Mass Imputation of Agricultural Economic Data Missing by Design A Simulation Study of Two Regression Based Techniques

The demand for information concerning all facets of the production of agricultural commodities is constantly increasing. This demand is placing significant burden on the relatively few large and mid-sized producers that account for a disproportionally large percentage of all agricultural production. In particular, the demand for economic data from farm operations is especially intrusive of the producer in terms of time and sensitivity. One way to reduce this burden is to simply collect significantly less data and then try to regain some of the lost precision by modeling the now unobserved data using data that is observed. This paper evaluates some of the characteristics of data sets that are incomplete by design and are “completed” via imputations obtained from two regression based imputation methods. Estimates of population means and correlations are evaluated for a set of 27 economic variables that have a fixed missing rate of 60 percent. Some standard error estimates are obtained for one of the methods and these are evaluated as well. Gains in RMSE can be made for many variables and correlations can be preserved for many pairs of incompletely observed variables.