What Can Air Pollution Tell Us About the Chinese Government's Foreign Affairs?

Note: This article has been adapted from my Master’s thesis, “Blue Skies for Allies? Quantifying China’s Foreign Relations Using Air Pollution Readings” (source code available here)

There is ample anecdotal evidence that Chinese authorities are able to exert control over urban air pollution, at least to some extent. For example, prior to the 2008 Beijing Olympics, it was widely understood that the Chinese government regulated factory output and the number of cars on the road in order to reduce air pollution in Beijing prior to the Opening Ceremony. The goal of these measures was to improve the image of China in the eyes of the international community, especially as the Olympics was going to draw special attention to Beijing. This raises the question of whether Chinese authorities have sought to control the level of air pollution during other politically important events. Official visits to China from representatives of foreign governments could fall into this category.

My Master’s thesis probes whether analyzing the level of air pollution in China during state visits can inform our understanding of how China views its bilateral relationships with different nations. I explore the relationship between state visits occurring in Beijing and the Air Quality Index (AQI) level in an effort to develop a novel approach to quantify China’s foreign relations. In order to investigate this question, I construct an original dataset on state visits from foreign officials from 155 different countries over the period of 2008 through 2018, which provides a unique look into the timing and composition of state visits to Beijing over this time span. I combine this with hourly data on air pollution particulate matter readings as well as weather data from this period. The dataset I constructed is itself a notable contribution to the research literature. To my knowledge, it represents the first detailed catalogue of information on state visits taking place in Beijing.

To take one example, from November 8 through November 10, 2017, President Xi Jinping hosted U.S. President Donald Trump for an official state visit between the two heads of state in Beijing. Winter months in Beijing are known to have elevated levels of air pollution, typically averaging upwards of an AQI score of 100. However, during President Trump’s three-day state visit, the air quality improved significantly. The chart below shows hourly air pollution particulate matter readings starting three days prior to Trump’s arrival in Beijing and ending three days after his departure:

alt text

In the 24 hours before Trump landed in Beijing, hourly AQI readings dropped by around 200 points, going from a level of just over 200 to nearly zero. The disappearance of grey, smoggy skies made for a good photo opportunity; photographs of President Trump and President Xi in the Forbidden City on the first day of Trump’s visit show the two leaders under blue skies. The fact that Trump’s visit was greeted with unpolluted air—particularly since other elements of Trump’s visit, such as the lavish state dinner held in the Forbidden City, were important symbolic gestures—may suggest that the Chinese government strategically intervenes in order to manipulate the level of air pollution during important political events like visits from heads of state from other countries.

Was the decrease in air pollution prior to Trump’s state visit a coincidence, or does it portend a broader trend? According to my analysis, when representatives of the U.S. are in town, pollution tends to go down: U.S. state visits are associated with a 12-point drop in AQI on average. This finding indicates that days in which American officials travel to Beijing to meet with their Chinese counterparts are routinely met with blue skies. However, other countries such as Japan experience higher levels of air pollution during their state visits. By exploiting the timing of state visits of foreign officials in Beijing and the estimated impact of a state visit on air pollution, I introduce a new approach to understanding China’s relationships with other countries. My findings are further strengthened by making use of a natural experiment, as one approach to addressing issues of endogeneity. Ultimately, this paper seeks to contribute to the literature on the political dimension of air pollution and to suggest avenues for further research into using air pollution as a novel way to make sense of the behavior of the Chinese government.

The main research questions I explore are:

  1. What impact does a state visit have on air pollution on a given day in Beijing?
  2. Which countries are associated with an increase in air pollution during state visits in Beijing?
  3. Which countries are associated with a decrease in air pollution during state visits in Beijing?

I hypothesize that state visits overall are associated with a decrease in air pollution. This result would bolster the theory that reducing air pollution is a mechanism the Chinese government makes use of to improve China’s image in the eyes of foreign leaders and foreign media, who are likely to cover a foreign official’s state visit to China.

In order to test my hypothesis that air pollution in Beijing drops around visits by foreign officials, I collected data on daily air pollution in Beijing and match this information with records of state visits to Beijing from representatives of other countries, merging the data by date. In order to construct this dataset, I collected data from three main sources. First, I gathered data on hourly air pollution particulate matter readings from the U.S. State Department. Second, for data on state visits, I constructed a dataset of state visits by compiling information on meetings taking place in Beijing from China Vitae, an initiative of the Carnegie Endowment for International Peace. China Vitae is an online repository of information, including new stories and biographical information, used to track the appearances and travel of leading Chinese officials. I wrote a web-scraping algorithm in Python to collect data on all political meetings occurring in Beijing since 2003. From this set of records, I am able to determine whether a foreign official from another country was in Beijing on a given day. The third main data source used in my analysis is hourly weather data, collected from an atmospheric monitoring station at Beijing Capital International Airport. Weather data is understood to have an important effect on the level of air pollution in China, and previous research has modeled the level of air pollution in Beijing using various weather variables. The weather data I use in this paper come from the National Oceanic and Atmospheric Administration (NOAA), a U.S. scientific government agency that collects weather data worldwide. The agency makes data available for download via File Transfer Protocol (FTP), accessible by weather station code.

Using the China Vitae data, the criteria I used to define an event with a foreign official were: 1) if an event took place with at least two participants, and 2) if at least one foreign official was present at the meeting. I also restrict events to those that occurred at the highest levels of the Chinese government: meetings with foreign officials and either the Chinese president or premier. My logic for doing so is as follows: if the Chinese government strategically intervenes to control air pollution in Beijing, I would guess that meetings occurring with the highest level of Chinese leadership are the most likely category of events that would be susceptible to manipulation. After merging the three datasets together and cleaning the data, the final sample has 4018 observations, where each record is a unique day occurring from January 1, 2008 and December 31, 2008. The dependent variable in my econometric specifications is AQI score, a measure of how polluted the air is at a given point in time. The main explanatory variable used in my analysis is a state visit dummy variable that takes the value 1 when at least one foreign official is present in Beijing for a state visit on a given day, and otherwise takes the value of 0. I also include regressions that are specific to particular countries. Overall, a state visit occurs on about 17 percent of the days between 2008 and 2018.

Estimating the relationship between air pollution and state visits using Ordinary Least Squares (OLS), I find that when a generic state visit—that is, treating state visits from all countries as if they were the same—occurs on a given day, the level of AQI is expected to increase modestly (somewhere in the range of zero to 5 points), but this estimate is not statistically significant. Weather conditions like wind speed and direction also have an important impact on air quality as well, as higher wind speeds correspond with lower pollution levels. Based on these findings, there is weak evidence that a state visit correlates with a large and statistically significant decrease in air pollution, which rejects my main hypothesis.

Yet this initial analysis treats countries as interchangeable, leaving unexplored how results might differ from country to country. I proceed to explore how state visits from specific countries relate to changes in air pollution, and find that some country visits correspond with reductions in air pollution while others correspond with increases in air pollution. For example, U.S. state visits coincide with a 12-point drop in AQI (8% reduction relative to the mean), while visits from Japanese government officials correlate with a 23-point increase in AQI (15% increase over the mean). These findings are robust to the inclusion of weather variables as well as controls for seasonality. However, these estimates are only slightly statistically significant (at the 10 percent level), which reflects the noise in the data.

In order to to visually capture how estimated changes in air quality during days with state visits vary across countries, the below figure displays how AQI is estimated to change when representatives from the top 21 most frequent visiting countries send representatives to Beijing. Point estimates and 90 percent confidence intervals are presented for these countries with at least 10 state visits between 2008 and 2018, along with estimates for G20 countries as well as for all countries grouped together. Each black point represents the estimated coefficient of the state visit variable for that country employing the model specification that includes all weather controls, weather interaction terms as well as seasonality dummies for year and month. Throughout my econometric analyses, this specification has been the model of choice, as it incorporates the full set of control variables and most often results in a higher R2. The red bars on the chart represent 90 percent confidence intervals; bars that do not cross the zero line indicate that a point estimate is statistically significant at the 10 percent level.

Estimates for 11 countries suggest that pollution drops when officials from these countries meet with the Chinese president or premier. Of these countries, Myanmar has the largest estimated drop, yet point estimates for all but the United States are statistically insignificant. Ten countries experience estimated increases in air pollution on days with state visits, yet these estimates are statistically imprecise as well, as indicated by the large error bars. Only the coefficients for Vietnam, Japan, North Korea and Germany—which all have estimated AQI increases greater than 20 points—are statistically significant at the 10 percent level. The magnitudes of the estimate for “g20” and “all countries” remain small, as they are quite close to zero. Since these variables pool observations across multiple countries, their larger sample size reduces their variance, as the tighter confidence intervals illustrate. However, since the confidence intervals cross zero, this indicates that these estimates are not statistically significant.

alt text

There are important limitations to this study, including the possibility that there are other variables not included in the analysis (such as the level of industrial output or country-specific characteristics), which would bias the regression estimates. Moreover, these initial empirical results presume a correlational relationship between state visits and air pollution, rather than a causal one, so there is certainly an issue of endogeneity. In order to strengthen the causal relationship I am putting forth—that the presence of a state visit causes a change in air pollution level—I exploit natural variation in weather patterns in Beijing, examining whether changes in pollution during state visits went in the opposite direction than what was expected based on that day’s weather conditions. I do so by conditioning my regressions on windy days and examining how the estimates change. My empirical results showed that higher wind speeds correspond with lower pollution levels. In addition, previous research suggests that days when the prevailing winds in Beijing originate from the South are associated with lower levels of air pollution. As a result, I test the robustness of my results by employing these alternate subsets of the data: 1) estimates for days with wind speeds above 3 meters per second, 4 meters per second, and 5 meters per second; and 2) estimates for days when the wind was blowing from the South. I find that some results are strengthened using this approach. For example, in the case of Japan, on days when wind conditions would ordinarily suggest a reduction in air pollution, a Japan state visit is accompanied by a large increase in air pollution. Since Japan state visits correspond with a statistically significant increase in pollution overall, this result lends additional support suggesting that I’ve accurately captured the causal mechanism linking state visits and air pollution.