home *** CD-ROM | disk | FTP | other *** search
- Comments: Gated by NETNEWS@AUVM.AMERICAN.EDU
- Path: sparky!uunet!paladin.american.edu!auvm!UNCVM1.BITNET!WDAVIS
- Message-ID: <STAT-L%92111917080847@VM1.MCGILL.CA>
- Newsgroups: bit.listserv.stat-l
- Date: Thu, 19 Nov 1992 16:49:24 EST
- Sender: STATISTICAL CONSULTING <STAT-L@MCGILL1.BITNET>
- From: Walter Davis <WDAVIS@UNCVM1.BITNET>
- Subject: Re: meaning of transformations
- In-Reply-To: Message of Wed, 18 Nov 1992 13:47:51 PST from <jwa@ANS.ORST.EDU>
- Lines: 63
-
- On Wed, 18 Nov 1992 13:47:51 PST Jerry Arnold said:
- >Michael Granaas writes:
- >
- >The following is an excerpt from 'Introduction to Quantitative
- >Genetics' chapter 17 by D. S. Falconer. I don't know if this exactly
- >addresses your question, but I think it's relevant to the discussion.
- >
- > It is tempting to suppose that each character has its 'natural'
- > scale, the scale on which the biological process expressed in the
- > character works. Thus, growth is a geometrical rather than an
- > arithmetical process, and a geometric scale would appear to be the
- > most 'natural'. For example, an increase of 1g in a mouse weighing
- > 20g has not the same biological significance as an increase of 1g
- > in a mouse weighing 2g: but an increase of 10% has approximately
- > the same significance in both. For this reason a transformation to
- > logarithms would seem appropriate for measurements of weight.
- > This, however, is largely a subjective judgement, and some
- > objective criterion for the choice of a scale is needed. Different
- > criteria, however, are often inconsistent in the scale they
- > indicate and, moreover, the same criterion applied to the same
- > character may indicate different scales in different populations.
- > Therefore the idea that every character must have its 'natural'
- > and correct scale is largely illusory.
- >
- >He goes on to discuss three reasons for using a transformation of
- >scale 1) make the distrib normal 2) make the variance independent of
- >the mean and 3) reduce non-additive interactions.
- >
- >
- >--
- >
- >Jerry Arnold Internet: jwa@ans.orst.edu
- >Dept of Animal Sciences Voicenet: 503-737-5043
- >Oregon State University
- >Corvallis, OR 97331
-
- hi all,
-
- I agree that variables do not have natural or 'true' scales. This
- problem comes up frequently when dealing with latent variable models
- when you introduce 'scaling' to students. The students tend to
- be quite uncomfortable at first about scaling the latent variable
- to the units of one of its variables.
-
- However, some scales (for some variables) have achieved a
- level of social consensus (degrees, centimeters, percents, etc.).
- Therefore, when interpreting results, it is much easier to explain
- how a model works, what it means, etc. when you are able to stay
- relatively close to the original units.
-
- On the other hand, specifying the correct functional form is
- vitally important. I want to be clear that I am not recommending
- that anyone estimate a poorly fitting model just because it's
- 'non-transformed'. I am trying to suggest that transforming
- variables can have tremendous costs in terms of substantive
- interpretability.
-
- hope this helps,
-
- Walter Davis <WDAVIS@UNCVM1>
- Department of Sociology
- Institute for Research in Social Science
- UNC - Chapel Hill
-