brazilian farmer applies pesticide to soybean

Herbicides widely used in agriculture increase infant mortality


Published 02.04.24

The use of glyphosate in Brazilian agriculture led to a deterioration of birth outcomes in surrounding populations

Humans have a long history of using herbicides and pesticides—substances that kill weeds and insects with limited harm to crops. However, these substances can also have negative health and environmental effects, prompting regulatory restrictions and sometimes prohibition. The most emblematic case is DDT, a once widely used insecticide that was later banned due to its perceived negative consequences.

This is particularly true for subclinical toxicity, defined as the effect on populations at large, not subject to direct poisoning but to low-level exposure through the ingestion of water or food. Landrigan (2018) notes that “(. . . ) recently, understanding has increased that acute poisoning is only the visible tip of a large iceberg and that pesticides are capable of causing a wide range of asymptomatic effects at levels of exposure too low to produce overt signs and symptoms (p.E.1)”. Small concentrations of pesticides are indeed recurrently detected in the bodies of the majority of individuals in Western societies, including those who live in urban areas and have no direct contact with the respective substances. The population potentially affected by subclinical toxicity is therefore much larger than that affected by direct exposure. However, it is not clear whether such low concentrations have any negative health implications. Small probabilistic effects spread over very large populations limit the potential role of experiments due to lack of statistical power, but, at the same time, mean that aggregate welfare losses may be substantial. Understanding the externalities of pesticide use can contribute to a public debate that, up to now, has been dominated by organised economic interest and scientific controversies (Mesnage and Antoniou 2017, Landrigan and Belpoggi 2018).

Glyphosate is the most heavily used herbicide in human history, accounting for 30% of the aggregate value of the international herbicide market in 2017 (Benbrook 2016, DataIntelligence 2020). In the European Union, where it is tightly controlled, it accounts for 34% of the total use of herbicides (2017 weight of active ingredients, from Antier et al. 2020). In Brazil, which currently alternates with the US as the top soybean producer, it accounts for 62% of total herbicide use and over 35% of total pesticide use (averages from 2009 to 2016, from Alcantara-de-la Cruz et al. 2020). Worldwide, glyphosate usage has increased fifteen-fold since the development of glyphosate-resistant (“Round-up Ready”) genetically modified seeds, particularly soybean seeds, in the mid-1990s. Some fear that, with the ongoing development of new varieties, its usage may grow by yet another 800% between 2017 and 2025 (DataIntelligence 2020).

In our recent research (Dias, Rocha and Soares 2023), we use a natural experiment to assess the effect of the agricultural use of glyphosate on human health in Brazil, during the period between 2000 and 2010, when soybean production expanded rapidly following the introduction of genetically modified seeds, resistant to glyphosate. More specifically, we study the effects of glyphosate on events surrounding birth – and infant mortality, in particular – because the exposure period can be clearly identified, as opposed to potential long-term effects of continued exposure, which would require longitudinal data with past residence history. In addition, existing lab evidence indicates that human embryos are particularly responsive to environmental conditions, and research has shown that glyphosate can affect both placental cells and the fetus in utero directly (Benachour and Séralini 2009, Poulsen et al. 2009).

There are two main challenges in estimating the causal effect of glyphosate on human health. First, the adoption of glyphosate-resistant seeds by soybean producers is linked to characteristics like local entrepreneurship, producers’ coordination capacity, and distribution infrastructure. All these characteristics are likely correlated to socioeconomic characteristics and, ultimately, with human health. To capture the effect of glyphosate without these correlations, we construct an instrumental variable based on the local natural suitability to this new technology. The idea is that natural suitability should not be correlated with other characteristics, like entrepreneurship, but should influence the adoption of glyphosate-resistant soybean seeds. Second, the adoption of glyphosate-resistant soybeans increases agricultural productivity, which in turn should affect the local economy, leading to effects on human health. To separate the effect of glyphosate on health from the effects of socioeconomic characteristics on health, we exploit the idea that when glyphosate is used in one particular area, it can contaminate water sources, be carried downstream to other areas, and affect people who consume contaminated water far away from the location where the pesticide was used. We leverage detailed information on water basins and the water flow within each of these basins. This idea is summarised in Figure 1.

Figure 1: Illustration of the identification strategy

Illustration of the identification strategy

Notes: The subdivisions in the map indicate municipalities within the same basin (we use what is called the level 3 ottobasins). The municipality marked in red is the reference municipality, or AMC (the Minimum Comparable Area, which is used due to the fact that some municipalities are created and others are merged during the period). The lighter color in the figure represents municipalities that are upstream from the reference municipality according to the subdivision of the basin (or level 4 sub-basins), while the darker color indicates municipalities that are downstream from the reference municipality. The intermediate color indicates municipalities that are at the same sub-basin as the reference municipality and, therefore, cannot be unequivocally considered upstream or downstream from it. Our main treatment variables are constructed considering the use of glyphosate in the lighter area, meaning considering only municipalities unequivocally upstream from the reference municipality. By excluding municipalities at the same level as our reference from this calculation, we also minimize concerns related to the correlation in socioeconomic characteristics between the reference municipality and the immediately surrounding areas.

Our main results, summarised in Figure 2, show a deterioration in birth outcomes downstream of areas where glyphosate use increased. We estimate significant increases in infant mortality, the incidence of preterm births, and the frequency of low birth weights. Our results indicate that the average increase in glyphosate use in the sample during this period led to an increase in the infant mortality rate of 0.88 per 1,000 births, or 5% of the mean. This corresponds to an increase of 503 infant deaths per year. Since we are looking at areas distant from locations of use, and focusing only on infant mortality, this number is likely to underestimate the overall effect of glyphosate use on human health.

Figure 2: Results for Infant Mortality Rate – Municipalities in the Brazilian Center-West and South Regions, 2000-2010

Results for Infant Mortality Rate – Municipalities in the Brazilian Center-West and South Regions, 2000-2010


Notes: This plot displays the result of a reduced-form specification in which IMR is regressed on the potential gain in productivity in the area upstream from each municipality interacted with year fixed-effects (with the coefficient in the last year before the introduction of the new technology, 2003, normalized to zero). Confidence intervals are computed at the 95% level.

Looking at causes of death, we show that the mortality response is consistent with what would be expected from exposure to glyphosate during pregnancy: 56% of the total effect come from perinatal period conditions and 19% come from respiratory conditions. We also conduct a series of additional exercises to provide evidence that the effect we estimate is associated with the use of glyphosate. First, we show that the documented effect is working through water bodies and that it is associated with something that is carried from surrounding soil into the water. Second, we show that the effect is associated with the expansion of soybean production and not a result of some spurious spatial correlation or overall expansion in agricultural activity. And third, we show that it is not due to some other form of water contamination brought about by the expansion in soybean production.

Scientists, especially biochemists, have recently reexamined claims that glyphosate is a safe pesticide with little to no effect on human health. These studies have typically used controlled laboratory experiments. Our work provides evidence that glyphosate can affect human populations at large in a real-world setting, at the levels of use typically observed in agriculture.

Few cases manifest the trade-off between agricultural productivity and the external effects of pesticides so clearly as soybean production in Brazil. Welfare losses spread over a large population put in perspective the major economic benefits of technology adoption, both at the local and aggregate levels. Since the type of externality documented here was unknown when current regulations were originally set in place, a new discussion must be initiated on the optimal regulatory mark for the future use and handling of glyphosate-based herbicides.


Alcantara-de-la Cruz, R, G Moraes-de Oliveira, L Bianco-de Carvalho, and M F Fernandes-da Silva (2020), "Herbicide Resistance in Brazil: Status, Impacts, and Future Challenges", in Pests - Classification, Management and Practical Approaches, IntechOpen, pages 1-25.

Behrman, R E and A S Butler (2007), Preterm Birth: Causes, Consequences, and Prevention, National Academies Press, Committee on Understanding Premature Birth and Assuring Healthy Outcomes.

Benachour, N and G E Séralini (2009), "Glyphosate formulations induce apoptosis and necrosis in human umbilical, embryonic, and placental cells", Chem. Res. Toxicol., 22(1):97–105.

Benbrook, C M (2016), "Trends in glyphosate herbicide use in the United States and globally", Environmental Sciences Europe, 28(1):3.

Bharadwaj, P, J Fenske, R A Mirza, and N Kala (2018), The Green Revolution and Infant Mortality in India, Technical Report 385, The University of Warwick.

Bustos, P, B Caprettini, and J Ponticelli (2016), "Agricultural Productivity and Structural Transformation: Evidence from Brazil", American Economic Review, 106(6):1320–1365.

DataIntelligence (2020), Herbicide market, size, share, opportunities and forecast, 2020-2027, Technical report, Herbicide Market, Size, Share, Opportunities and Forecast, 2020-2027.

Dias, M, R Rocha, and R R Soares (2023), "Down the River: Glyphosate Use in Agriculture and Birth Outcomes of Surrounding Populations", The Review of Economic Studies, 90(6):2943–2981.

Feder, G, R Just, and D Zilberman (1985), "Adoption of Agricultural Innovations in Developing Countries: A Survey", Economic Development and Cultural Change, 33:255–298.

Poulsen, M S, E Rytting, T Mose, and L E Knudsen (2009), "Modeling placental transport: correlation of in vitro BeWo cell permeability and ex vivo human placental perfusion", Toxicology in Vitro, 23(7):1380–1386.