A graphical aid for sensitivity analysis of estimated regression coefficients

A graphical aid for sensitivity analysis of estimated regression coefficients

0.00 Avg rating0 Votes
Article ID: iaor1988820
Country: United States
Volume: 35
Issue: 2
Start Page Number: 269
End Page Number: 284
Publication Date: Apr 1988
Journal: Naval Research Logistics
Authors: ,
Abstract:

A problem often arising in applied linear regression modeling is determining the appropriate weight to attach to each observation. In the extreme, the problem extends to the deletion of observations from a data base. Since the choice of weights, fitting criteria, and estimation procedure depend on the specific objectives of the modeling, universally applicable guidelines are virtually nonexistent. However, the sensitivity of analytic conclusions to weights assigned to ‘suspect’ observations can be conveniently assessed using a graphical display. This report develops such a display based on a modeling of outliers which leads naturally to estimators based on a weighted least-squares criterion, and a data-analytic method for determining how much downweighting to impose on a specific subset of observations. This technique is illustrated with several examples, including one relating air pollution to human mortality.

Reviews

Required fields are marked *. Your email address will not be published.