Defending Against Air Pollution in China

Personal-level defensive measures against particulate matter (PM) pollution, including reducing outdoor time and wearing anti-pollution facemasks (APFs), are becoming increasingly important in both developing and developed nations. Nevertheless, there is a lack of empirical research on the overall efficiency of APFs and the size of the Peltzman effect in real-world settings. The uncertainty in the existing literature may have partly contributed to pushback against mask-wearing in the US and Australia when wildfires caused heavy PM pollution. Moreover, while information nudges have been widely applied as behavioral interventions to encourage desirable behaviors, there is a lack of understanding of the associated mechanisms and the optimal design. Acknowledging some of the gaps in existing literature, we conduct a large-scale field experiment through randomized control trials (RCTs) with 2,296 participants in Shenyang, the capital city of Liaoning Province in northeastern China. Intervention treatments include (i) financial incentives in the form of free masks; (ii) passive information nudge where each participant is provided with a brochure containing detailed information about PM pollution, its associated health damages, and required defensive measures; and (iii) three cascading information nudges where, on top of the brochure, the participants are asked to fill daily information about the air pollution level, checkmark the associated health effects, and/or the need to wear facemasks. Follow-up surveys are conducted to collect information from participants regarding their daily outdoor time and mask-wearing time, as well as their monthly health conditions throughout the entire winter heating season (from November 2018 through March 2019) when the pollution was the heaviest during the year.

2018-11-03

2019-03-31

defensive behaviors against air pollution

Defensive behaviors are measured by outdoor time (i.e. time spent outdoors) and mask intensity (i.e. time spent outdoors divided by time spent on wearing facemasks) each day throughout the winter heating season (from November 2018 to March 2019) during the study period.

Health outcomes

Health outcomes are measured by monthly doctor visits (taking a value of 0 or 1) due to respiratory or cardiovascular diseases throughout the whole winter heating season during the study period

Our experimental interventions involve one incentive treatment, namely a mask treatment, and three cascading information nudges, i.e. the passive information nudge, the active pollution information nudge and the active APF information nudge. The mask treatment provides three free KN95-type APFs while the three cascading information nudges are designed in the form of a combination of an information brochure and three different versions of calendars.

The “passive information nudge” provides each participant with one information brochure and version 1 of the calendar (V1 calendar). On the V1 calendar, participants are asked to fill out their daily information regarding how many hours they stay outdoors (outdoor time) and how many hours they wear facemasks when being outdoors (mask-wearing time) on each day during the whole winter heating season (from November 2018 through March 2019). The information brochure contains detailed information about smog and PM pollution, the health effects of the smog, defensive measures taken on smog days such as reduced outdoor activities and mask-wearing, roles of APFs in reducing the health effects of smog, different brands of APFs available on the market, the difference between APFs and regular cotton facemasks, and proper ways of wearing APFs. Each participant receiving an information brochure reads it through with verbal explanations from an enumerator during the baseline survey and then is asked to bring home one copy of the brochure after the survey. The “active pollution information nudge” provides one information brochure and a version 2 calendar (V2 calendar). The only difference between the passive information nudge and the active pollution information nudge is that the latter gives the V2 instead of the V1 calendar. The V2 calendar builds on the V1 calendar but requires the participants to carry out two additional tasks, i.e. checking one of four boxes corresponding to four intervals of daily mean AQI (i.e. 0-100, 101-200, 201-300, and 301 and higher) and one of four boxes corresponding to the health impact of the air quality (i.e. no impact, some impact, big impact, and very big impact). The “active APF information nudge” gives the participants one information brochure and a version 3 calendar (V3 calendar). The only difference between the active pollution information nudge and the active APF information nudges is the calendar version: the V3 calendar is like the V2 except that the V3 requires the participants to check an additional box among four boxes corresponding to the necessity of wearing an APF (i.e. no need, somewhat necessary, necessary, and very necessary).

The three different versions of calendars require different levels of effort made by the participants to fill out their daily information. Compared to the V1 calendar, the V2 and the V3 calendars need participants to make extra efforts to find out the daily AQI for a given day and then check a box indicating the health impact of the air quality on that day (both for the V2 and the V3) or an additional box showing the necessity of wearing facemasks on that day (only for the V3).

The additional tasks of filling out AQI information and checking boxes on the V2 or V3 calendars are relatively easy to be carried out. First, the participants can get easy access to daily AQI information, since nowadays in China, daily AQI information is routinely provided together with weather forecasts to the public through many media sources and public messages sent to cellphone users. Second, when checking boxes on the calendar, the participants can easily refer to a table presented at the lower corner of each page of the calendar. The table in the V2 has two columns: the first lists four intervals of daily mean AQI (i.e. 0-100, 101-200, 201-300, 301, and higher), and the second column lists the health impact corresponding to each interval of AQI. The table in V3 has three columns, including the same first two columns as those in V2, and a third column listing the necessity of wearing facemasks that corresponds to the information presented in the first and the second columns.

We use a 2 times 4 design to obtain eight experimental arms/interventions corresponding to different combinations of information nudges and the mask treatment. For the control group, only the V1 calendar is provided.

The randomization is conducted at the neighborhood level through the random assignment of eight experimental arms (interventions), which represent different combinations of a mask treatment and three cascading information nudges, to 40 neighborhoods. Half (20) of the neighborhoods (arms of 2, 4, 6, and 8) receive the mask treatment where each participant is given three free KN95-type APFs while another half (20) do not receive the APFs. Three-quarters (30) of the neighborhoods (arms of 3, 4, 5, 6, 7, and 8) receive three cascading information nudges: 10 neighborhoods (arms of 3 and 4) receive the passive information nudge, 10 neighborhoods (study arms 5 and 6) provided with the active pollution information nudge, and 10 neighborhoods (study arms 7 and 8) given the active APF information nudge. One quarter (10) of the neighborhoods (arms of 1 and 2) receive the V1 calendar but not the brochure; among them, five neighborhoods receive the APFs while the other five do not receive APFs. The control group is the neighborhoods only receiving the V1 calendar, but not the information brochure or free APFs.

The randomization is conducted at the neighborhood level.

Yes

40 neighborhoods.

Using clustered sampling, a total of 40 neighborhoods are randomly drawn from a population of 3,785 neighborhoods in nine urban districts in Shenyang. The number of neighborhoods assigned to each district is determined according to the proportion of its population in the total population of the urban areas of the city. To control for potential spillover effects between adjacent neighborhoods, we randomly draw the neighborhoods by imposing a condition that the distance between any two of them is greater than 2 km.

The planned number of observations: 2,400 participants. In each neighborhood, 60 adult participants (18 years and older) are randomly drawn from the neighborhood roster, subject to a screening condition that they would spend the entire winter heating season of November 2018 − March 2019 in Shenyang and had lived in Shenyang for at least one year by November 2018. The screening condition is set with the consideration that some households in Shenyang spend the winter season in warm climatic zones in southern China. The sample is stratified into two age groups: one half is between 18 and 64 years old and the other half is 65 years and older.

Sample size: 2,296 participants

With financial incentives provided for the participation, we recruit 2,296 participants out of the 2,400 participants who had been randomly selected and contacted through phones before the baseline survey and the RCTs and had agreed to participate in our study, achieving a participation rate of 95.63%.

We are mainly interested in estimating the main treatment effect of each treatment rather than the interaction effects of different treatments. For this purpose, we conduct two rounds of power analyses to determine the needed number of neighborhoods and participants, using data collected from a pilot study and the first period of our follow-up survey, respectively. In the pilot study carried out in January 2018, we randomly select 121 participants from six neighborhoods and conduct surveys on their past behaviors. After the survey, the participants are asked to participate in a randomized controlled experiment that involves an information treatment and a mask treatment. The information treatment provides information about PM pollution, smog, and its health damage as well as the role of APFs against pollution, while the mask treatment provides three KN95-type APFs at a 30% discount rate. We use a 2\times2 design and randomly assign four interventions at the neighborhood level; then we conduct a follow-up survey, which asks the participants to record their outdoor time and mask-wearing time spent on each day for two weeks. In the end, we collect the follow-up survey data from 74 participants, representing a response rate of 61%, and use these data to estimate intra-cluster correlations (ICC) of outdoor time and mask-wearing time, and the treatment effects on outdoor time and mask-wearing time. Based on the estimated ICC and effect size, we calculate the needed number of neighborhoods to achieve a statistical power of 80% to detect the treatment effect. Our calculation shows that the needed number of neighborhoods varies from 13 to 54 and the needed number of participants drawn from each neighborhood is around 40, depending on the outcome variable (outdoor time vs. mask-wearing time) and the type of treatments (the mask treatment and the information treatment). Considering the budget and time constraints, we decide to have 40 neighborhoods randomly selected and 60 participants randomly drawn from each neighborhood in the formal stage of the RCTs. Our pilot study also suggests two ways to modify our experiment design. One is to use more nuanced ways of information provided both in the form of an information brochure and calendars and the other is to give APFs free of charge instead of at a 30% discount rate. We incorporate the above modifications in our final design of treatments. We then conduct another round of the power analyses to make sure that after we make the modifications, the number of the neighborhoods (40) and the participants (60 from each neighborhood) that are initially determined using data from the pilot stage still has enough power to detect the treatment effect. For this round of power analyses, we use data collected from the first two weeks of follow-up surveys in November 2018 as our baseline surveys were rolled out from November 4th to 18th. In the second round of the power analyses, we also estimate the ICC of outdoor time and mask-wearing time, and the effect size of treatments including the information and mask treatments. The calculation shows that given 60 participants to be randomly selected in each neighborhood, the required number of neighborhoods to achieve a statistical power of 80% ranges from 8 to 80 depending on the types of treatments (for the treatment effects to be captured for different pairs of study arms, refer to Appendix Table 1), but the mean number of the needed neighborhoods is 25. We, therefore, continue our formal RCTs with 40 neighborhoods randomly drawn from all neighborhoods in Shenyang and 60 participants randomly chosen from each neighborhood.