The health capital theory posits that dietary behavior and food consumption are fundamental not only to individual well-being but also as investments in health (Grossman, 1972). These investments are inherently dynamic, allowing individuals to adjust their behaviors in response to external shocks, such as changes in environmental conditions. Cropper (1981) extends this framework, arguing that health shocks, such as air pollution exposure, prompt individuals to modify their behaviors, including dietary choices, as a means of mitigating adverse effects. Dietary adjustments are particularly noteworthy as they offer a relatively low-cost and accessible strategy for managing health risks, especially in contexts where access to healthcare or other resources is constrained.1 This issue is especially relevant in rapidly urbanizing and industrializing countries like China, where worsening air quality has become a pressing public health concern. Beyond the direct health risks, pollution also imposes broader economic costs by altering consumer behavior, such as increased spending on healthier foods, which may strain household budgets and exacerbate welfare losses.

While behavioral responses to air pollution, such as avoidance actions (Zhang and Mu, 2018; Ito and Zhang, 2020) and increased medical expenditures (Deschênes et al., 2017; Barwick et al., 2024), have received significant attention, limited research examines how environmental shocks influence dietary behavior—a critical yet underexplored aspect of health investments.2 Dietary responses provide valuable insights into household decision-making under environmental risks, reflecting a direct and immediate coping mechanism beyond avoidance or healthcare costs. However, investigating these behaviors poses significant empirical challenges. The scarcity of high-frequency household consumption data has historically hindered efforts to capture short-term adjustments, while the endogeneity of pollution levels—stemming from unobserved factors that simultaneously affect air quality and consumer behavior—further complicates causal analysis. Addressing these gaps is essential not only for advancing theoretical understanding of environmental determinants of health but also for informing policies that strengthen health resilience, reduce disparities, and mitigate welfare losses in the face of environmental shocks.

This study addresses critical gaps in the literature by examining how households adjust their food consumption in response to air pollution. We leverage a unique high-frequency dataset encompassing over 30,000 households across 25 major Chinese cities from 2014 to 2019, with each household meticulously tracked for an average of three years. The data collection methodology involved the use of barcode scanners for daily recording of food purchases, supplemented by monthly reports on household demographics and personal information. This granular, daily tracking captures detailed information on food categories, quantities, and purchasing channels, enabling precise monitoring of short-term fluctuations in household food consumption in response to environmental factors such as air pollution. We incorporate Urban Household Survey (UHS) data, which provides monthly household-level information on income, expenditures, and categories like dining out and medical expenses, offering a comprehensive view of household economic behavior to enhance the analysis. Additionally, we link household consumption data with city-level PM2.5 pollution data to investigate the relationship between air pollution and household dietary adjustments.3

To address the potential endogeneity of pollution levels, we employ an instrumental variable (IV) strategy that leverages changes in wind direction as an exogenous shock to local air pollution, building on established methodologies in the literature (Deryugina et al., 2019; Chen et al., 2023). This approach exploits the fact that wind patterns create exogenous variation in pollution levels, independent of household characteristics or behaviors. Recognizing that the relationship between wind direction and pollution varies across cities, we further examine heterogeneity, distinguishing between cities where pollution levels are highly sensitive to wind patterns due to geographic or industrial configurations and those where wind effects are less pronounced. By addressing potential endogeneity concerns, this approach enables us to credibly estimate the causal impact of air pollution on household food consumption and dietary behavior.4

The first-stage results provide strong validation of our instrumental variable (IV) approach, demonstrating a robust and statistically significant relationship between wind direction and local PM2.5 concentrations. Dominant wind patterns linked to major pollution sources exert the most pronounced effects on pollution levels, reflecting the geographic and industrial configurations of the cities studied. For instance, in Beijing, wind blowing from the heavily industrialized regions of Hebei and Shanxi significantly increases PM2.5 levels, underscoring the city’s reliance on external pollution transport. In contrast, Guangzhou, located at the center of the Pearl River Delta industrial cluster, exhibits a statistically insignificant relationship between wind direction changes and pollution levels, as local emissions dominate and pollution sources are distributed across all directions. These heterogeneous results highlight the critical role of geographic and industrial contexts in shaping the relationship between wind patterns and pollution, confirming the relevance of wind direction as an instrument. This robust first-stage relationship establishes a credible foundation for estimating the causal impact of air pollution on household dietary behavior.

Our findings reveal that air pollution significantly increases household food expenditures, particularly on healthier food items.5 Specifically, a one-standard-deviation increase in PM2.5 (about 25 µg/m³) is associated with a 9.3 yuan rise in weekly spending on healthy foods such as vegetables, fruits, and dairy products. This represents approximately 8 % of the within-household standard deviation in total weekly food expenditures and 11 % of the standard deviation in healthy food expenditures. Analysis of food quantities further highlights a notable increase in the intake of key nutrients, including proteins and micronutrients such as vitamins and iron, alongside a modest rise in overall energy intake. This shift toward healthier food consumption translates into a reduction in nutrient deficiencies, with a one-standard-deviation increase in PM2.5 narrowing the gap between actual and optimal nutrient intake by 3.5 % based on the standards outlined in the Dietary Guidelines for Chinese Residents.6

Spending on unhealthy foods, such as puffed snacks and oils, remains unchanged in response to air pollution, and no significant adjustments are observed in carbohydrate or fat intake. This suggests that dietary changes are specifically targeted toward nutrient-dense, health-promoting foods rather than reflecting a broad increase in caloric intake. These findings highlight the selective nature of dietary responses to pollution, driven primarily by health concerns rather than indiscriminate changes in consumption patterns. Collectively, the results indicate that air pollution serves as a catalyst for short-term improvements in dietary quality, prompting households to reallocate expenditures toward healthier food categories as a strategy to mitigate perceived health risks.

An analysis of temporal dynamics reveals that air pollution’s effects on household food consumption are immediate and short-lived. Pollution levels in the current week significantly impact food consumption, while levels in previous weeks do not, and the effect dissipates when data are aggregated to bi-weekly or monthly intervals. This indicates that dietary adjustments are temporary responses to short-term air quality fluctuations rather than permanent changes in preferences or habits. These findings align with studies like Zhang and Mu (2018), which document short-term avoidance behaviors, and Deryugina et al. (2019), which emphasize the transient nature of responses to environmental shocks, consistent with broader literature on temporary protective actions (Deschênes et al., 2017).

To examine inequalities in the effects of air pollution on dietary adjustments, we analyze variations across income levels and household compositions. High-income households exhibit the strongest response, increasing healthy food spending by 15.2 yuan for a one-standard-deviation increase in PM2.5, compared to 9.8 yuan among middle-income households and no significant change for lower-income households. Similarly, households with elderly members show a greater response, with a 23.4 yuan increase in weekly healthy food spending compared to 7.55 yuan in households without elderly members. This disparity likely reflects heightened health concerns among the elderly (Fischer et al., 2003; Lim et al., 2012) and underscores the roles of health consciousness and financial capacity in shaping dietary responses to environmental shocks.

Moreover, to understand the full scope of changes in household dietary behavior, we investigate the changes in household dining out behavior. Using monthly expenditure data from the Urban Household Survey, we find that air pollution reduces household spending on dining out, consistent with previous findings that individuals reduce outdoor activities to avoid exposure to polluted air (Bresnahan et al., 1997; Zivin and Neidell, 2009). However, the magnitude of this reduction is relatively small, suggesting that the effect does not materially alter our baseline findings on at-home dietary adjustments.7

Finally, we analyze city-level data from the National Supermarket Retail Price Data to test whether air pollution affects food prices and could explain the observed dietary changes. The analysis reveals no significant relationship between air pollution and food prices, consistent with Ito and Zhang (2020). This suggests that dietary adjustments are driven by health concerns rather than cost changes, with air pollution acting as a catalyst for short-term shifts toward healthier food choices, not an economic shock affecting affordability.

We perform a series of robustness checks. First, we adopt alternative instrumental variables, including city-specific wind directions, thermal inversions, wind speed, and a double lasso procedure to select the most pollution-sensitive wind directions, all of which yield consistent results. Second, we consider alternative measures of pollution, including non-linear effects of PM2.5, the Air Quality Index (AQI), and satellite-based PM2.5 estimates, with similar findings. Third, we alter model specifications, such as excluding weather covariates or city-by-year fixed effects, and the results remain robust. These checks confirm the robustness of our key finding that air pollution prompts short-term increases in healthy food consumption.

This study contributes to the literature on environmental and health economics by uncovering how air pollution dynamically influences household behavior. Building on Grossman’s (1972) health capital framework, we show that households respond to pollution not only through traditional avoidance actions or healthcare utilization (Deschênes et al., 2017; Ito and Zhang, 2020) but also by making significant dietary adjustments. These reallocations toward nutrient-dense foods represent a critical yet underexplored coping mechanism for mitigating health risks. By identifying this indirect pathway, our findings advance the understanding of health production functions and the broader interplay between environmental stressors and household decision-making.

Our findings also suggest potential welfare implications of short-term behavioral responses to environmental shocks.8 Unlike prior research that focuses on long-term impacts such as productivity losses (Chang et al., 2012) or chronic health outcomes (Chen et al., 2013), we show that air pollution induces changes in consumption patterns and increases short-term volatility in household spending, both of which may be indicative of welfare losses given households’ preference for stability in consumption (Carroll, 2001).9 These findings suggest that conventional willingness to pay (WTP) estimates for clean air underestimate the full economic costs of pollution by ignoring these short-term adjustments.10 Moreover, our use of high-frequency household data—uncommon in developing country studies—represents a methodological innovation, complementing datasets like Nielsen Homescan in developed economies (Allcott et al., 2019) and providing a robust framework for analyzing transient shocks.11

Finally, this research bridges environmental and development economics by examining how pollution affects household behavior in a rapidly urbanizing economy. While much of the literature focuses on developed countries (Currie and Neidell, 2005; Jbaily et al., 2022), we provide evidence from a developing context, where pollution levels are more severe, and coping mechanisms are constrained.12 Our findings highlight the disproportionate burden of pollution on low-income households, as higher-income households demonstrate greater dietary adaptability, mitigating health risks more effectively. These disparities underscore the need for targeted interventions to enhance resilience among vulnerable populations. By integrating short-term welfare losses and behavioral adjustments into cost-benefit analyses of air quality improvements, policymakers can address the unequal impacts of pollution more equitably.

The paper is organized as follows: Section 2 introduces the research background and theoretical model. Section 3 presents the data and descriptive statistics. Section 4 outlines the empirical strategy. Section 5 reports the IV estimation results, including the effects of wind direction on pollution and food expenses. Section 6 discusses the main findings on air pollution's impact on dietary behavior. Section 7 conducts robustness tests, and Section 8 concludes.