A survey of some new alternatives to Wald's variance component test

Rok, strany: 1999, 91 - 102

Perhaps the best known and most often applied test of whether a variance component differs from zero is Wald's test as generalized by Seely and El-Bassiouni [J. F. Seely, Y. El-Bassiouni: Applying Wald's variance component test, Ann. Statist. 11 (1983), 197–201]. In this survey, we propose the use of near replicate lack of fit tests for testing variance components. These are exact $F$ tests and we present examples that demonstrate substantial increases in power using these tests rather than Wald's test. Moreover, we present a rationale, based only on the design matrix, for identifying situations in which these tests should work better than Wald's test. The techniques are quite general and leave open the possibility of finding other exact $F$ tests that can out perform Wald's test in specific situations. Nonetheless, our experience is that Wald's test performs well over a wide array of models.

ISO 690:

Christensen, R., Bedrick, E. 1999. A survey of some new alternatives to Wald's variance component test. In Tatra Mountains Mathematical Publications, vol. 17, no.3, pp. 91-102. 1210-3195.

APA:

Christensen, R., Bedrick, E. (1999). A survey of some new alternatives to Wald's variance component test. Tatra Mountains Mathematical Publications, 17(3), 91-102. 1210-3195.