home *** CD-ROM | disk | FTP | other *** search
- Path: sparky!uunet!noc.near.net!hri.com!spool.mu.edu!uwm.edu!zaphod.mps.ohio-state.edu!howland.reston.ans.net!paladin.american.edu!auvm!UWPG02.BITNET!WSIMPSON
- From: WSIMPSON@UWPG02.BITNET (BILL SIMPSON)
- Newsgroups: bit.listserv.stat-l
- Subject: r for nonlinear regression
- Message-ID: <01GTS0KGXV2QAH4QWX@uwpg02.uwinnipeg.ca>
- Date: 21 Jan 93 16:50:20 GMT
- Sender: STATISTICAL CONSULTING <STAT-L@MCGILL1.BITNET>
- Lines: 27
- Comments: Gated by NETNEWS@AUVM.AMERICAN.EDU
- X-Envelope-to: stat-l@mcgill1.bitnet
- X-VMS-To: in%"stat-l@mcgill1.bitnet"
- MIME-version: 1.0
- Content-transfer-encoding: 7BIT
-
- I have been using Kaleidagraph on the Mac to do some nonlinear regression.
- This is real nonlinear regression with user-specified function (i.e. I am
- not talking about polynomial fit or log, exponential, or power fit done by
- doing ordinary linear regression on log x &/or y).
-
- The routine returns the estimated parameters for the function and chi-squared
- and r.
- I asked the company about these things. The chi-squared is
- sum((y-yhat)/s)^2 where s is SD of each point (only makes sense if you
- have repeated observations at each x level and you use the mean and SD).
- This formula for chi-squared is the same as SSE if sd=1 ie if unweighted least
- squares is used. Does anyone have a reference for this?
-
- I was wondering what an r could possibly mean for a nonlinear fit. They say
- they use:
- r=sqrt(1-(chi-squared/sum w(y-ybar)^2)), where w is the weight obtained from
- the sd of the observation. If unweighted, this is the same formula one
- might use for linear regression: sqrt(1-(sse/ssto)). I don't see how it makes
- any sense for nonlinear regression.
-
- This can't make sense because it measures degree of fit TO A LINE, not to the
- nonlinear regression function. (It seems to me.) The only r that would
- make sense would be a nonparametric r such as Spearman's.
-
- Any help much appreciated.
-
- Bill Simpson
-