Scientific Program > Topic 3 > Session 3B >
Presentation 3B1. Interpretation of Regression Output: Diagnostics, Graphs and the Bottom Line

Presenter
Wesley O. Johnson (USA) johnson@wald.ucdavis.edu

 

Presentation Abstract

A standard approach in presenting the results of a statistical analysis of regression data in scientific journals is to mainly focus on the question of statistical significance of regression coefficients and factors. The reporting of p-values in conjunction with a description of the various positive and negative associations between the response the variables in question ensues. The bottom line tends to be, "yes the treatment works" or "there was no statistically significant association found between the response and a particular variable," when the real questions of interest beyond these initial assessments ought to be, "how well does the treatment work," and "was the experiment powerful enough to detect an important relationship that was not detected here?" The point of view taken in this talk will be that such a presentation of results, while important, constitutes only a first order approximation to a complete analysis and interpretation of data. Emphasis will be given in the talk on the interpretation of the model on the scale of the observable quantities in the data, rather than on the generally scaleless dimension of regression coefficients. The bottom line to be discussed is the interpretation of the data in a way that quantifies the, perhaps complicated, effect of variates on the mean or median response. This will be accomplished mainly by a simple graphical representation of the effects. While the approach presented is not novel, it is nonetheless rare in my experience to see statistical analyses in scientific journals or presentations that take this extra step in presentation. In addition, emphasis will also be given in the talk on the diagnostic assessment of the selected model, and its compatibility with the data. The diagnostic assessment, when possible, will again considered from the point of view that the ultimate goals of inference are best expressed on the scale of observable quantities in the data.

 

Manuscript
Download in Adobe Acrobat format (107 Kb).

 

Back to top

ICOTS-6, The Sixth International Conference on Teaching Statistics - International Program Committee (IPC) Website.
Copyright © 2001 by the IASE. All rights reserved. This information is subject to change without notice. This page was last modified on July 2, 2002.
For questions or comments, contact the Webmaster, Dagan Ben-Zvi.