Statistics and the Art of Model Construction
Abstract
Behavioral models should be based on reliable knowledge. Reliable knowledge is achieved when a scientific community reaches consensus on the interpretation of available evidence. When properly used and interpreted, statistical models can aid in the process of principled argument that leads to consensus. This report reviews the state of the art for the use of statistical methods in behavioral modeling. The major topics covered are the construction of measurement models to quantify behavioral constructs, the construction of path models to describe relationships between behavioral constructs, issues associated with model appraisal and amendment, and methods of searching for alternatives to or refinements of existing models. Examples of specific topics include methods of evaluating the underlying nature of constructs (i.e., continuous or categorical), methods of constructing models that combine constructs from different research domains (e.g., individual growth, individual differences, and group processes), the place of significance tests in model evaluation, and methods of searching for new insights regarding behavior. Potential military applications are illustrated. Each section concludes by considering how the issues and methods discussed contribute to the process of principled argument that is needed to ensure that behavioral models are based on reliable knowledge.
Document Details
- Document Type
- Technical Report
- Publication Date
- Nov 14, 2003
- Accession Number
- ADA433422
Entities
People
- Ross R. Vickers
Organizations
- Naval Health Research Center