Subject: [isostat] Resampling and/or permutation test packages. |

From: "Whittinghill, Dexter C." |

Date: Thu, 18 May 2006 11:51:34 -0400 |

To: isostat@oberlin.edu |

Folks, Here is what I asked on 4/10 (with some minor corrections!): Is there a nice, general, resampling and/or permutation test package out there? Of course I have heard of "Resampling Stats". I was wondering if anyone out there has purchased or used one extensively AND would recommend it? And of course, if it is easy to do in R or S, you folks can tell me that, too. [I am thinking of getting it for me (to use with my engineering grant buddies).] Here is what I heard, permuted and grouped so you can see a thread of evolution (just for fun). I give it to you anonymously, but a couple of people give themselves away, and one other you can probably guess. 1. I have used Resampling Stat in courses and really like it. The students quickly catch on to using it. It is easy to learn, adaptable but specialized. It is not expensive. 2. I have used the "Resampling Stats" about 10 years ago. I believe most in the software can be accomplished more easily with R or S+. 3. See www.insightful.com/Hesterberg/bootstrap for info about the S+resample library, and material for teaching intro stats using resampling. I don't recommend Resampling Stats. Pretty limited, e.g. when I looked at it you couldn't even do scatterplots. 4. Of Course it is easy to do in R/Splus :-) 5. I've got a student doing Bootstrap CI's for a senior project. He/we(?) are using Fathom to understand the basics and R for "production" work (e.g. running hundreds of samples and checking coverage of the CI). There is a "boot" package for R that has lots of nice features - but it is operated more like a program language, so less suitable for general students. 6. Both Stata and SPlus have nice bootstrapping capabilities. 7. I've been happily resampling in R or Stata without incident. The "bootstrap" and "permute" programs within Stata are quite flexible. 8. There is a "bootstrap" command in Stata. It is general in the sense that you can use it with many different regression or modeling commands. 9. I have been playing around a bit with R this spring, but I haven't located the "easy to use" permutation/randomization test procedure I was hoping to find. If you become aware of one, please let me know. 10. A team headed by Tim Hesterberg (mainly) and I (as exposition expert) wrote a student-level (we hope) introduction to the bootstrap and permutation tests as a supplement to Moore & McCabe. A PDF is available free online at www.whfreeman.com/ips and click on Supplement Chapters. Or you can order hardcopies. Relevant here is that Tim prepared an Splus library with special menus designed to make these procedures easy for students to use. The library will load all the data sets from Moore & McCabe if you choose that as well. Go to www.whfreeman.com/resample for links. To use the library, you must of course use S-plus for PCs. Note that student S-plus is free (link again on the site above). You wealthy faculty (or your institution) will have to pay the academic fee for regular S-plus. Details at the site above. Two plugs: 1. I think this exposition is the most interesting thing I have recently been involved with. I wanted to make it part of the core text -- if folk like you start teaching resampling methods to undergrads (we should), maybe it will make the next edition. 2. S-plus is my favorite stat software. It (and the free version R) are staples of research in many evolving fields of statistics. It's worth learning. I think I will have to go back to S/S+ via R. Enjoy. Dex ************************************* Dexter C. Whittinghill III Department of Mathematics Rowan University 201 Mullica Hill Rd. Glassboro, NJ 08028 e-mail: whittinghill@rowan.edu Phone: 856-256-4500 x3879 FAX: 856-256-4816 We are using Outlook 2003. *************************************