Human Behavior in the Presence of Wildfire Smoke: New Methods Reveal Differences in Group Responses

There is a growing awareness about the harmful effects of wildfire smoke and how those effects vary across socioeconomic groups. A new study by Berke et al. in Nature Human Behaviour suggests that differences in groups’ behavior, influenced by socioeconomic realities, may help explain why disadvantaged groups are disproportionately impacted by wildfire smoke.

Understanding behavioral responses to climate events, like wildfire smoke, can help governments produce better policy to protect residents. At present, public health guidance emphasizes individual self-protection as a means to mitigate the risks of smoke exposure. Governments encourage the “private provision of protection,” such as staying indoors, limiting air infiltration into the home, and buying protective equipment. This new research argues that individuals vary in their compliance with these suggestions, often due to factors beyond their control. As a result, these policies risk becoming both inadequate and inequitable.

To reach these conclusions, Burke et al. embraced a set of non-traditional, high-frequency data sources, including Google search queries, Twitter posts, and location data from smartphones. They analyzed individuals’ awareness and sentiment during wildfires, along with their movements, to better understand how behaviors differed across income brackets. Their findings are useful for state and local governments seeking to produce more effective and equitable climate policy.

Authors measured smoke exposure and income level using traditional data sources.

To understand smoke exposure, Burke et al. developed their own measure of daily exposure to ambient particulate matter (PM_2.5) from wildfire smoke. The authors collected data on particulate matter during periods in which satellites identified the presence of overhead smoke affecting a significant share of the U.S. population. Satellite images were used as a means to identify PM_2.5 measures that could credibly be attributed to wildfire smoke, rather than from other pollution sources.

The authors also used data collected from private pollution sensors, configured both indoors and outdoors, to understand how ambient outdoor smoke infiltrated homes. They collected hourly data from 1,520 indoor PurpleAir pollution monitors, then used data from outdoor PurpleAir monitors nearby to construct outdoor PM_2.5 concentrations for each home. These measures were primarily used to study whether infiltration of smoke into homes differed depending on various household and neighborhood characteristics. Finally, Burke et al. collected data on median household income from the U.S. Census Bureau’s American Community Survey.

Many behavioral responses to wildfire smoke exposure were income-dependent.

Berke et al. used Google search queries to understand whether increased smoke exposure also increased interest in purchasing individual protective equipment. They analyzed the popularity of specific search queries (ex. ‘air quality’) as a means to understand individuals’ awareness of smoke exposure. The authors interpreted an individual’s own purposeful Google search about smoke exposure as an indicator that they were aware of and interested in learning about the effects of their exposure.

On average, smoke exposure increased online search activity related to protective behavior. Searches for technologies that helped limit smoke exposure (ex. ‘air filter’) increased on days when smoke concentration levels were higher. It’s notable, however, that higher income households had greater search activity compared to lower income households when it came to protective equipment. The authors found that lower income households were not as interested in purchasing protection as higher income households, despite the two groups’ similar levels of awareness with regard to smoke exposure.

Burke et al. also analyzed Twitter posts published since 2016 to estimate sentiment and preferences regarding smoke exposure. Using natural language processing, they found that exposure to ambient smoke yielded more negative sentiment. However, there was a distinction between levels of unhappiness among different socioeconomic groups. Low-income households had more mild sentiments relative to higher income households, who tended to be more disgruntled as expressed through their Twitter activity.

Finally, Burke et al. used location data from smartphones to understand individuals’ movement during periods of smoke exposure. They studied the share of people estimated to be completely at home or completely away from the home. To ensure that they were measuring change in location as a direct impact of smoke exposure, rather than proximity to the wildfire itself, the authors developed a measure of distance to the nearest active wildfire and analyzed whether individual responses differed depending on fire proximity.

They found that greater smoke exposure was associated with a tendency to avoid outdoor air. On days with heavy smoke concentration, cell phone location activity showed that many people chose to either shelter in their own home or leave the area completely when exposure became severe. The authors did find a distinction between socioeconomic groups, where individuals residing in wealthier counties were more likely to remain fully at home during heavy smoke days relative to lower income households.

There was a weak correlation between income and infiltration of ambient smoke into the household.

Despite higher-earning individuals’ tendency towards greater protection and awareness, Burke et al. nevertheless found high levels of indoor smoke exposure in wealthy neighborhoods. They suggested that this finding likely resulted from the dominant policy guidance to open windows and close doors as a means of managing smoke infiltration, rather than any residents’ advocating for high quality building materials. Thus, it’s reasonable to assume wealthier households practicing the same infiltration management tactics as lower income households would not differ in their indoor smoke exposure levels.

Policy reliance on individual self-protection during wildfires was inadequate in mitigating risks from smoke exposure, particularly for disadvantaged groups.

Beyond studying how individuals generally respond to wildfire smoke exposure, Burke et al. showed how these behaviors differed by socioeconomic status. On average, lower income households were less likely to shelter at home on heavy smoke days or consider purchasing protective equipment than their higher income counterparts. This finding is consistent with lower income groups often facing constrained decision-making when it comes to deviating from regular routines and acquisition of resources (like protective equipment). Furthermore, lower income households were found not to have as extreme negative sentiments about smoke exposure, suggesting that their focus may be more directed toward fulfillment of basic needs.

Despite these findings, many current policy approaches to mitigating risk from smoke exposure encourage basic self-protection measures, such as sheltering at home and closing windows. This policy approach is inadequate, as it doesn’t consider key nuances in behavioral responses to smoke exposure. Lower income groups face more resistance to adopting protective measures, according to the authors’ findings about mobility. Any policy promoting such measures could be biased against disadvantaged groups.

Limitations to findings included representativeness of pollution monitors, ambiguity of search queries, and selection of level for spatial analysis.

Burke et al. noted that data collected from PurpleAir monitors were likely their least representative sources or datasets, as wealthier and more educated households were more likely to own PurpleAir monitors. They noted, however, that previous studies didn’t find support for socioeconomic and demographic information being strong predictors of infiltration levels. They therefore claimed that the representativeness of PurpleAir monitor data should not be cause for concern.

The authors didn’t go so far as to assume that an item-specific search is indicative of an individual’s intention to buy protective equipment, as they could not observe whether individuals actually made a purchase beyond the initial Google search. They hinted that another limitation is how Google searches are, at most, a proxy for true actions. In other words, we cannot know whether a search increases the likelihood of an inevitable purchase.

Another limitation of the study was that most analysis took place at the county level. As a result, any variation in estimated individual behavioral responses was largely ignored.

The authors’ creativity in data collection presents a blueprint for using new methods to analyze human behavior during climate events.

Policymakers should consider tailoring risk mitigation measures to better respond to the behaviors and perceptions of residents, particularly those in marginalized groups. The authors’ use of multiple non-traditional data sources, including cell phone location, social media activity, and online search queries, presented more nuanced behavioral insights that differed by income group. Harnessing such creativity in data collection may yield more insightful findings that can, in turn, encourage more effective and equitable climate policy.

---

Burke, M., Heft-Neal, S., Li, J. et al. Exposures and behavioural responses to wildfire smoke. Nat Hum Behav 6, 1351–1361 (2022). https://doi.org/10.1038/s41562-022-01396-6

Jia Coco Liu, Ander Wilson, Loretta J. Mickley, Keita Ebisu, Melissa P. Sulprizio, Yun Wang, Roger D. Peng, Xu Yue, Francesca Dominici, Michelle L. Bell, Who Among the Elderly Is Most Vulnerable to Exposure to and Health Risks of Fine Particulate Matter From Wildfire Smoke?, American Journal of Epidemiology, Volume 186, Issue 6, 15 September 2017, Pages 730–735, https://doi.org/10.1093/aje/kwx141

Stone, S., L. Anderko, M. Berger, C. Butler, AND W. Cascio, et al. Wildfire Smoke: A Guide for Public Health Officials, revised 2019. U.S. EPA Office of Research and Development, Washington, DC, EPA/452/R-19/901, 2019. https://cfpub.epa.gov/si/si_public_record_report.cfm?dirEntryId=347791&Lab=NHEERL