Title: Regression Analysis Applied to Happenstance Data
Author: George E.P. Box
Publication Source: Report No. 149, October 1996, Center for Quality and Productivity Improvement, University of Wisconsin - Madison
Note: Click on the "Download r149" button on the page linked above for a free copy of this paper
Abstract
[Based on the author's abstract.] Care is needed in interpreting results when regression analysis is applied to happenstance data. Although a fitted regression model may be useful for prediction of future values of a series, it may be totally misleading for explaining causation relationships between the variables. Furthermore, such analysis can be disastrously affected by lack of independence between residual errors.
To download Adobe Acrobat (for free) click here.