home *** CD-ROM | disk | FTP | other *** search
- Comments: Gated by NETNEWS@AUVM.AMERICAN.EDU
- Path: sparky!uunet!paladin.american.edu!auvm!MARLIN.JCU.EDU.AU!BHMJS
- Message-ID: <199211230803.AA00928@marlin.jcu.edu.au>
- Newsgroups: bit.listserv.stat-l
- Date: Mon, 23 Nov 1992 18:03:45 +1000
- Sender: STATISTICAL CONSULTING <STAT-L@MCGILL1.BITNET>
- From: Michael Smithson <bhmjs@MARLIN.JCU.EDU.AU>
- Subject: Meaning of transformations
- Lines: 45
-
- The recent discussion on the interpretability of transformations is
- pertinent, I think, to concerns shared by a number of us. I'd like
- to add my own penny's-worth, by way of claiming that under some
- conditions we may not have to suffer the cost in substantive inter-
- pretability to which Walter Davis refers.
-
- If you have monopolar raw scales (whether unbounded or not), then
- a fuzzy set interpretation for certain kinds of commonly used
- transformations is possible. The key idea is to think of the monopolar
- raw scale as if it's a fuzzy set with strong members denoted by high
- scores and weak or marginal members by low scores (e.g., annual
- income as an indicator of how strongly someone is a member of the
- set of really high income-earners).
-
- Here's a potted example (although it's much easier to get across with
- the help of graphs):
-
- A right-skewed distribution could be thought of as a very exclusive
- fuzzy set, in the sense that there are only a few strong members
- (these are the outliers up on the right-hand extreme of the scale).
- A right-skew correcting transformation compresses the right-
- hand (upper) end of the scale relative to the left-hand (lower) end,
- and in some cases extends the left-hand end somewhat as well.
- The effect is to bring the high-scoring outliers closer to the bulk of
- the scores further down the scale (i.e., closer to those who are not
- good members of the set). The transformed scale is therefore like a
- less exclusive set than its original counterpart. In fuzzy set terms,
- the transformation has "dilated" the set. Taking the log of income,
- for instance, is akin to rescaling everyone in terms of the extent
- to which they are members of the set of people who earn at least
- a 'moderately high' income.
-
- The analogous argument may be made to interpret left-skew correct-
- ing transformations as fuzzy set "concentrators", whose scales behave
- like more exclusive fuzzy sets than their originals.
-
- I also have fuzzy set interpretations of kurtosis-altering transformations,
- but I fear this posting is long enough already. If anyone is interested in
- further information or a reference for this approach, please feel free
- to get in touch with me.
-
- Mike Smithson
- Behavioural Sciences bhmjs@marlin.jcu.edu.au
- James Cook University
- Queensland 4811 Australia
-