2.1 Data

The data used in our analysis contain information about MSW discarded every day. We obtained the data from a waste collection company,Footnote ⁴ which serves 252 communities in Changping, a district in northwest Beijing. The data contain detailed discarding records of these communities from 2019 to 2020, including the weight, time and place of discarding. Since the accumulation of domestic waste in communities would cause great problems, the garbage collection company was still in normal operation during the period of the lockdown. This ensured that we can still obtain data on waste discarded during the lockdown. Owing to service limitations, the company only provides garbage collection services other than organic waste. In other words, our sample only includes refuse and recyclable solid wastes.

We show the spatial dispersion of our sample communities in figure 1 using the geographical information of the communities. Changping District has a resident population of 2.166 million. Owing to the large areas of mountains and parks in Changping District, our sample communities are relatively concentrated as shown in online appendix figure A1.

Figure 1.

Map of Changping District, Beijing, China.

Notes: Panel (a) shows the relative location of Changping District in Beijing. Panel (b) shows the geographical distribution of sample communities.

We subsequently merge the data on municipal waste with the COVID-19 lockdown policy data of China. We refer to He et al. (Reference He, Pan and Tanaka2020) and Dang and Trinh (Reference Dang and Trinh2021) and collect the lockdown information from Oxford University's COVID-19 Government Response Tracker (OxCGRT). On January 23, 2020 China began to take strict measures to control the COVID-19 epidemic, including store closing, suspension of work and schools, stay-at-home requirements, and traffic control to restrict public gatherings and mobility (Hale et al., Reference Hale, Angrist, Cameron-Blake, Hallas, Kira, Majumdar, Petherick, Phillips, Tatlow and Webster2020). Unlike many other countries, China's lockdown was implemented simultaneously. However, food materials purchases and municipal waste disposal services necessary for life were not restricted.

According to the government announcements and news collected by OxCGRT, Beijing implemented a lockdown policy similar to the one in Wuhan and canceled the annual grand event of celebrating the Spring Festival, closed major scenic spots, and advocated that citizens reduce mobility on January 23. In the following two days, Beijing suspended the long-distance bus service. Furthermore, China Railway canceled hundreds of trains in the following week, and Beijing required people traveling from Hubei Province to be isolated. Overall, while the measures in Beijing were not as strict as those in Wuhan, it also implemented similar measures to restrict mobility.Footnote ⁵ Figure A2 (online appendix) plots the newly confirmed cases and the total number of confirmed cases in Beijing from January to May, which shows that there were a large number of cases in Beijing in January.

We add to this data information about local weather and the characteristic of each community. The weather data were collected from the China Meteorological Data websiteFootnote ⁶ and contain daily temperature and wind speed.Footnote ⁷ Temperature is measured in centigrade, and wind speed is measured in meters per second. We manually collected the community characteristics from a real estate transaction website called Lianjia.com (at https://bj.lianjia.com/ershoufang/) to analyze the community heterogeneity. Data were collected on average house price, construction year, the total number of households, and property service information.

2.2 Empirical strategy

Most empirical work estimating the impact of the COVID-19 lockdown is built on Regression Discontinuity Design (RDD), such as that of Dang and Trinh (Reference Dang and Trinh2021) and Takaku and Yokoyama (Reference Takaku and Yokoyama2021). However, China's 2020 Lunar New Year was on January 25, which is very close to the imposition of the COVID-19 lockdown (January 23, 2020) and may thus become a confounding factor affecting MSW generation. Before the Lunar New Year, Chinese families have the custom of cleaning. Even if China had not implemented the lockdown policy, more solid waste before the Lunar New Year would have been anticipated than after. The RDD method directly compares the weight of waste discarded in each community before and after the lockdown, and the results will be biased by the New Year's cleaning customs.

To overcome the bias induced by the Chinese Lunar New Year., we refer to He et al. (Reference He, Pan and Tanaka2020), Fu and Gu (Reference Fu and Gu2017), and Abrams (Reference Abrams2021) and conduct a DiD regression where the comparison group for each community is itself in 2019:

(1)\begin{equation}MS{W_{cdy}} = \alpha + {\beta _1}\textrm{trea}{\textrm{t}_y} \times \textrm{Afte}{\textrm{r}_{cd}} + {\beta _2}\textrm{Afte}{\textrm{r}_{cd}} + {\beta _3}\textrm{trea}{\textrm{t}_y} + \beta {X_{dy}} + {\pi _d} + {\mu _c} + {\varepsilon _{cdy}},\end{equation}

where $MS{W_{cdy}}$ is the weight of MSW thrown away in c community on day d in year y. $\textrm{trea}{\textrm{t}_y}$ equals 1 if the year is 2020; otherwise equals 0. $\textrm{Afte}{\textrm{r}_{cd}}$ is 1 on the solar calendar day when the lockdown policy is implemented regardless of year. ${\beta _1}$ is the parameter of interest. ${X_{dy}}$ is a vector of observed time-varying control variables, such as temperature and wind speed. Since the severe cold and bad weather may affect people's decision regarding waste discarding at the same time, we added weather conditions to the control variables.Footnote ⁸ ${\pi _d}$ and ${\mu _c}$ respectively denote the weekday and community fixed effects. We estimate equation (1) with a window of one month before and after the lockdown.Footnote ⁹

Considering the bias induced by the Chinese New Year cleanup custom, we conduct a robustness check by using the corresponding lunar calendar of 2019 as the control group. Specifically, the lockdown day (January 23, 2020) happened two days before the Chinese New Year and corresponds to January 23, 2019, in the solar month. However, in the lunar calendar, the lockdown day corresponds to February 3, 2019, which is also two days before the Lunar New Year in 2019. $\textrm{Afte}{\textrm{r}_{cd}}$ is 1 after February 3 in 2019 and after January 23 in 2020 for regression based on the lunar calendar.

We also conducted an event study of the treatment and control groups to ensure the parallel trend in 2019 and 2020 before lockdown day. The basic event study model is given by

(2)\begin{equation}MS{W_{cdy}} \!=\! \alpha \!+\! \mathrm{\Sigma }_{j ={-} 9}^{10}({\beta _{1j}}\textrm{trea}{\textrm{t}_y} \!\times\! \textrm{da}{\textrm{y}_j}) \!+\! {\beta _2}\textrm{Afte}{\textrm{r}_{cd}} \!+\! {\beta _3}\textrm{trea}{\textrm{t}_y} + \beta {X_{dy}} + {\pi _d} \!+\! {\mu _c} + {\varepsilon _{cdy}},\end{equation}

where $MS{W_{cdy}}$ is the weight of MSW, and $\textrm{trea}{\textrm{t}_y}$ is 1 if the year is 2020. $\textrm{da}{\textrm{y}_j}$ is a dummy for the day on which time $j\;$ is before or after the calendar date when the lockdown policy is implemented regardless of year. Similar to equation (1), the event study analysis also includes the weather controls and year, weekday and community fixed effects.

Table 1 reports the summary statistics of all the variables. The average number of households in the sample community is 1,186. The average daily waste weight of each community is 156.3 kg and 125.6 kg during January and February of 2019 and 2020, respectively. One possible reason is that many people in Beijing return to their hometown during the Spring Festival and the waste weight counted in our sample does not include organic waste, the average daily waste weight of each community is relatively small.

Table 1.

Summary statistics

Notes: The t-test statistics in the last column test whether the means of a variable in the 2019 and 2020 periods are equal.

4.1 Effects of MSW hoarding

One possible explanation for the previous finding is hoarding behavior. The lockdown policy may not directly influence the generation of MSW; instead, people merely hoard garbage at home and reduce the frequency of waste discarding to reduce the risk of COVID-19 exposure.Footnote ¹¹ If this mechanism works, the total amount of MSW generated has not decreased, but the frequency of people throwing garbage has decreased. Our sample communities are all buildings and apartments, implying that people may have to throw away a large amount of garbage every time due to the difficulty of hoarding garbage for a long time in their homes.

To access the MSW hoarding mechanism, we estimate the effect of lockdown on the average weight of MSW per packageFootnote ¹² with equation (1). The weight of each package of waste refers to the weight of the garbage discarded at one time. If there is a hoarding effect, the average weight of each discard will increase. If the amount of waste generation reduces, the weight of waste discarded at one time will reduce. We expect the coefficient of the cross-term to be significantly positive, which suggests people hoarding garbage and a decrease in the frequency of discarding.

Table 3 reports the effect of lockdown on the average weight of MSW per package. The coefficient of the cross term is −2.89 and is statistically significant at the level of 1 per cent. Even when we add the weather controls and change the control group based on the lunar calendar, the results remain robust. We further estimate the effect of log MSW per package and find that the lockdown reduces the average weight of MSW per package by 55.2 per cent. These results are suggestive of the fact that the decrease in MSW after the lockdown occurred not because they hoarded garbage and reduced the frequency of discarding, but because less waste seems to have been generated.

Table 3.

Response of MSW per package to COVID-19 lockdown

Notes: This table shows the effects of the COVID-19 lockdown on the average weight per package of MSW. Standard errors are clustered at the community level and are reported in parentheses.

We also conduct a time-event analysis of the average MSW per package. Figure 3 plots the results specifically based on solar and lunar calendars, which confirm the finding with a DiD model. The average MSW per package is significantly lower after the lockdown, which confirms that the hoarding effect could not explain our previous findings. Instead, the significant decrease in the average MSW per package suggests that the generation of MSW decreased after the lockdown.

Figure 3.

Event study analysis: average MSW weight per package.

Notes: This figure reports the effects of the COVID-19 lockdown on the average weight of MSW per package and 95 per cent confidence intervals for the period from 9 days before to 10 days after the lockdown. Panel (a) reports the effect measured by the solar month model and panel (b) reports the results of the lunar month model. All models also include weather controls, community dummy, weekday dummy, and year dummy. The reference group is 10 days prior to the lockdown window. Standard errors are clustered at the community level.

4.2 Shifts of discard timing

Another alternative mechanism that would explain our finding is that the household shifted the time of discarding garbage. First, the severity of the virus may have affected the decision-making of discarding. Specifically, households delayed the time of discarding in the short run after obtaining the information on COVID-19 cases. The COVID-19 cases may also endogenously correlate with the lockdown policy and bias our estimation. Second, at the other extreme, households delayed the time of discarding until the lockdown policy was lifted in April. If the long-term shift effect is established, we will see that the amount of MSW after the lifting of the lockdown will exceed the level the day prior to it.

To test the short-term shifts in MSW discarding due to the information effects of COVID-19 cases, we refer to Brodeur et al. (Reference Brodeur, Cook and Wright2021) and include the lag variables of the new cases in Changping District, Beijing, and the entire country in equation (1). Table 3 reports the coefficients of the regression. Panel A reports the change in the total MSW, and panel B reports the effect of the average MSW. All the coefficients of the interaction are significantly negative. The coefficient based on the lunar calendar is −163.355 which is even lower than that in the baseline result (−134.157). This result makes sense, if the information on new cases prevents discarding, then we will see the gap in MSW before and after the lockdown widen when controlling the case information. Thus, our finding of the decrease in MSW is robust and is not driven by short-term shifts (table 4).

Table 4.

Robustness check: MSW response considering COVID cases

Notes: This table shows the robustness check of the effects of the COVID-19 lockdown on the weight of MSW. Standard errors are clustered at the community level and are reported in parentheses.

We next address the extreme situation that households shifted the time of discarding in a relatively long period and discarded all the garbage generated after the end of lockdown. We expect that the MSW will increase to a level higher than that before the lockdown in April. Therefore, we extend the post-lockdown window in the event study specification to 70 days after the lockdown, i.e., to April 2, when the lockdown policy was relaxed. The associated equation is

(3)\begin{equation}MS{W_{cdy}} = \alpha + \mathrm{\Sigma }_{j ={-} 9}^{70}({\beta _{1j}}\textrm{trea}{\textrm{t}_y} \!\times\! \textrm{da}{\textrm{y}_j}) \!+\! {\beta _2}\textrm{Afte}{\textrm{r}_{cd}} \!+\! {\beta _3}\textrm{trea}{\textrm{t}_y} \!+\! \beta {X_{dy}} \!+\! {\pi _d} \!+\! {\mu _c} \!+\! {\varepsilon _{cdy}}.\end{equation}

Figures 4 and 5 report the long-run event study analysis of the total MSW and average MSW per package, respectively. Both total MSW and average MSW decreased dramatically at the beginning of the lockdown day and then increased slightly after approximately 20 days. MSW remained relatively stable for nearly 50 days, and the MSW weight after the opening in April did not increase to the level before the lockdown. We refer to the model based on the lunar month, but the result stays robust. These findings refute the hypothesis that households shifted the MSW discarding in the long run. Instead, both the total MSW weight and average MSW weight remain low after 70 days of the lockdown, suggesting that the lockdown decreased the generation of MSW.

Figure 4.

Long-term event study analysis: total MSW weight.

Notes: This figure reports the effects of the COVID-19 lockdown on the total weight of MSW and 95 per cent confidence intervals for the period from 9 days before to 70 days after the lockdown. The long-term window ends on April 2, 2020. Panel (a) reports the effect measured by the solar month model and panel (b) reports the results of the lunar month model. All models also include weather controls, community dummy, weekday dummy, and year dummy. The reference group is 10 days prior to the window. Standard errors are clustered at the community level

Figure 5.

Long-term event study analysis: average MSW weight per package.

Notes: This figure reports the effects of the COVID-19 lockdown on the average weight of MSW per package and 95 per cent confidence intervals for the period from 9 days before to 70 days after the lockdown. The long-term window ends on April 2. Panel (a) reports the effect measured by the solar month model and panel (b) reports the results of the lunar month model. All models also include weather controls, community dummy, weekday dummy, and year dummy. The reference group is 10 days prior to the window. Standard errors are clustered at the community level.

4.3 Restriction of offline activities

To confirm that the decrease in MSW is the reduction of MSW generation and not the result of not being discarded, we further distinguish the effect of measures taken by the lockdown policy. In China, various measures were implemented, including restricting offline activities (e.g., dining, takeout, express delivery), closing workplaces and schools, and restricting public transportation. Although we could not distinguish the effect of each measure taken in January, the prevention measures to control a new round of the COVID-19 epidemic in Beijing offer an opportunity to test the effect of offline activity restriction.

After 56 consecutive days of zero new cases in Beijing, a new round of localized epidemics broke out at the Xinfadi market in Fengtai District, Beijing on June 12. The number of confirmed cases was 45 on that day, and several consecutive new cases were declared. The Fengtai District in Beijing rapidly took measures to control the epidemic on June 13. As for our sample community in Changping District, activities such as eating out and picking up express delivery were not restricted. However, mobility and public gathering activities were restricted, and schools closed again.

Utilizing this new round of control measures on June 13 as a placebo for the restriction of offline activities, we conduct the event study analysis with equation (2). Figure 6 reports the event study analysis of the placebo lockdown on June 13. We find insignificant decreases in both total MSW and average MSW per package after the new lockdown policy. The main difference between the June 13 lockdown and the first lockdown is that it did not strictly control offline activities, such as shopping, eating and express delivery. The insignificant coefficient in the placebo test indicates that offline activity restriction is the main driving force that reduces MSW. This finding implies that the lockdown changes consumer behavior and thus reduces the generation of MSW.

Figure 6.

Placebo event study analysis.

Notes: This figure reports the effects of the placebo epidemic response on the weight of MSW and 95 per cent confidence intervals for the period from 9 days before to 10 days after the COVID-19 cases broke out without lockdown policy in Changping. Panel (a) reports the change in total MSW weight and panel (b) reports the change in average MSW weight per package. All models also include weather controls, community dummy, weekday dummy, and year dummy. The reference group is 10 days prior to the event window. Standard errors are clustered at the community level.

A multi-country survey has documented an increase in MSW in some countries other than China because of increased consumption of packaged food, fresh food, and food delivery during the lockdown (Leal Filho et al., Reference Leal Filho, Voronova, Kloga, Paço, Minhas, Lange Salvia, Dias Ferreira and Sivapalan2021), which is the opposite of our result. However, our empirical results of the second round of lockdown point out that MSW does not decrease if offline consumption activities are not restricted. Unlike most countries, China's lockdown limits almost all consumption, especially food takeout and express delivery. Our findings are in line with the descriptive analysis of Fan et al. (Reference Fan, Jiang, Hemzal and Klemeš2021), which finds a decrease in household waste in Shanghai and an increase in both Brno and Singapore. We provide evidence that the restriction of consumption decreases the amount of MSW at the source.

4.5 Fear of going out

A possible explanation for the decrease in MSW is that people are fearful of going out due to newly confirmed cases. We collect the geographical location and date of the local COVID cases to conduct an event-study analysis, which examines the effect of locally confirmed cases on MSW discarding behavior.

(4)\begin{equation}MS{W_{cd}} = \alpha + \mathrm{\Sigma }_{j ={-} 9}^{10}({\beta _{1j}}\textrm{COVI}{\textrm{D}_c} \times \textrm{da}{\textrm{y}_j}) + {\beta _2}\textrm{Afte}{\textrm{r}_{cd}} + \beta {X_d} + {\pi _d} + {\mu _c} + {\varepsilon _{cd}},\end{equation}

where $MS{W_{cd}}$ is the weight of MSW for community c on day d, which takes logarithmic form. $\textrm{COVI}{\textrm{D}_c}$ is a dummy variable that takes the value of 1 if the community is within a certain range of the new case. We follow Liu and Tang (Reference Liu and Tang2021) and select the 3 km, 5 km, and 10 km communities near the newly confirmed cases as the treatment group because they have a high exposure risk of infection. The window of interest is 10 days before and after the date of the case report in 2020 and we omit the −10 day, which means that the reference group is 10 days prior to the occurrence of the new case. All the local cases in the Changping District are confirmed after the implementation of the lockdown policy.

Figure A3 in the online appendix reports the effect of confirmed cases. The insignificant change of the coefficients suggests that the newly confirmed local cases have no effect on the discard of MSW. We use various definitions of the high exposure communities and the results are robust. The average MSW per package shows a similar pattern (see figure A3b). This finding shows that the adjustment of discarding behavior because households avoid going out for fear of infection is not the main reason for the reduction of MSW.

4.4 Suspension of work and school

Finally, we simply address the effect of the suspension of work and school that moved people's daily activities from their workplaces and schools to their homes. The concern is that the shift in people's place of daily activity may also shift where they generate and throw away the garbage, which may also affect the changes in MSW after lockdown.

However, we find certain facts that make the effect of shifts in places of daily activities less problematic for our analysis. First, the winter holiday of primary and secondary school students began on January 18 in Beijing, which means that the impact of school suspension on their daily place of activity can be ignored. Second, if the suspension of work shifts the place where people generate and discard garbage, an upward bias of the MSW weight is anticipated after the lockdown. However, we find that the total MSW still significantly decreased after the lockdown. Moreover, the decrease in the average MSW indicates the reduction of MSW generation. Overall, even if we consider the effect of the shift in discarding place, the lockdown still significantly reduced MSW.