Air Pollution

A feature of air pollution as an environmental health risk is that people do not passively “receive” exposure. When pollution rises or when people learn that it has risen, households and firms often respond by changing behaviour and making private investments that reduce exposure. These defensive actions include: reducing time outdoors; shifting the timing and mode of travel; buying masks and air purifiers; changing ventilation and building practices; seeking medical care or medication; and (in the medium run) relocating to cleaner places. A growing empirical literature now documents these behaviours, particularly in settings where air quality information is salient and where defensive technologies are widely available, but there are still major gaps for many LMIC contexts.

Defensive behaviour matters for several reasons. First, it is part of the welfare cost of pollution: even if defensive actions successfully reduce health harms, they use real resources (money, time, foregone enjoyment) and are themselves costly (Deschênes et al. 2017). Second, defensive behaviour complicates empirical measurement of pollution damages: if people reduce exposure when pollution is high, reduced-form estimates that relate ambient concentrations to outcomes may understate the underlying biological effect of exposure. Third, defensive investments are often highly unequal, with richer households and formal firms better able to protect themselves, raising distributional concerns and potentially weakening political pressure for public regulation. Finally, exploring margins to increase low-cost defensive behaviour may be a highly cost-effective strategy to reduce damages from air pollution.

Avoidance responses to air quality alerts

Air quality alert systems have been used extensively to study defensive behaviour because they offer clean quasi-experimental variation: when pollution crosses a regulatory threshold, an alert is issued, and researchers can compare behaviour just above and just below the cutoff. The foundational paper is Neidell (2009), who uses a regression discontinuity design around smog alert thresholds in Southern California and finds that alerts reduce outdoor facility attendance by 3–11%. Crucially, health effect estimates that fail to account for this avoidance behaviour are substantially attenuated: the true biological effect of ozone is larger than naive regressions suggest, because alerts trigger the very self-protection that dampens observed health outcomes. This is a key methodological point for the literature reviewed in Section 2.1: standard reduced-form damage estimates are lower bounds wherever avoidance is endogenous to information provision. Graff Zivin and Neidell (2009) extend this by documenting intertemporal dynamics: alert responses attenuate sharply when alerts are issued on consecutive days, consistent with rising costs of repeatedly postponing outdoor activities. Beyond individual venues, Keiser et al. (2018) show that ozone warnings reduce monthly visitation at 33 major US national parks, confirming that pollution-induced avoidance extends to leisure and recreation and imposes welfare costs well beyond health outcomes alone.

The scale of what disclosure can achieve is most vividly demonstrated by Barwick et al. (2024), who study China’s 2013 launch of a nationwide real-time air quality monitoring and public disclosure programme as a natural experiment. The programme triggered cascading behavioural responses: stronger avoidance of outdoor exposure, increased spending on protective products, and shifts in residential demand towards cleaner areas. These responses collectively reduced the mortality impact of air pollution, with conservative estimates implying a 9% reduction in pollution-related mortality. The programme’s health benefits outweigh its costs by an order of magnitude, making it one of the highest-return environmental interventions documented in the economics literature. The finding is especially relevant for LMICs: China in 2013 had severe pollution, weak prior disclosure infrastructure, and a large share of the population with limited access to independent information, conditions that characterise many developing cities today.

Yet other work suggests caution about what alerts alone can achieve without additional interventions. Chen et al. (2018) apply a regression discontinuity design to Toronto’s air quality alert programme using administrative health data on 2.6 million residents from 2003 to 2012. They find that alerts reduced respiratory morbidity outcomes but had no detectable effect on mortality or cardiovascular hospitalisations, leading them to conclude that issuing alerts alone has limited public health impact and that enforced emission control actions are warranted. Aguilar-Gomez (2025) reaches a similar conclusion for Mexico City’s Contingencia Ambiental programme. Using sensor-level traffic data and geotagged accident records, she shows that alerts alone did not improve air quality or health outcomes; the programme only became effective once a driving restriction component was added. Alerts did induce voluntary trip reductions before restrictions took effect, suggesting information shifts behaviour at the margin, but larger reductions in exposure require regulatory backing.

The contrast across these settings raises as many questions as it answers. In China, disclosure alone generated large behavioural responses and measurable mortality reductions; in Toronto and Mexico City, alerts produced at most modest morbidity reductions and required enforcement to achieve broader effects. Whether this reflects differences in baseline pollution levels, the credibility of information sources, the costs of avoidance, or something else entirely is not clear from the existing evidence. It is also worth noting that these are quite different interventions: China’s programme was a large-scale, nationally coordinated roll-out of real-time monitoring with substantial media coverage, while Toronto’s and Mexico City’s were pre-existing alert systems operating in contexts where some information was already available. Importantly, in megacities in South Asia with the highest levels of air pollution, it remains an open question how alerts should be designed when ambient air pollution levels remain persistently high. Do alerts induce avoidance behaviour on the most polluted days? Or might they inadvertently signal non-alert days as safe, even when ambient levels remain harmful? The design of such alert systems should be an active area of research for economists.

For most of sub-Saharan Africa, South Asia beyond India’s largest cities, and much of Latin America, there is very little evidence with credible causal identification on how disclosure or alert programmes affect behaviour or health outcomes. Low-cost sensor networks are now expanding across many of these cities, creating natural experiments that did not exist a decade ago. Whether disclosure delivers similar returns in these settings, and what design features drive the difference, are open empirical questions with direct policy relevance.

Defensive investments: Masks, purifiers, and the limits of private protection

Where consumer markets for protective equipment exist, purchasing data provides a revealed-preference window into how households value clean air. Zhang and Mu (2018) use daily mask purchase data and ambient air quality readings in China to show that households respond to particulate pollution with large and strongly nonlinear expenditure responses: a 100-point increase in the Air Quality Index raises purchases of all masks by roughly 55% and of anti- PM2.5 masks by roughly 71%, with especially sharp increases during extreme pollution episodes. This pattern is consistent with episodic avoidance: households scale up protection when pollution becomes severe rather than sustaining costly protection continuously. Wang and Zhang (2023) sharpen this by exploiting China’s staggered rollout of real-time air quality monitoring as a quasi-natural experiment. They find that improved information provision increased PM2.5 respirator expenditures by around 32%, with larger effects on heavily polluted days.

Complementary evidence from Delhi reaches similar conclusions through a different method. Baylis et al. (2023) implement a field experiment offering randomly priced pollution masks to low-income households and find near-zero baseline demand that rises more than fivefold when respondents receive information on the health effects of air pollution. In contrast, correcting beliefs about peer disapproval of mask-wearing has no effect, prior experience with masks does not raise future demand, and a government campaign distributing five million masks across Delhi increased usage temporarily but did not raise willingness to pay. The results point to limited information about health risks, rather than social norms, income, or experience, as the primary barrier to demand for clean air among low-income households.

Air purifiers extend the same logic indoors. Ito and Zhang (2020) use scanner data and quasi-experimental variation in pollution from China’s Huai River heating policy to estimate willingness-to-pay (WTP) for cleaner air through purifier demand, finding households are willing to pay around $1.34 per year to remove 1 µg/m3 of PM10, with substantial heterogeneity by income and exposure. The methodological contribution here is that variation in the quality of purifier purchases can recover values for cleaner air in settings where hedonic housing or wage gradients are difficult to estimate.

Both lines of evidence share an important limitation. Defensive investments provide a lower bound on WTP: households using masks or purifiers still face residual exposure outdoors, in transit, and at work, and they may suffer discomfort, anxiety, and forgone leisure that protective devices do not address. A useful synthesis by Allen and Barn (2020) reviews randomised evidence on purifiers, facemasks, and behavioural modifications. They find that HEPA-filter purifiers can substantially reduce indoor PM2.5 and improve subclinical cardiopulmonary indicators; that well-fitting N95 respirators reduce exposure with some evidence of cardiovascular benefit; and that evidence on health impacts of behavioural modifications such as staying indoors or closing windows is comparatively thin. The distinction between reducing exposure and improving health matters: defensive spending only translates into welfare gains if it delivers meaningful dose reductions, which is far from guaranteed for lower-quality products or intermittent use.

Outside China, a striking feature of the emerging literature is low adoption despite extreme pollution. Chowdhury et al. (2026) study middle-income households in Dhaka, where fewer than 1% own an air purifier despite indoor PM2.5 averaging 150 µg/m3 in winter. Using two field experiments across 3,400 households, they systematically rule out the standard barriers to adoption. Free monitors improved beliefs about indoor pollution severity but did not raise willingness to pay. Free purifiers reduced uncertainty about effectiveness but households used them fewer than 40 minutes per day even when compensated for electricity costs. Showing households a real-time 44% drop in PM2.5 from purifier operation produced no increase in willingness to pay. Interest-free payment plans had no effect. The study also shows that purchase-based estimates of willingness to pay for clean air are methodologically fragile: assumptions about expected usage shift implied valuations by an order of magnitude, and survey-reported usage is unreliable. Using high-frequency objective usage data and randomised variation in the marginal cost of operation, they recover a tightly bounded flow-value willingness to pay of essentially zero to $0.30 per µg/m3 per year. The results suggest that low demand for clean air is not primarily a market failure correctable by information or credit provision, raising difficult questions for policies that rely on sustained complementary household action.

Avoidance through mobility and time use: commuting, travel modes, and short-run WTP

Defensive behaviour also occurs through the allocation of time and the choice of travel mode, which can change the degree to which individuals are exposed to outdoor air. Li et al. (2022) study pollution avoidance through travel mode choice in Beijing, modelling trade-offs between indoor and outdoor travel modes for compulsory work trips. They estimate a short-term willingness-to-pay to avoid PM2.5 exposure and interpret it as a lower bound on longer-run WTP for cleaner air, since people can adjust more margins (including location choice) over longer horizons. Beyond the specific estimates, the broader contribution is to highlight that a sizeable share of defensive behaviour is not a product purchase at all, but a re-optimisation of daily routines under constraints.

The general lesson for policy is that restrictions or advisories that assume people can easily “stay indoors” may understate real costs, especially for workers with inflexible schedules, outdoor occupations, or long commutes. This matters for distribution: the ability to avoid exposure by shifting time or travel is often tightly linked to job type, housing location, and access to transport.

Medium-run defensive behaviour: Sorting and migration

While much of the avoidance literature studies high-frequency behaviour, another important defensive margin is relocation. Migration can be interpreted as a form of avoidance that changes exposure permanently (or at least over multi-year horizons). Chen et al. (2022) estimate the effect of air pollution on migration in China using changes in thermal inversion strength over five-year periods as an instrument for medium-run pollution changes. They find large responses of migration inflows and outflows to pollution, including sizeable compositional effects: cleaner places attract and retain more educated migrants early in their careers, while dirtier places lose population through net out-migration.

Khanna et al. (2025) quantify the aggregate consequences of this re-sorting using a spatial equilibrium model fitted to 18 years of Chinese pollution and migration data, instrumenting for pollution using upwind power plants and thermal inversions. Skilled workers emigrate more from polluted cities than unskilled workers, concentrating in cleaner cities where their relative scarcity is lower and their productivity contribution smaller. The resulting aggregate productivity loss from pollution-induced re-sorting is roughly as large as the direct health effects of pollution, suggesting that standard benefit-cost analyses that count only health damages may understate the true economic costs of air pollution by approximately half. Hukou policy restrictions that limit low-skilled workers’ ability to follow amplify these distributional consequences: unskilled workers bear both the pollution and the productivity losses from the departure of skilled colleagues.

For LMICs, the migration margin raises two further measurement issues. First, pollution-driven sorting complicates the interpretation of place-based health and productivity gradients, because population composition becomes endogenous to pollution. Second, if wealthier households and skilled workers are more mobile, relocation becomes a highly unequal defensive strategy, potentially leaving behind more vulnerable populations with fewer private options and weaker political voice.

Defensive investments by firms and institutions

A compelling frontier for defensive behaviour lies in workplaces and institutions. Since many adults and children spend a large fraction of waking hours at work or school, indoor air quality in these settings can be a major component of total exposure, and employers and schools may have strong incentives to invest in filtration.

Garg et al. (2025) study small garment firms in Dhaka in a randomised experiment and find that air purifiers generate productivity gains of around 10% and profit increases of around 18%, with payback under three months during peak production months. Yet adoption outside the experiment remains negligible, with survey evidence pointing to uncertainty about returns and perceived irrelevance during slack periods as key barriers. These results connect to classic themes in development economics: uncertainty about returns, inattention, and the importance of complementarities. They also highlight that private defensive investment at the firm level can have spillovers: cleaner indoor air at work affects workers as well as owners, and these benefits may not be fully internalised if labour markets are imperfect.

A growing literature on schools reaches broadly similar conclusions about the returns to filtration while adding important nuance. Gilraine (2023) exploits the Aliso Canyon gas leak as a natural experiment in which air filters were installed in every classroom within five miles of the site. Using a spatial regression discontinuity design, he finds that filters raised mathematics and English test scores by 0.1–0.2σ, at less than 1% of the cost of achieving comparable gains through class size reductions. Bharti et al. (2025) conduct an RCT in grade-2 classrooms in Lahore, Pakistan, where baseline indoor PM2.5 averaged 191 µg/m3, an order of magnitude higher than in contexts like Milan or Bogota. Air purifiers reduced indoor concentrations by 25% and raised test scores by 0.15σ within three weeks, with gains concentrated in mathematics and fluid intelligence.

The study also finds reductions in disruptive classroom behaviour, with score gains largest in above-median disruption classrooms, suggesting a learning-environment channel alongside the direct cognitive pathway. At $4.72 per student-year, returns are estimated at 36 times intervention costs. However, the treatment had no detectable effect at a second follow-up six weeks later, because school officials had been switching the purifiers off due to complaints about cold temperatures and electricity costs, and switched them back on only in anticipation of the pre-announced survey. The purifiers were effective on exam day itself but the preceding weeks of clean air exposure were absent, so cumulative gains had not accumulated. Bonan et al. (2025) find complementary evidence from an RCT in Italian schools: air purifiers reduced indoor PM2.5 by 32% and lowered student absences, though only when outdoor pollution was relatively low, consistent with the concave C–R relationship discussed in Section 2.1.

For full reference list see the end of the conclusion chapter.