home *** CD-ROM | disk | FTP | other *** search
- Newsgroups: bit.listserv.edstat-l
- Path: sparky!uunet!utcsri!torn!csd.unb.ca!UNBVM1.CSD.UNB.CA
- From: WRK0000 <WRK0@UNB.CA>
- Subject: ON QUARTILES
- Message-ID: <13DEC92.10550753.0036@UNBVM1.CSD.UNB.CA>
- Lines: 26
- Sender: usenet@UNB.CA
- Organization: The University of New Brunswick
- Date: Sun, 13 Dec 1992 13:46:09 GMT
-
- Yes, a computer statistics package must get the calculation for
- quartiles RIGHT.
-
- But there are other concerns when teaching:
-
- (1) Given encouragement, the number of formulae appearing in a
- statistics course wants to proliferate beyond reason.
- I'm glad to drop any I can, and this is a prime candidate.
-
- (2) My class is taking 5 other courses. My student has limited
- think time for my course, and there are MUCH more important
- things in a statistics course to think about.
-
- (3) One of those more important things to think about is this:
- The precision of the sample quartile is order 1/sqrt(N).
- so any K-th observation where K is within the span,
- N/4 plus-minus constant * sqrt(N),
- is a reasonable estimate. The assorted K's of the competing
- formulae as all within plus-minus 2 of each other.
-
- (I accept any reasonable quartiles on an exam and tell this to
- any student who inquires.)
-
- Bill Knight / University of New Brunswick / Canada
-
-
-