Read an Excerpt
CHAPTER 1
Heterogeneous Yield Impacts from Adoption of Genetically Engineered Corn and the Importance of Controlling for Weather
Jayson L. Lusk, Jesse Tack, and Nathan P. Hendricks
Although agriculture has historically experienced one of the highest rates of productivity growth in the US economy (Jorgenson, Gollop, and Fraumeni 1987), there is evidence that agricultural productivity growth is beginning to slow (Alston, Andersen, and Pardey 2015; Alston, Beddow, and Pardey 2009; Ray et al. 2012). The decline in productivity growth has coincided with concerns about food price spikes, social instability, food insecurity, population growth, drought, and climate change (Bellemare 2015; Ray et al. 2013; Roberts and Schlenker 2013; Schlenker and Roberts 2009; Tack, Barkley, and Nalley 2015a,b). This confluence of problems has prompted interest in determining whether certain technologies can promote gains in crop yields, and none has been more controversial than biotechnology.
Many previous studies have investigated whether adoption of genetically engineered (GE) crops has increased yield (e.g., see reviews in Fernandez-Cornejo et al. 2014; Klümper and Qaim 2014; NASEM 2016), and the consensus from the microlevel data and experimental studies is that adoption of GE crops, particularly insect-resistant Bt varieties targeting the corn borer, have generally been associated with higher yield. However, ample skepticism remains, with high-profile popular publications purporting that GE crops have failed to live up to their promise of yield increases (e.g., Foley 2014; Gurian-Sherman 2009; Hakim 2016).
A variety of factors might explain the divergence in views about the yield effects of GE crops, but one of the main issues is that adoption of GE crops does not appear to have had much effect on trend yields when investigating national-level yield data (Duke 2015), nor do yield trends appear much different in developed countries that have and have not adopted GE varieties (Heinemann et al. 2014). As the NASEM (2016, 66) put it, "The nation-wide data on maize, cotton, or soybean in the United States do not show a significant signature of genetic-engineering technology on the rate of yield increase." This raises the question of whether the yield-increasing effects of GE crops observed in particular locations and experiments can be generalized more broadly and, if so, whether the impact on crop yields varies spatially.
In this chapter, we show that simple analyses of yield trends mask important weather-related factors that influence the estimated effect of GE crop adoption on yield. Our analysis couples county-level data on corn yields from 1980 to 2015 and state-level adoption of GE traits with data on weather variation and soil characteristics. Using state-level adoption data does not induce measurement-error bias because state-level aggregate adoption is necessarily uncorrelated with the deviation of a particular county's adoption from the state-level aggregate. Using state-level adoption data does induce serial correlation of the error term, which we address with two-way clustering.
A number of important findings emerge from our analysis. First, changes in weather and climatic conditions confound yield effects associated with GE adoption. Without controlling for weather, adoption of GE crops appears to have little impact on corn yields; however, once temperature and precipitation controls are added, GE adoption has significant effects on corn yields. Second, the adoption of GE corn has had differential effects on crop yields in different locations even among corn-belt states. However, we find that ad hoc political boundaries (i.e., states) do not provide a credible representation of differential GE effects. Rather, alternative measures based on soil characteristics provide a broad representation of differential effects and are consistent with the data. In particular, we find that the GE effect is much larger for nonsandy soils with a larger water-holding capacity. Overall, our studies show that GE adoption has increased yields by approximately 18 bushels per acre on average, but this effect varies spatially across counties ranging from roughly five to 25 bushels per acre. Finally, we do not find evidence that adoption of GE corn has led to lower yield variability, nor do we find that current GE traits mitigate the effects of heat or water stress.
The adoption of GE crops does not necessarily imply that farmers perceived yield benefits, because there are several other benefits associated with the adoption of GE crops — primarily through a reduction in the cost of production. The nonyield benefits have come in the form of labor savings, reduced insecticide use, and improved weed and pest control, which has facilitated the ability to adopt low- and no-till production methods, alter crop rotations, and utilize higher planting densities (Chavas, Shi, and Lauer 2014; Fernandez-Cornejo et al. 2014; Klümper and Qaim 2014; Perry, Moschini, and Hennessy 2016; Perry et al. 2016). Revealed preferences of US farmers indicate producer benefits over and above the substantially higher price of GE corn relative to conventional corn (Shi, Chavas, and Stiegert 2010). The rapid adoption of GE corn by farmers also provides evidence of these benefits. GE corn was first grown commercially in the United States in 1996. In just four years, a quarter of the corn acres were planted with a GE trait, and in less than 10 years, adoption had spread to more than half the US corn acres. In 2016, 92 percent of US corn acres were planted with GE corn, with 81 percent of the total GE corn acreage being planted with "stacked" varieties that are both insect resistant and herbicide tolerant. It is also important to recognize that GE crops can increase production through the expansion of designated crop-planting areas (i.e., the extensive margin) because greater yields and lower costs of production provide incentives to expand crop production (Barrows, Sexton, and Zilberman 2014).
Nonetheless, discussion of yield impacts of GE crops remains at the forefront of public discussions about whether and to what extent bio technology can contribute to food security and help mitigate the effects of climate change. In response to the finding that GE adoption does not appear to alter national-level yield trends, NASEM (2016, 16) recommended that research "should be conducted that isolates effects of the diverse environmental and genetic factors that contribute to yield." Our objective here is to help fill this gap in the literature.
The next section reviews some of the research on the yield effects of GE crops, and we delineate our contribution to the literature. The third and fourth sections discuss the data and methods, followed by the presentation of results. The last section concludes.
1.1 Background
GE crops currently on the market do not increase yield per se. However, they can reduce the gap between actual and potential yield by reducing the adverse effects of weeds and insects (NASEM 2016). It is also possible that crops with GE traits can reduce yields if introduced into less productive varieties not ideally suited to a particular growing region (Chavas, Shi, and Lauer 2013).
Figure 1.1 shows the national trend in US corn yield and the adoption of GE corn from 1980 to 2016. The figure suggests, in the words of Duke (2014, 653), that "yields have continued to increase at the same rate as before introduction [of GE crops]." Leibman et al. (2014) similarly investigated aggregate yields and found after adoption of GE corn a small (0.5 bushels/acre) trend increase; however, no tests of statistical significance were performed. These sorts of aggregate comparisons make no attempt to control for potentially confounding factors such as weather, which could have coincidentally been worse in the 1980s before the adoption of GE corn. Controlling for weather in a national-level trend analysis is difficult due to the nonlinear impacts of weather on yield and highly spatially heterogeneous weather conditions within the country. This motivates the use of disaggregate data to test the impact of GE adoption on yields.
These aggregate investigations can be contrasted with the large literature from agronomic experimental studies that attempt to hold constant many factors such as location and germplasm. Nolan and Santos (2012) summarize the results of more than 30 such studies mainly published between 2000 and 2003. None of the reviewed studies report a statistically significant negative effect associated with Bt GE corn, and nearly all reported positive yield effects associated with the Bt GE trait, with yield gains as high as 19 percent. In their analysis, Nolan and Santos (2012) combined data sets from multiple experiments conducted by 10 different state agricultural extension services from 1997 to 2009. They found, after controlling for weather, agronomic inputs, management, and soil characteristics, that planting of Bt GE corn led to yield gains of around 14 bushels per acre, although when the only GE trait present was herbicide tolerance, yield was unaffected or slightly negative. Despite finding that yield was affected by location, weather, and soil characteristics, the authors did not investigate whether these factors interacted with the GE effect (i.e., whether GE yield gains were higher or lower in different locations, in different weather patterns, or in different soils).
As shown by Chavas, Shi, and Lauer (2013), however, there is likely ample heterogeneity in the effects of GE adoption on mean yield and yield variance. Using experiment data from agricultural experiment stations in Wisconsin from 1990 to 2010, Chavas, Shi, and Lauer (2013) found that GE traits had variable effects on corn yields, depending on the type of GE trait introduced and how long the trait had been used in production, with mean yields significantly increasing relative to conventional non-GE corn for some traits (namely, Bt targeted at the European corn borer) but not others (namely, herbicide-tolerant-only GE corn and GE corn with Bt targeted only at corn rootworm). Additional analysis of the same data by Chavas, Shi, and Lauer (2013) suggests that some of the yield gains attributable to GE hybrids were a result of improvements in non-GE germplasm and the wider availability of higher-quality germplasm. However, regardless of the GE trait analyzed, the authors found a consistent effect on yield variance, with GE crops reducing the variance of corn yields. The authors conclude that GE crops have helped farmers reduce their risk exposure.
As was the case in Nolan and Santos (2012), Chavas, Shi, and Lauer (2013) did not investigate whether the yield effects of GE traits were affected by location, management practices, soil type, and so on. However, there are reasons to believe the potential yield effects of GE adoption are not uniform across location or time. Currently available GE traits rely on Bt to provide protection against the European corn borer and/or corn rootworm, and/or tolerance to certain herbicides (primarily glyphosate). While there are fewer agronomic reasons to suggest that herbicide tolerance would convey significant yield benefits, insect resistance can plausibly lower the gap between potential and realized yield. As discussed by Nolan and Santos (2012), conventionally applied insecticides only provide 60 percent to 80 percent protection against corn borer and rootworm, whereas Bt provides near 100 percent protection. As such, the effect of Bt GE corn relative to conventional corn depends on pest pressure. It has long been known that corn borer and corn rootworm pressures are affected by soil characteristics and weather (e.g., Beck and Apple 1961; Huber, Neiswander, and Salter 1928; Turpin and Peters 1971; MacDonald and Ellis 1990), and prior research has hinted at the fact that yield effects of Bt corn might depend on soil characteristics via their effects on insect populations (Ma, Meloche, and We 2009). Pest pressure is also likely to vary spatially according to the density of corn production, which depends on soil and climatic conditions.
In a paper most similar to the present inquiry, Xu et al. (2013) used aggregate, nonexperimental data and found that GE adoption led to a 19.4 bushel per acre increase in corn yields in the central Corn Belt (Illinois, Indiana, and Iowa). What explains the contrast between the apparent lack of impact of GE adoption in aggregate trend yields shown in figure 1.1 and the results from Xu et al. (2013)? There are a variety of possibilities. For example, Xu et al. (2013) look at county (rather than national) yields, and they control for confounding factors related to weather and fertilizer use. However, it is unclear from Xu et al. (2013) what the impacts are of ignoring these factors. Moreover, these authors only considered limited geographic heterogeneity (they only explored central Corn Belt to noncentral Corn Belt), and they did not consider other factors like soil characteristics or how weather and soil characteristics may influence GE-adoption effects on yield. In addition, the authors did not consider the effects of GE adoption on yield variability.
Another confounding factor that exists when exploring national yield trends is the fact that the number of acres planted to corn has increased significantly over the same period of time that GE traits have been adopted. For example, in the 10 years from 1980 to 1989 prior to adoption of GE corn, 75.7 million acres of corn were planted on average each year in the United States. By contrast, in the most recent 10-year period from 2007 to 2016, during a period of near full adoption of GE traits, on average 91.2 million acres of corn were planted each year in the United States, a 20.5 percent increase. Some of the acreage expansion is a result of GE adoption, as GE traits have increased the viability of continuous corn (planting corn after corn rather than rotating with soybeans; Chavas, Shi, and Lauer 2014), a practice that has historically been associated with significant yield drag (Gentry, Ruffo, and Below 2013). Ethanol policies, among other factors, also led to a dramatic increase in corn prices over the period of GE corn adoption, which both increased the prevalence of continuous corn (Hendricks, Smith, and Sumner 2014) and led to the expansion of corn onto acres that would previously have been considered marginal lands. Combined, these factors suggest that national corn yields would have been higher in recent years had it not been for the expansion of corn acreage.
1.2 Data
We utilize a large panel of roughly 28,000 yield observations spanning 819 counties from 1980 to 2015. We chose 1980 as the starting point for the time series, as this gives us a roughly equal number of years pre- and post-GE adoption, which started in 1996. Roughly 13,000 (45 percent) of the yield observations correspond to the pre-GE period. These data were collected via USDA NASS Quick Stats and correspond to total production divided by harvested acres in each county. As in Xu et al. (2013), we omit any county where (i) more than 10 percent of harvested cropland is irrigated or (ii) yield data was reported for less than two-thirds of the pre-GE years or two-thirds of the post-GE years. Figure 1A.1 in the appendix to this chapter shows that there exists extensive cross-sectional and temporal variation of yields. Note that all tables and figures with a leading S are contained in the accompanying supplementary material.
The limiting factor for the cross-sectional (spatial) representation of the data is the availability of GE adoption data. We utilize the same NASS data as Xu et al. (2013), which reports GE adoption at the state-year level for 13 states: Illinois, Indiana, Iowa, Kansas, Michigan, Minnesota, Missouri, Nebraska, North Dakota, Ohio, South Dakota, Texas, and Wisconsin. These data were first recorded in 2000 for all but North Dakota and Texas, which were recorded starting in 2005, several years after adoption had already started to occur in some areas. We interpolate missing data using predictions from a generalized linear model with a binomial family and a logit link function. A pooled model with state fixed effects provides similar predictions as separate models for each state. We use the latter here. Our interpolation procedure follows the seminal work of Griliches (1957), who modeled the diffusion of hybrid corn seed as logistic growth. Figure 1A.2 provides the observed adoption data in addition to the predictions from both models used to interpolate missing values. Table 1A.1 provides summary statistics for both the observed and observed-plus-interpolated GE adoption rate variables. Figure 1A.3 provides a spatial map of the in-sample counties studied in the analysis.
(Continues…)
Excerpted from "Agricultural Productivity and Producer Behavior"
by .
Copyright © 2019 National Bureau of Economic Research, Inc..
Excerpted by permission of The University of Chicago Press.
All rights reserved. No part of this excerpt may be reproduced or reprinted without permission in writing from the publisher.
Excerpts are provided by Dial-A-Book Inc. solely for the personal use of visitors to this web site.
