Reply to Boomsma and Molenaar
暂无分享,去创建一个
Boomsma (B) and Molenaar (M) raise several useful issues , for which we thank them. But they do not offer alternatives. It is as true in scientific discussion as in sports: You can't beat sump'n with nuttin'. Merely asking worried questions is not a sound reason to reject a method for which there is some strong support. Example: B and M criticize the experiments by Shevokas and At kinson on the grounds of possible bias. But they give no evidence of this. Atkinson's and Shevokas's work certainly might have been b iased , t hough both repor ted being skeptical about resampling before starting the work. But in the absence of any contrary evidence, there would seem a presumption that the exis t ing f indings are meaningful even if not conclusive. The mere possibility of a con founding factor—which almost never can be ruled out—is hardly reason to dismiss those studies out of hand. More fundamentally, B and M do not present or cite any other re search on resampling. This is what is needed—studies wi th other designs and different students. We will be happy to cooperate with them or with anyone else in con ducting such studies. We appreciate the nice things tha t B and M say about our RESAMPLING STATS program. As to their cr i t icisms, we lack space to comment on many of them. Most important, however, is that our original art icle is not mainly about the program; rather, it is about resampling as a general method, whether it be done with our program or with another. And we greatly hope that B's and M's concentrat ion on our computer p rogram does not d is t ract the reader's attention from our central point: Resampling is more likely to produce a sound answer for the researcher and student than is the conventional formulaic method. B and M say that the BASIC pro gram we presented for comparison could be greatly shortened. But they do not say how. Nor do they show a program that would rebut our c la im that RESAMPLING STATS is much clearer. They also criticize our selection of a pro gram because it supposedly con tains a "false" suggestion. In fact, we selected the very first program in the Quantitative Literacy Series book on s imula t ion by Gnanadesikan, Scheaffer, and Swift. We hoped that choosing a published program, the first in this standard book, would be the best way to avoid the charge that we had set up a straw man. But this, too, is a matter that deserves investigation and not just vague criticism. We have, there fore, recent ly suggested to the editor of a technical journal (who
[1] Julian L. Simon,et al. Probability and Statistics: Experimental Results of a Radically Different Teaching Method. , 1976 .
[2] Donald Fraser,et al. Randomization Tests for a Multivariate Two-Sample Problem , 1958 .
[3] Julian L. Simon,et al. Basic research methods in social science , 1985 .
[4] M. Dwass. Modified Randomization Tests for Nonparametric Hypotheses , 1957 .
[5] B. Efron. The jackknife, the bootstrap, and other resampling plans , 1987 .