Title: One Line or Two? Perspectives on Piecewise Regression
Ewing, Robert - IOWA STATE UNIVERSITY
Submitted to: ASA-CSSA-SSSA Annual Meeting Abstracts
Publication Type: Abstract Only
Publication Acceptance Date: November 16, 2006
Publication Date: November 16, 2006
Citation: Ewing, R.P., Meek, D.W. 2006. One Line or Two? Perspectives on Piecewise Regression [CD-ROM]. In: ASA-CSSA-SSSA Annual Meeting Abstracts, November 12-16, 2006, Indianapolis, IN.
Sometimes we are faced with data that could reasonably be represented either as a single line, or as two or more line segments. How do we identify the best breakpoint(s), and decide how many segments are “really” present? Most of us were taught to distrust piecewise regression because it can be easily abused; a defensible decision strategy is therefore needed. The decision process starts by finding the breakpoint. Interestingly, the most obvious approach – fitting two independent lines – generally gives the worst breakpoint estimate of all the methods tested. Much better are finding the minimum sum of squares for two non-independent lines, and hidden Markov methods. Meanwhile, the most appropriate method for deciding between one or two lines depends on your expectations and understanding of the data. For example, an unexpected break requires more justification than an expected one, and some decision criteria (e.g., the Akaike Information Criterion) are less strict than others (e.g., the Bayesian Information Criterion).