Performance Results and Characteristics of Adopters of Genetically Engineered Soybeans in Delaware

Agricultural and Resource Economics Review, Oct 2004 by Bernard, John C, Pesek, John D Jr, Fan, Chunbo

In order to estimate the Cox model, the treatment of ties must be considered. The model was developed for continuous events. However, it is common for events to occur simultaneously (to be tied). Since fanners may make the decision to plant GM soybeans at any time of the year but only the year of the decision is given, times in our study are only accurate to within a year, and thus ties will occur. When this is the case, the Cox model is estimated with the exact method.5

The effect of the change on the survival probability (the probability of continuing to use conventional soybeans) can also be given. The current survival function changes from S^sub c^(t) to S^sub c^(t)^sup exp(β^sub j^Δx^sub j^)^ If the coefficient β^sub j^ is one, the survival function is unchanged. If it is less than one, the survival function will increase (recall that O

Yields and Weed Control Costs

As specified, however, there existed the potential for selection bias in the models. The concern stems from the possibility that unobserved factors within the error terms may also affect the farmer's adoption decision, implying GESoybeans would be endogenous in the models. Because this theoretical concern suggests potentially serious consequences for the analysis of results, the first step was to determine the appropriate methodology for estimating the above models [equations (4) and (5)]. Ordinary least squares would be most efficient only if adoption was exogenous; otherwise, a two-stage least squares procedure would be required.

The adoption variable, GESoybeans, was tested for endogeneity in each model using the procedure outlined in Wooldridge (2003). This involved a two-stage process by which GESoybeans was first estimated as a function of the structural variables in (4) and (5), and instrumental variables selected from the earlier duration analysis. For the second stage, the residuals were included as explanatory variables in the yield and cost models and tested for significance. While significance on these coefficients would indicate GESoybeans was endogenous, the resultingp-values were 0.4406 for the yields model and 0.5159 for the costs model. Consequently, given the lack of evidence of endogeneity in either model, each was estimated using ordinary least squares.

Results and Discussion

Adoption Decision Model

The adoption decision model results, based on 104 observations, are reported in table 3. As observed from this table, the only significant effects at the 5% level were farm size and use of a computer for finances. Consistent with expectations, both of these factors increase the probability of early adoption. However, the other farm operation and technology use variables were not found to be significant. For technology, the use measure of narrow row spacing did not affect speed of adoption. We believe this result likely stemmed from the fact that nearly all farmers in the sample had already adopted this technique, and therefore the variable Narrow was not as good an indicator of the use of new technologies as anticipated. Of the two remaining operations variables, soybean income and grain storage capability, the insignificance of the former at even the 10% level was the greater surprise. At the same time, however, it should be noted that the wide confidence intervals for the hazard ratios of many of the covariates suggest this study cannot rule out an effect of these or the remaining covariates on farmers' adoption process.