However, farmer surveys give a more accurate picture of the total farmlevel economic effects of GE-crop adoption in terms of the secondary behavioral changes resulting from adoption (e.g., adoption of conservation tillage and changes in the timing of pesticide application). Moreover, it is rarely the case that a farmer would or could choose to adopt a GE cultivar to replace a non-GE cultivar that is an isoline or near-isoline, so relying on agronomic experimental data to measure the economic differences can be biased. Also, only farmer surveys can reveal the value of the changes in nonpecuniary characteristics that can occur with the adoption of GE cultivars.
Social scientists often are able to statistically control for certain influencing factors for which there are data (apart from the GE-crop treatment) by using multiple regression techniques in econometric models. That is, differences in economic conditions and crop or management practices that also influence yield or other outcomes are held constant so that the effect of adoption can be isolated. For example, in research on GE crops, economists control for many factors, including output and input prices, pest infestation levels, farm size, operator characteristics, and management practices such as crop rotation and tillage. In addition, economists control for self-selection and simultaneity (of GE adoption and pesticide use decisions) using particular types of econometric models. To account for simultaneity of decisions and self-selectivity, a two-stage model may be used. The first stage consists of the adoption-decision model for GE crops. The second stage then uses the findings from the first stage to examine the impact of using GE crops on yield, farm profit, and pesticide use.
The Counterfactual. Ideally, measuring the impact of a treatment requires the observation of the results that would emerge in the absence of the treatment: a counterfactual. Aside from controlled experiments, it is not possible to observe this counterfactual outcome. Rather, the counterfactual is inferred by methods such as those summarized above (e.g., controlling for all other influencing factors). Moreover, regarding environmental impacts, Ferraro (2009) argues that “elucidating casual relationships through counterfactual thinking and experimental or quasi-experimental designs is absolutely critical in environmental policy and that many opportunities for doing so exist.” The use of the two-stage estimation procedure to correct for selection bias exemplifies such a quasi-experimental design. However, Ferraro also admits that “not all environmental programs are amenable to experimental or quasi-experimental design.” In those cases, firm conclusions cannot be drawn about the causative factors inducing GE-crop adoption or other outcomes of interest.