In article <[EMAIL PROTECTED]>, David Heiser <[EMAIL PROTECTED]> wrote:
>"Stan Brown" <[EMAIL PROTECTED]> wrote in message >news:[EMAIL PROTECTED] >... >... >... > Generally they want to see the role of >> traditional NHST reduced to a greater or lesser extent, in favor of >> confidence intervals, Bayesian inference, and/or meta-analysis. >> Whether you agree with any particular author or not, the interplay >> is likely to get you thinking. It will certainly make a difference >> in how I teach my introductory stats courses this fall. >> Many thanks to Louis and Rich for the suggestion! >I'm going to have to read that. Sounds interesting. >I think that whether to or not to teach hypothesis testing depends on who >makes up the class. If there is any body in the class who is going into >psychology, education, social scinces, biogeneics, biology, medicine or who >will have to write a research paper as a requirement in some future class, >then you need to teach hypothesis testing. There is hypothesis testing, and there is what is now being done. What is actually being tested is rarely quite what is claimed to be tested anyhow, and the point null hypothesis itself is essentially never true; so why should anyone care what happens to be the probability that it is rejected if it happens to be true? The real problem is whether one should act as if it is true, or do something else. If it is close enough to being true, it may well pay to treat it as so; if one does not make such slightly wrong simplifying assumptions, it is rarely possible to do anything. How close is "close" is not a statistical question. One has to balance the loss for improper rejection with the loss for improper acceptance. The rational Bayesian approach is that these are linear combinations (integrals) of the loss as a function of the state of nature with the "prior measure", which is the weight forced by rationality. >If the class is engineers, skip hypothesis testing and focus on confidence >intervals. BTW, confidence intervals are also not justified. One needs to balance the risk of the parameter not being in the interval against the cost of the size of the set. >If your text is Moore and McCabe, or some other text that has several >chapters devoted to significance testing, you may have to include hypothesis >testing. It is hard to overcome religious indoctrination. >David Heiser PDE (plain dumb engineer) -- This address is for information only. I do not claim that these views are those of the Statistics Department or of Purdue University. Herman Rubin, Department of Statistics, Purdue University [EMAIL PROTECTED] Phone: (765)494-6054 FAX: (765)494-0558 . . ================================================================= Instructions for joining and leaving this list, remarks about the problem of INAPPROPRIATE MESSAGES, and archives are available at: . http://jse.stat.ncsu.edu/ . =================================================================
