I estimate the causal effects of air pollution on healthcare costs in France by combining administrative data on healthcare reimbursements with reanalysis data on air pollution concentrations and weather conditions. I adopt an instrumental variable approach where I exploit daily postcode-level variation in nitrogen dioxide, ground-level ozone and particulate matter concentrations induced by variation in wind speed. I explore effect heterogeneity by patient and location characteristics and by medical speciality. This study presents evidence for substantial healthcare costs caused by exposure to pollution levels that are predominantly situated below current European legal limits. The effects are several orders of magnitude larger than those estimated in the previous literature, suggesting that the healthcare costs of air pollution have been severely underestimated. I find significant heterogeneity of effects across location and patient characteristics, indicating that air pollution reduction policies have the potential to reduce health inequalities.
This study investigates the impact of a couple’s break-up on the economic resources of the household by studying changes in income and food purchases around the time of separation in a panel of French households. I estimate a household fixed effects model to account for unobserved time-invariant household characteristics while controlling for additional time-varying covariates. Household income and food purchases decrease suddenly and significantly at the time of separation and remain lower than pre-separation levels for several years after the break-up. The decrease in food purchases appears to translate into a slight decrease in the female partner’s body mass index (BMI). The share of unhealthy food purchases increases shortly before, during and after separation, indicating that the composition of food purchases changes as well. The decline in food purchases and BMI mainly affects households in the lowest pre-separation income tercile, suggesting that these changes are due to insufficiency of financial resources.
with Marie-Christine Boutron-Ruault, Marie-Aline Charles, Olivier Allais & Guy Fagherazzi
The Developmental Origins of Health and Disease (DOHaD) framework suggests that early-life experiences affect long-term health outcomes. We tested this hypothesis by estimating the long-run effects of exposure to World War II-related food deprivation during childhood and adolescence on the risk of suffering from hypertension and type 2 diabetes at adulthood for 90,226 women from the French prospective cohort study E3N. We found that the experience of food deprivation during early-life was associated with a higher risk of developing type 2 diabetes (+0.7%, 95% CI: 0.073–1.37%) and hypertension (+2.6%, 95% CI: 0.81–4.45%). Effects were stronger for individuals exposed at younger ages. Exposed individuals also achieved lower levels of education, slept less, and were more frequently smokers than unexposed individuals. These results are compatible with both the latency and the pathway models proposed in the DOHaD framework which theorise the association between early life exposure and adult health through both a direct link and an indirect link where changes in health determinants mediate health outcomes.
with Olivier Allais & Pascal Leroy
We estimate the impact of retirement on food expenditure and food quantities purchased, using detailed home-scan panel data on food purchases and household characteristics in France. We identify a causal relationship by exploiting the French legal minimum age for retirement as an exogenous shock to retirement behavior. Upon retirement, households significantly decrease their expenditure on food and the amount of food purchased. Households with lower pre-retirement income appear to be more severely affected. Our results indicate that the decrease in food quantities purchased at the aggregate level is driven by a decline in purchases of food from animal origins. A reduced consumption of animal based food products is likely to undermine the diet balance of retirees.
with Olivier Allais, Philippe Caillou & Michèle Sébag
We assess the adverse impact of residential pesticide diffusion on residents living close to agricultural lands, exposed to pesticides via spray drift and volatilisation beyond the treated areas. This population is largely absent in studies to date. We exploit sensitive health data in combination with newly available data on pesticide pollution. For the sake of a clear focus, we will rely on the body of knowledge relating the exposure to some molecules at precise stages of the pregnancy, to the impaired development of specific cognitive and biological systems. Accordingly, the study will focus on the short and medium-term pesticide impact on newborns and children. We use quasi-experimental methods and new machine learning approaches for causal inference to face the main challenges of non-linearity of the effects, high dimensionality of the potential causes (cocktail effect), data incompleteness, and hidden confounding factors.
with Olivier Allais and Antoine Nebout
We investigate whether individual preferences such as attitudes towards risk, time and ambiguity are correlated with an individual’s exposure to air pollution through her choice of residence and how this impacts health outcomes. For this, we add a module with questions concerning individual preferences for the new wave of data collection of the French cohort study CONSTANCES. This project is currently at the data collection stage.
with Fabien Forge
We study the determinants of climate change adaptation using both machine learning and economic theory. For farmers, crop choice is one of the most effective and cheapest way of mitigating the effects of climate change. Yet it is unclear whether farmers’ adapt in reaction to past weather realisations or in anticipation of climate change. We attempt to answer this question by testing two theories: one in which farmers are only backward looking and a second in which they are also forward looking. Since these two behavioural models do not leave in the same parameter space, we follow Fudenberg et al. (2020) and measure how `complete’ each theory is by comparing their predictive performance to a predictive upper bound defined using machine learning.