A Universal Complexity Criterion for Model Selection in Dynamic Models of Cooperative Work based on the DSM

Year: 2013
Editor: Scheurmann, E.; Maurer, M.; Schmidt, D.; Lindemann, U.
Author: Schlick, C.M.; Schneider, S.; Duckwitz, S.
Series: DSM
Institution: RWTH Aachen University, Germany
Section: New Approaches Complexity Management and Matrix Methods
Page(s): 99-105
ISBN: 978-3-446-43803-3

Abstract

This paper presents a complexity criterion for model selection in dynamic models of cooperative work in new product development. The complexity criterion is based on a task-oriented variant of the design structure matrix. Vector autoregression models of cooperative work are introduced and used to calculate a closed-form solution for a metric of emergent complexity. The metric was invented in basic research. Based on the complexity metric, a universal principle for model selection is formulated. The principle provides a natural safeguard against overfitting as it defines a method to reduce the part of the field data that looks like random performance fluctuations by using a more elaborate − but in the sense of Occam’s Razor not unnecessary complex − model. Finally, the results of two validation studies are presented. The results show that the complexity criterion is not only highly accurate for making model selection decisions in specific NPD environments, it appears to be an effective universal criterion for model selection in the class of vector autoregression models of arbitrary open systems.

Keywords: Project dynamics, cooperative work, complexity management