pollution policy in host cities.

Readings from embassy monitors may also be considered more credible than locally sourced information on air quality. Namely, previous work has documented the manipulation of monitor readings (13–15), the strategic siting of monitors (16), and strategic responses to gaps in monitor coverage (17). An important feature of our setting is that embassy monitors are installed by a third party with no obvious motive to suppress information on poor air quality.

High monitored pollution levels may lead to public pressure from residents or even internationally, forcing policymakers to take actions to control pollution (18). On this point, we provide suggestive evidence that embassy monitor readings make pollution more salient to residents in host cities, showing that Google searches for “air quality” and related keywords increase after an embassy installs a monitor. However, testing for specific mechanisms behind monitor-induced reductions in pollution is challenging in our setting since we would require data on the preexisting availability, quality, credibility, and salience of air-quality information for many cities in many countries.

The Effect of Monitoring on Pollution

Embassy Monitors.

We collect historical air-quality reports from the US Government website AirNow (19, 20). This allows us to identify the first day that air-quality information was released from each US Embassy monitor.^‡

To understand trends in air-quality monitoring in low- and middle-income countries, we combine the embassy monitor installation data with information on global air-quality monitoring from the World Health Organization’s (WHO) outdoor air pollution database (3). Using these data, Fig. 1 shows the availability of PM${}_{2.5}$ monitoring in non-Organization for Economic Cooperation and Development (non-OECD) countries in our sample between 2001 and 2018. The black solid line shows the staggered rollout of monitors and accompanying live tweeting of air-quality reports by US Embassies, beginning with Beijing in 2008.^§ The black dashed line shows the share of countries recorded in the WHO database as having some form of local government monitoring.

Fig. 1.

State of air-quality monitoring in non-OECD countries. “Local PM2.5” corresponds to the share of countries with some form of local government-administered monitoring of PM${}_{2.5}$. “U.S. Embassy” corresponds to the share of countries with a monitor installed by a US Embassy. “Any” is the share of countries with a monitor administered by either the local government or a US Embassy. Data on the presence of local PM${}_{2.5}$ monitoring are from (3) (2011 to 2018).

This country-level annual measure of monitoring likely overstates the availability of air-quality information in low- and middle-income countries. Unlike the embassy monitoring, which is continuous and widely accessible, local monitoring need not be accessible to the public nor track pollution consistently across sites and throughout the year. Regardless, prior to 2008, there were virtually no PM${}_{2.5}$ monitors in non-OECD countries. By 2018, ~32% of non-OECD countries in our sample had some form of PM${}_{2.5}$ monitoring; 18% had only local monitoring, 7% had only embassy-administered monitoring, while 7% had both forms of monitoring.

Method.

Ref. 21 provides gridded estimates of fine particulate concentration levels across the world by combining satellite measurements of aerosol optical depth with chemical transport modeling (22). Using these data, we calculate average PM${}_{2.5}$ levels in circles of increasing radius from 1 to 200 km around the center of 466 cities in 136 non-OECD countries (23). For the main analysis, we consider the 50-km circle but the conclusions are robust to alternatives. In SI Appendix, Table S1, we confirm that satellite-derived estimates of PM${}_{2.5}$ are highly correlated with readings from embassy monitors. The results remain similar when using annual averages of gridded estimates of PM2.5 from the Modern-Era Retrospective analysis for Research and Applications (MERRA) dataset (SI Appendix, Fig. S6) (24).

We assess the impact of monitoring on pollution using variation in whether and when monitors were installed across cities from 2001 to 2019. For our main specification, we exclude Beijing, home to the first and most heavily publicized monitor installed in 2008. However, SI Appendix, Fig. S5 demonstrates that the results remain similar when including Beijing and using pollution data from 1998 to 2020.

We estimate the impact of monitor installation on pollution using the following event study framework:

$$Y_{i,t}={\alpha }_i+{\gamma }_t+{\beta }_i\times t+{\displaystyle \sum _{e=0}^T{\tau }_e}\mathbf{1}[t-M_i=e]+{ϵ}_{i,t},$$ [1]

where $Y_{i,t}$ is the PM${}_{2.5}$ concentration level at site i in year t. The specification includes site fixed effects, α_i, that capture time-invariant differences in pollution across locations (e.g., persistently high- versus low-pollution cities). We also include year fixed effects, γ_t, that capture global shocks to pollution (e.g., due to a global recession) and city-specific linear time trends, ${\beta }_i\times t$. The independent variables of interest, $\mathbf{1}[t-M_i=e]$, correspond to being e years after the year that the monitor was installed at site i (i.e., t is the year of sample and M_i is the year the monitor was installed) and τ_e are the parameters of interest.^¶ SEs are clustered by city.

To ensure that our estimates of the impact of monitoring on pollution levels in treated cities are robust to the presence of arbitrary heterogeneity in treatment effects, we utilize the methodology specified in ref. 25.^# Their imputation approach begins by estimating fixed effects α_i and γ_t as well as the coefficient β_i corresponding to the city-specific linear time trends using only untreated observations. The estimated parameters are then used to impute counterfactual pollution levels for treated sites in the absence of treatment. Consequently, the parallel trends assumption required for causality when using this methodology is that counterfactual pollution at treated embassies in the absence of monitoring is a linear combination of the fixed effects and control variables.

Results.

Fig. 2 plots the estimated effect of embassy air-quality monitoring on PM${}_{2.5}$ for each event year along with 90 and 95% CIs (shaded dark and light gray, respectively). The squares are coefficient estimates for the pretreatment period and the circles are coefficient estimates for the posttreatment period. Pretreatment and posttreatment parameters are estimated separately in the imputation approach.

Fig. 2.

Event study estimates of the effect of monitor installation on PM${}_{2.5}$ at embassies in non-OECD countries. Squares denote pretreatment coefficient estimates and circles denote posttreatment coefficient estimates; pre- and posttreatment coefficients are estimated separately. Dark gray shaded areas correspond to 90% pointwise CIs and light gray shaded areas correspond to 95% pointwise CIs. Both sets of CIs are based on SEs clustered by city.

The pretreatment coefficient estimates indicate that, conditional on city fixed effects, year fixed effects, and city-specific linear time trends, average pollution levels were trending similarly across treatment versus control cities prior to embassies in the treatment cities receiving a monitor. Indeed, we fail to reject the null hypothesis that sites with versus without monitors have common trends in pollution in the years prior to the treatment sites receiving the monitor. As further evidence in favor of the parallel trends assumption, after residualizing pollution by city fixed effects, city-specific time trends, and year fixed effects, residualized pollution levels are similar across treatment and control cities prior to 2008 when the first monitor was installed in Beijing (SI Appendix, Fig. S4, Bottom). This supports the assertion that our posttreatment estimates capture the effect of monitoring on pollution rather than simply differential trends in pollution across treatment and control embassies unrelated to the installation of the monitors.

The posttreatment estimates indicate that the installation of embassy monitors leads to statistically significant reductions in pollution levels. As might be expected, the effect of the monitor installation increases over time, suggesting that implementing pollution control is a gradual process. Fig. 3 demonstrates the robustness of our findings. Fig. 3A reproduces the main estimates cutting off at 4 y of treatment and Fig. 3B further restricts the set of treatment locations to those with at least 4 y of data after receiving a monitor. Fig. 3B thus presents event study estimates informed by a balanced panel of cities, noting that all pretreatment estimates are informed by all treatment cities since the earliest monitor was installed in 2008 and our sample begins in 2001. Both the pre- and posttreatment estimates informed by a balanced panel of cities are similar to our primary estimates in Fig. 2. Changes in the composition of treatment cities are thus unlikely to be driving the downward trend in the posttreatment estimated effects.

Fig. 3.

Event study estimates of the effect of monitor installation on PM ${}_{2.5}$ at embassies in non-OECD countries. Squares denote pretreatment coefficient estimates and circles denote posttreatment coefficient estimates; pre- and posttreatment coefficients are estimated separately. Dark gray shaded areas correspond to 90% pointwise CIs and light gray shaded areas correspond to 95% pointwise CIs. Both sets of CIs are based on SEs clustered by city. A reproduces the main estimates from Fig. 2 cutting off at 4 y of treatment. B computes posttreatment effects informed only by the set of treatment locations with at least 4 y of data after monitor installation. C and D present estimates from specifications that include country-by-year fixed effects rather than year fixed effects. The specification considered in C is estimated on the full sample of countries while in D we restrict the sample to only countries that received an embassy monitor between 2001 and 2020.

Fig. 3 C and D report event study estimates based on specifications that include country-by-year fixed effects rather than year fixed effects. Fig. 3C reports estimates using countries with and without embassy monitors, while Fig. 3D restricts the sample to only countries that received a monitor during our sample period. The estimated reductions in pollution remain economically and statistically significant although smaller in magnitude than the previous estimates.

In SI Appendix, Table S2, we present estimates of the average effect of monitoring on pollution for a variety of alternative specifications and sample restrictions, again using the imputation approach specified in ref. 25. The estimate in SI Appendix, Table S2, column 1 indicates that the average effect of embassy monitor installation in the main sample is –3.9 µg/${\text{m}}^3$ per year. In SI Appendix, Table S3 we further show that the estimates are similar in magnitude whether pollution is measured using a circle based on distances of 1, 50 (the base specification), or 200 km around the city center. This suggests that the reductions in pollution from installing a monitor are not highly localized. SI Appendix, Table S2, column 2 reports the estimate when we restrict the sample to include only cities in countries that receive an embassy monitor between 2001 and 2020. After adding country-by-year fixed effects to account for country-specific trends in pollution levels, we estimate that monitor installation led to a 1.89 µg/${\text{m}}^3$ decrease in pollution levels regardless of whether the sample includes countries that did not receive monitors between 2001 and 2020 (SI Appendix, Table S2, columns 3 and 4). In SI Appendix, Table S2, column 5, we document that the estimates remain similar when we exclude China and India, the first two countries to receive embassy monitors. Finally, SI Appendix, Table S2, column 6 reports that the monitoring program reduced pollution growth rates by 0.02 percentage points on average.

What could have caused such significant reductions in pollution? A stated aim of the monitoring program was to raise awareness of environmental issues locally (27). US diplomats and observers claimed that the program was successful in this goal, drawing attention to pollution in host countries and triggering a “profound change” in local air pollution policy (28).

As evidence of this, Fig. 4 shows that the real-time information on air quality provided by US Embassy monitors increased the salience of local air pollution. Fig. 4 plots event study estimates of the effect of monitor installation on Google search intensity for the keyword “air quality” collected from the Google Trends API (29). We see a gradual rise in search intensity after the installation of the monitor, with no differential trends in search intensity in the period prior to treatment across sites that receive versus do not or have not yet received monitors. SI Appendix, Fig. S7 documents that the event study estimates are similar for the keywords “pollution” and “AQI.”

Fig. 4.

Event study estimates of the effect of embassy monitor installation on Google search intensity for “air quality.” Squares denote pretreatment coefficient estimates and circles denote posttreatment coefficient estimates; pre- and posttreatment coefficients are estimated separately. Dark gray shaded areas correspond to 90% pointwise CIs and light gray shaded areas correspond to 95% pointwise CIs. Both sets of CIs are based on SEs clustered by city.

The Benefits of Reductions in Pollution

Measuring Benefits to the Local Population.

We next measure the benefits from the reductions in pollution due to the installation of embassy monitors. To measure the annual benefits enjoyed by residents of cities with monitors, we use an approach standard to regulatory impact analyses in the United States (30) and the European Union (31). The first step in this process is to calculate the number of premature deaths from pollution exposure avoided in the year t = 2019 due to the installation of an air-quality monitor at site i using the following equation:

$$M_{i,t}=PO{P}_{i}M{R}_{i,t}(1-(1/\mathit{exp}(\widehat{\rho }*\mathrm{\Delta }{\text{PM}}_{i,t}))),$$ [2]

where $\mathrm{\Delta }{\text{PM}}_{i,t}$ is the change in PM${}_{2.5}$ levels in 2019 implied by our event study estimates of monitor-induced decreases in pollution. Both the functional form relating pollution levels to mortality and the coefficient estimate $\widehat{\rho }=0.0135$ come from ref. 32. Annual country-specific all-cause all-age mortality rates $M{R}_{i,t}$ are collected from the WHO (33). Finally, for our central estimates, POP_i is the population in a circle of radius 10 km around monitor location i in the year 2015, calculated using gridded data on population derived from underlying NASA satellite measurements (34).

The results of this calculation are in SI Appendix, Table S4, which also documents results from the following sensitivity analyses. First, SI Appendix, Table S4A shows the reductions in premature mortality implied by smaller estimates of $\mathrm{\Delta }{\text{PM}}_{i,t}$, based on specifications that include country-by-year fixed effects and focus only on countries that receive embassy monitors during our sample period. In addition, SI Appendix, Table S4 presents results based on calculating population POP_i in circles of radii 1, 10, and 20 km around the embassy site as well as considering the whole population of the city exposed to monitor-induced decreases in pollution. Considering the entire city’s population exposed likely results in a reasonable upper bound on avoided premature mortality given that our estimates indicate that embassy monitor installations result in reductions in pollution even up to 50 km away from the embassy site. Finally, SI Appendix, Table S6 documents that the qualitative conclusions drawn remain the same when calculating population POP_i using city-level data on population density from (35) rather than satellite-derived measurements.

The next step in this process is to multiply the total number of deaths avoided $M_{i,t}$ by the value of statistical life (VSL) calculated using the formula derived in ref. 36:

$${\text{VSL}}_{i,t}={\text{VSL}}_{\text{USA},2016}\times {\left(\frac{{\text{GNI}}_{i,t}}{{\text{GNI}}_{\text{USA},t}}\right)}^{ϵ}\text{}.$$ [3]

In our implementation, we set ${\text{VSL}}_{\text{USA},2016}$ equal to $9.6 million (2016 US dollars). ${\text{GNI}}_{i,t}$ is the gross national income per capita, and ϵ is the income elasticity. Following ref. 36, our central estimates of the monetized benefits from pollution reductions are calculated using $ϵ=1$, but SI Appendix, Table S5 also provides estimates based on ϵ equal to 0.52, 1.1, and 2.

For each embassy monitor i, this methodology gives us the monetized benefit from reductions in premature mortality in 2019 due to monitor-induced decreases in PM${}_{2.5}$. Even under the conservative assumption that only people within 10 km of the embassy site are exposed, our methodology indicates that installing an embassy monitor results in a median across treated embassies of 303 fewer pollution-caused premature deaths and median monetized mortality benefits of $127 million (2016 US dollars [USD]). Under the reasonable upper-bound assumption that the entire city enjoys the benefits of monitor-induced decreases in pollution, the corresponding median reduction in premature deaths is 895 and the median mortality benefit is $465 million (2016 USD).

The distributions of avoided premature mortality and mortality benefits in 2019 from the monitoring program are right skewed. For example, the 5th percentile (95th percentile) of mortality benefits assuming that only people within 10 km of the embassy site are exposed is $10.4 million ($3.5 billion). This skewness is driven almost entirely by differences in population across cities. The results become smaller if we utilize smaller estimates of monitor-induced decreases in pollution, reduce the population exposed, or increase the income elasticity. However, regardless of the assumptions utilized in the calculation, the monetized mortality benefits from the embassy monitoring program are large, indicating that information or monitoring interventions can lead to substantial mortality benefits via reductions in pollution levels.

Measuring Financial Savings to the US Department of State.

To quantify the financial savings to the US Department of State from decreases in air pollution due to the embassy monitoring program, we first collect information on the wage premium paid to US diplomats to compensate them for living standards that differ from conditions in the United States. This wage premium, called the posthardship differential, is a proportional increase in salary determined by the US Department of State based on measures of social and physical isolation, violence, crime, healthcare, and environmental conditions (which include air pollution). The State Department publishes current and historical rates by diplomatic post and provides information on hardship categories (37). For our main analysis, we utilize the posthardship differential rates for 201 locations in 144 countries from 2001 to 2019.^‖

We use an event study framework to quantify the impact of monitoring on the posthardship differential.^** Fig. 5 plots the effects of monitoring on the wage differential along with 90% (dark gray) and 95% (light gray) CIs. The dependent variable (posthardship differential) is measured as a proportion; a value of 0.01 indicates that diplomats in that location receive a wage of 1.01 times base salary, where base salary is determined by their rank. In Fig. 5, the squares denote coefficient estimates for the pretreatment period and the circles denote coefficient estimates for the posttreatment period. The pretreatment estimates indicate that the posthardship differential was not trending differently for treatment versus control locations prior to the treatment locations receiving a monitor. This evidence supports the parallel trends assumption that the counterfactual wage premium at treated embassies in the absence of monitoring would have trended similarly to the predicted wage premium levels implied by the fixed-effects and control variables estimated using only untreated observations.

Fig. 5.

Event study estimates of the effect of embassy monitor installation on the posthardship differential (a percentage increase in wage to compensate diplomats for adverse conditions abroad) received by US diplomats stationed at locations in non-OECD countries. Squares denote pretreatment coefficient estimates and circles denote posttreatment coefficient estimates; pre- and posttreatment coefficients are estimated separately. Dark gray shaded areas correspond to 90% pointwise CIs and light gray shaded areas correspond to 95% pointwise CIs. Both sets of CIs are based on SEs clustered by city.

The posttreatment estimates in Fig. 5 indicate that, after US Embassies started to report air-quality information, the wage premium paid to US Embassy staff in those locations declined relative to the wage premium in control cities. This decline is consistent with the relative decrease in pollution presented in Fig. 2.

The posthardship differential is calculated using a formula that compensates US diplomats for the effects of pollution exposure. Consequently, the reduction in this compensation caused by the installation of the monitor reflects the monetized benefit of the associated pollution reduction enjoyed by the US Department of State. This interpretation implicitly assumes that the monitor impacts diplomat pay differentials only through changes in pollution. We believe this assumption to be reasonable since the posthardship differential is calculated based on a predetermined formula specified by administrative process, with pollution as one of many inputs, rather than negotiations between embassy workers and the State Department.

To calculate the annualized total reduction in compensation due to the monitoring program, we first estimate the average effect of embassy monitor installations on the posthardship differential. We then compute location-specific wage bills based on average staffing levels at US embassies, the distribution of rank of the foreign service, and diplomatic pay scales in 2019 (38). Salary schedules for 2019 were collected from (39). The distribution of staff across pay scales was collected from (40).

The estimate in SI Appendix, Table S8, column 2 indicates that the installation of an embassy air-quality monitor led to a 7.6% decrease in the posthardship differential. This 7.6% decrease corresponds to a 1.6 percentage point reduction in posthardship differential, as the average differential across treated cities in the year prior to monitor installation was 22%. Since the average US diplomat made $92,314 in 2019, our estimated effect represents a reduction in wages of approximately $1,477/y (i.e., 0.016 × $93,314). To put this in context, the installation of a monitor results in reductions in pollution levels of between 2 and 4 µg/${\text{m}}^3$ per year. Combining these estimates, US diplomats are compensated between $369 and $738 per µg/${\text{m}}^3$ of annual pollution exposure (i.e., between $1,477 $÷$ 2 and $1,477 $÷$ 4).

Across sites that received an embassy monitor between 2001 and 2020, the median number of diplomats employed in 2019 was 23, with a minimum of 7 and a maximum of 118.^†† Our estimates indicate that the monitors saved the median embassy $33,971/y in posthardship differential payments (i.e., 23 workers × $1,477 in saving per worker). Further, our estimates suggest that the annualized wage bill reduction from the monitor enjoyed by the US Department of State exceeds the annualized cost of the monitor for all treated embassies.^‡‡

These findings do not necessarily imply that the US Department of State should install monitors at all embassy locations. Although the United States does not directly bear the costs associated with reducing local air pollution in host cities, US policymakers may internalize some of the political pressure and economic costs of reducing air pollution resulting from the information on air quality. Providing credible information on poor air quality in a country may also come with diplomatic costs.

Conclusions and Policy Implications

The WHO notes in the most recent update of its air-quality guidelines that coverage of air-quality monitoring “is inadequate$\dots$ to accurately estimate exposure,” particularly in less developed countries (page 5 of ref. 42). However, expansion of regulatory grade monitoring is costly, and there is very little evidence on the benefits of improved information on air quality. This paper demonstrates that a large-scale global monitoring intervention by the US Department of State lowered pollution in low- and middle-income countries, likely through the channel of increased salience of air pollution among the local population. The intervention led to substantial health benefits enjoyed by the over 300 million people living in cities home to a US Embassy monitor as of 2019.

The costs of air-quality monitors have fallen precipitously in the last 10 y due to the advent of products such as Purple Air monitors. In addition, annual and even monthly measurements of pollution across the world from satellite-derived data are freely available. However, we caution that reporting precise information on air quality at an hourly or a daily level was likely important for the efficacy of the embassy monitoring program for several reasons.

First, temporally granular information on air quality is necessary for residents to incorporate pollution into their plans on the amount of time to spend outdoors on a given day. In addition, days with very poor air quality are especially salient, noting that satellite data are better suited to measure average pollution levels rather than extreme values (43). Accurately measuring the poorest air-quality days may better serve the purpose of motivating residents to place pressure on local policymakers to reduce pollution than average air-quality levels (18). For example, a New York Times article in 2013 reports “One Friday more than two years ago, an air-quality monitoring device atop the United States Embassy in Beijing recorded data so horrifying that someone in the embassy called the level of pollution ‘Crazy Bad’ in an infamous Twitter post” (44). Finally, most satellite-derived and ground-based measurements are noisier than the readings from the regulatory-grade monitors installed by the US Department of State. Implementing and enforcing local air-quality regulations on the basis of the former measurements may prove difficult.

Data, Materials, and Software Availability

All of the data used in this paper are from publicly available sources (3, 19, 20, 22–24, 29, 33–35, 37–40). All data and code used for analysis (satellite data, monitor installation dates, diplomat wages, and staffing levels) have been deposited in OpenICPSR (45).