Files
Abstract
"Straw Man" linear regression crop yield models using time as the single independent variable were created for corn and soybeans at the crop reporting district and State levels in Indiana, Illinois, and Iowa. Since yield trends may shift over time, several approaches to finding the "best fitting" model were evaluated, including simple one-line models, objectively selected 2- and 3-line segmented models, and models for which shifts in yield trends were systematically tested for 5- and 7-year groups of data. Comparisons of the models were based on relative percent differences of predicted 1979 yields from actual 1979 yields, and residual MSE values. Results show simple linear models had less fluctuations in relative predicted differences (1%-22% as compared to 0%-39%) than the 2- and 3-line models, but consistently higher residual MSE values. Simple linear models were the first choice in all corn yield models, while soybean yield models had no consistently superior model. A simple linear model was then selected as the single "Straw Man" model for all states and crops, and yield predictions for 1980 were generated.