Online Program

Return to main conference page
Thursday, May 30
Data Science Techologies
Practice and Applications
Data Science Applications E-Posters, I
Thu, May 30, 3:00 PM - 4:00 PM
Grand Ballroom Foyer

A Maximum Likelihood Method for Correlated Discrete and Continuous Outcomes with Selection, Lagged Effects and Variance (306343)

Satheesh Aradhyula, University of Arizona 
*Rhoda Nandai Muse, University of Arizona, Mathematics Department 

Keywords: Selection, Lagged effects, joint discrete, heteroscedastic variance and continous outcomes

Agricultural technology adoption and crop yields remain low in Sub-Saharan Africa. Farmers are risk averse but quantifying this effect on technology adoption has been difficult. Empirical studies on risk have used ad hoc two-step methods for capturing the first two moments of yield. These estimated moments are then subsequently used as regressors in the adoption equation as proxies for risk. These two-step procedures for estimating the role of risk on technology adoption decisions results in biased and inconsistent estimates. Improved seed use (discrete variable) affects yield (continuous variable) and is correlated with the yield error term resulting in an endogenous or selection term. We use maximum likelihood method to simultaneously estimate crop yields and technology adoption while accounting for this endogeneity. We also allow for the variance of crop yields to vary at the household level. Expected yields and yield variance, lagged once, are used as regressors in the adoption equation to explicitly account for the risk aversion. This is the first study to jointly account for endogeneity, heteroscedasticity of crop yields, and the use of variance of crop yields as regressors. We use longitudinal household survey data from Kenya for three years. We found that higher past season yield significantly increased the likelihood of improved seed adoption in the next season. When the selection/endogenous term is not included, we found that the sign of the covariance between the two equations changed from negative to positive and finally adoption of improved seed is negatively affected by yield variance.