“Oscar Kempthorne”的意思、由来-开放百科全书

Oscar Kempthorne (January 31, 1919 – November 15, 2000) was a British statistician and geneticist known for his research on randomization-analysis and the design of experiments, which had wide influence on research in agriculture, genetics, and other areas of science.

Born in St Tudy, Cornwall and educated in England, Kempthorne moved to the United States, where he was for many decades a professor of statistics at Iowa State University.

Randomization analysis

Kempthorne developed a randomization-based approach to the statistical analysis of randomized experiments, which was expounded in pioneering textbooks and articles. Kempthorne's insistence on randomization followed the early writings of Ronald Fisher, especially on randomized experiments.^[1]

Kempthorne is the founder of the "Iowa school" of experimental design and analysis of variance.^[2] Kempthorne and many of his former doctoral students have often emphasized the use of the randomization distribution under the null hypothesis. Kempthorne was skeptical of "statistical models" (of populations), when such models are proposed by statisticians rather than created using objective randomization procedures.

Kempthorne's randomization-analysis has influenced the causal model of Donald Rubin; in turn, Rubin's randomization-based analysis and his work with Rosenbaum on propensity score matching influenced Kempthorne's analysis of covariance.^[3]

Model-based analysis

Oscar Kempthorne was skeptical towards (and often critical of) model-based inference, particularly two influential alternatives: Kempthorne was skeptical of, first, neo-Fisherian statistics, which is inspired by the later writings of Ronald A. Fisher and by the contemporary writings of David R. Cox and John Nelder; neo-Fisherian statistics emphasizes likelihood functions of parameters.^[4]

Second, Kempthorne was skeptical of Bayesian statistics, which use not only likelihoods but also probability distributions on parameters.^[5] Nonetheless, while subjective probability and Bayesian inference were viewed skeptically by Kempthorne, Bayesian experimental design was defended. In the preface to his second volume with Hinkelmann (2004), Kempthorne wrote,

Bibliography

Writings about Oscar Kempthorne

See also

Notes

1. ^Randomization-based inference and randomized experiments were introduced by Charles S. Peirce, whose writings were recognized by Kempthorne (in later years). Kempthorne praised Peirce repeatedly in Hinkelmann and Kempthorne (2008), for example as being "the foremost philosopher of science" (page 8 in the first edition, 1994). Answering the question "What do you see as the important problems of statistics?", Kempthorne replied finally, "I feel that much of philosophy needs to be read—particularly the writings of the great American philosopher, C. S. Peirce." (Folks, p. 336)
2. ^The "Iowa school" is named on page 262 in Bailey and Speed* {{cite journal|doi=10.2307/1403405|title=Factorial Dispersion Models|author=T. P. Speed and R. A. Bailey|journal=International Statistical Review / Revue Internationale de Statistique|volume=55|issue=3|date=December 1987|pages=261–277|jstor=1403405}}
3. ^See Chapter 8 in Hinkelmann and Kempthorne 2008.
4. ^Kempthorne often distinguished between the randomization-based analysis of early Fisher and the model-based analysis of (post-Neyman) Fisher, for example in Kempthorne's comments on Debabrata Basu's paper "The Fisher randomization test" in the Journal of the American Statistical Association (1978).
5. ^However, Kempthorne recognized that the planning of experiments used scientific knowledge and beliefs, and therefore Kempthorne was interested in optimal designs, especially Bayesian experimental design:

The optimal design is dependent upon the unknown theta, and there is no choice but to invoke prior information about theta in choosing the design. I would like to say there has never been the slightest argument about this. In the design of experiments, one has to use some informal prior knowledge. (Folks, 334)

Kempthorne's skepticism towards Bayesian inference focused on the prior's use in analyzing data from randomized experiments; for analyzing data from randomized experiments, Kempthorne advocated using the objective randomization-distribution, which was induced by the randomization specified in the experimental protoocol and implemented in the actual experimental plan.

External links