Abstract

The yield of products in the dry milling industry is largely determined by the physical properties of the corn kernel. The main objective of this paper is to investigate several statistical models of dry milling yield prediction based on physical characteristics of corn. Data consisting of one hundred corn samples representing a range of genetic traits and quality differences are used. For each corn sample, 16 physical and chemical properties plus six dry milling product yields were measured in a controlled laboratory environment.

For each corn sample, we consider a vector of dry milling product yields and a vector of physical corn characteristics. Several single product models are investigated, two of which implicitly take into account the simplex sample space of product yields. A multivariate model is considered that consists of mapping the sample space from a simplex to unrestricted Euclidean space. Comparison are performed using a jack-knife-like approach.

Keywords: Dry milling, quality characteristics, yield prediction, production function, linear models, compositional data, Cobb-Douglas, translog, continuation ratios, jackknife, multivariate analysis.