Evolution of a Change-Point Model Solution

Alix I. Gitelman

Using data collected at several sites to study seed germination in Downy Brome (a weed in the grass family), several models are fit to provide answers to the following questions: (1) At what value of {\em growing degree days (GDD)}, a temperature/time metric, does seed germination become non-zero? (2) After germination is non-zero, what is the relationship between GDD and germination? A first stab at answering these questions is a manual, two-step approach in which the first question is addressed, and under its solution the data are reduced to include only non-zero germination samples, wherein the second question is addressed. A second approach automates these two-steps using a {\em change-point model}---the GDD value at which germination becomes non-zero is estimated, and a linear model is fit to the non-zero data. From there, more complex (more realistic?) Bayesian hierarchical change-point models are considered to account for differences across sites. Data and motivational questions for this exploration were provided by Dan Ball, Associate Professor of Weed Science with the Columbia Basin Agricultural Research Center in Pendleton, Oregon.