Econometric Evaluation of New Technology with an Application to Integrated Pest Management: Discussion

Abstract

To analyze technical change in agriculture, economists have had to select from a limited menu of research approaches. The first course or approach, introduced by Griliches, is to treat technical change as the adoption of a particular improved input, such as a crop variety or piece of equipment. Diffusion of the innovation among profit-maximizing firms is modeled as a contagious, stochastic process with a resulting S-shaped curve time. This is primarily a micro model, but often it is applied to secondary data. The second major approach is to examine factor-augmenting technical change by means of production functions or, more recently, with dual cost functions (Binswanger). In these more aggregate models, technical change is merely represented by a time trend variable. There are three other entrees on the research menu: (a) treatment of joint adoption of improved inputs by simultaneous equation models such as adoption of high-yielding grains and fertilizer, (b) explicit inclusion of both price and yield risk with the diversification offered by technology mixes, and (c) models incorporating fixed factors to depict the adjustment cost of adoption. The two studies offered to us today are likely to expand the research menu. The Hall and Duncan (HD) study extends and continues the work of labor economists by applying a qualitative choice model to pest consultant hiring decisions. Zilberman offers a tasty dish that has a novel blend of micro realism and input market rigor. The topping is the potential for aggregating the micro decisions to evaluate the effects of government commodity policies on natural resource utilization. His effective-water could be effective pesticides or any other scarce resource. I will direct my comments to the HD paper. The primary foods-for-thought that HD serve (in about 100 equations) are alternatives to Heckman's self-selection bias model. HD's estimator corrects for both self-selection and sample-proportion bias. Incidently, there have been several modifications offered to the Heckman model, yet its use seems to be widespread because it is easily understood and can be modified to the special problems at hand. Both the self-selection and choice-based biases can occur frequently; the latter problem is often encountered because it usually is less expensive to use observable choices as a basis for sampling. The analysis of the discrete technology choices of individual farmers several years with better econometric models is a research area that is underexploited. HD are to be commended for extending the work of Rodriquez and Burrows, who apply Heckman's model to pest management choices. Unfortunately, the HD study is warmed over in the sense that it uses the IPM choices of about fifty California farmers that have been used for five or six other studies. The data are decisions made in the early 1970s, but now there are new technologies, new pest species in the area, and different relative prices. The ideal farmer adoption sample would be different from that used in the HD study. There is a need for observations on many farmers choosing for several years among the newest technical developments. Multiple years and locations are needed to give variation in relative prices, pest levels, and weather conditions. Fixed factors such as soil fertility and managerial ability should be collected as well as prices of variable inputs, product prices, yields, and any direct adoption costs. Such a survey compiled for several years is very expensive to collect; one can see why good data are reheated.