Bounded-Influence Inference in Regression.
PRINCETON UNIV NJ DEPT OF STATISTICS
Pagination or Media Count:
Two new classes of tests for regression models, likelihood ratio type tests based on quadratic forms of robust estimators, are introduced. Both can be viewed as generalization of the classical F-test. By means of the influence function their robustness properties are investigated and optimally robust tests that maximize the asymptotic power within each class, under the side condition of a bounded influence function, are constructed. Finally, an example based on real data shows that these tests are valuable robust alternatives to the F-test. Author
- Statistics and Probability