Calibration of predictions in regression
UNSPECIFIED. (2000) Calibration of predictions in regression. COMMUNICATIONS IN STATISTICS-THEORY AND METHODS, 29 (9-10). pp. 1973-1986. ISSN 0361-0926Full text not available from this repository.
A regression predictor is well-calibrated if the predictions it gives are equal to the average responses that would be observed in an independent sample. The usual least squares predictor does not have this property, but its calibration can be improved by shrinking the predictions by a factor which depends on the signal-to-noise ratio of the regression model. We suggest a semi-Bayesian approach to estimating this factor, giving an estimate closely related to the so-called Stein Shrinkage Factor. The results are illustrated on a large medical data set.
|Item Type:||Journal Article|
|Subjects:||Q Science > QA Mathematics|
|Journal or Publication Title:||COMMUNICATIONS IN STATISTICS-THEORY AND METHODS|
|Publisher:||MARCEL DEKKER INC|
|Number of Pages:||14|
|Page Range:||pp. 1973-1986|
Actions (login required)