The extensive adoption of genetically engineered (GE) varieties in U.S. agriculture has dramatically changed the patterns of pesticide use. How this process ultimately affects environmental risk remains an open question. Previous studies have typically relied on aggregate trends to infer the impact of GE crop adoption on pesticide use, which fails to address selection bias and unobserved heterogeneity. We overcome this limitation by analyzing applied pesticide toxicity using farm-level fixed effects models, estimated with rich plot-level data on >200,000 seed and pesticide choices by U.S. maize and soybean farmers during the 1998–2016 period. We find that applied toxicity was, on average, lowered by the adoption of GE varieties across four target organism groups: mammals, birds, fish, and honey bees. However, most of the toxicity benefits conferred by GE adoption dissipated over time. For herbicide tolerant varieties, this was due to the increased use of old-line herbicides by GE adopters, a likely consequence of the growing problem of glyphosate weed resistance. Applied honey bee toxicity saw the sharpest increase during the GE era, but most of this increase was driven by the adoption of neonicotinoid seed treatments, rather than GE insect resistant traits.

美国农业广泛采用基因工程 (GE) 品种极大地改变了农药的使用模式。这个过程最终如何影响环境风险仍然是一个悬而未决的问题。以前的研究通常依赖于总体趋势来推断采用转基因作物对农药使用的影响，但未能解决选择偏差和未观察到的异质性问题。我们通过使用农场级固定效应模型分析施用的杀虫剂毒性来克服这一局限性，该模型使用 1998-2016 年期间美国玉米和大豆农民选择的超过 200,000 种种子和杀虫剂的丰富地块级数据进行估计。我们发现，平均而言，通过在四个目标生物群中采用 GE 品种降低了应用毒性：哺乳动物、鸟类、鱼类和蜜蜂. 然而，随着时间的推移，采用 GE 所带来的大部分毒性益处都会消失。对于耐除草剂品种，这是由于 GE 采用者越来越多地使用旧系列除草剂，这可能是草甘膦杂草抗性问题日益严重的结果。应用的蜜蜂毒性在 GE 时代增长最快，但这种增长的大部分是由于采用新烟碱类种子处理，而不是 GE 抗虫性状。