Accounting for Neighborhood Influence in Estimating Factors Determining the Adoption of Improved Agricultural Technologies

Researchers have traditionally applied censored regression models to estimate factors influencing farmers' decisions to adopt improved technologies for the design of appropriate intervention strategies. The standard Tobit model, commonly used, assumes spatial homogeneity implicitly but the potential for the presence of spatial heterogeneity (spatial autocorrelation or dependence) is high due to neighborhood influence among farmers. Ignoring spatial autocorrelation (if it exists) would result in biased estimates and all inferences based on the model will be incorrect. On the other hand, if spatial dependence is ignored the regression estimates would be inefficient and inferences based on t and F statistics misleading. To account for neighborhood influence, this study applied a spatial Tobit model to assess the factors determining the adoption of improved maize varieties in southern Africa using data collected from 300 randomly selected farm households in the Manica, Sussundenga and Chokwe districts of Mozambique during the 2003/04 crop season. Model diagnosis confirmed the spatial Tobit model as a better fit than the standard Tobit model. The estimated results suggest that farm size, access to credit, yield and cost of seed significantly influence maize variety adoption at less than 1% error probability while age of household head and distance to market influence adoption decisions at 5% error probability. The marginal effect analysis showed that convincing farmers that a given improved maize variety would give a unit more yield than the local one would increase adoption rate by 18% and intensity of use by 10%. Given that improved maize seeds are relatively more expensive than local ones, making credit accessible to farmers would increase adoption and intensity of use of improved maize varieties by 24% (15% being the probability of adoption and 8% the intensity of 2 use of the varieties). On the other hand, increasing seed price by a unit over the local variety would decrease the adoption rate by 12% and area under the improved variety by 6%. Targeting younger farmers with extension messages or making markets accessible to farmers would marginally increase the adoption and use intensity of improved maize varieties by only 0.4%. These results suggest that increasing field demonstrations to show farmers the yield advantage of improved varieties over local ones in Mozambique are essential in improving the uptake of improved varieties, which may be enhanced by making credit available to farmers to address the high improved seed costs. Alternatively, assuring farmers of competitive output markets through marketing innovations would enhance improved maize variety adoptions decisions. It may be concluded that the significance of the paper is its demonstration of the need to include spatial dependency in technology adoption models where neighborhood influences are suspected. Such an approach would give more credence to the results and limit the errors in suggesting areas to emphasize in individual or group targeting. The results thus have implications beyond the study area. Furthermore, the paper contributes to the scanty literature on the application of spatial econometrics in agricultural technology adoption modeling.