home *** CD-ROM | disk | FTP | other *** search
- Path: sparky!uunet!elroy.jpl.nasa.gov!sdd.hp.com!cs.utexas.edu!ut-emx!ccwf.cc.utexas.edu!rcole
- From: rcole@ccwf.cc.utexas.edu (Richard W. Cole)
- Newsgroups: sci.math.stat
- Subject: RE: Testing for Linearity
- Message-ID: <85461@ut-emx.uucp>
- Date: 16 Dec 92 20:16:33 GMT
- Sender: news@ut-emx.uucp
- Reply-To: rcole@ccwf.cc.utexas.edu (Richard W. Cole)
- Organization: The University of Texas at Austin, Austin TX
- Lines: 39
- Originator: rcole@sylvester.cc.utexas.edu
-
- >>Why don't you try calculating the correlation coefficient: the closer it
- >>is to 1 or -1, the more linear your data.
-
- >I do not think that the corr. coef. would tell much at all right away.
- >If he is asking about linearity I'm sure he has in mind fitting a straight
- >line.
- >He says:
-
- >>" I have a question regarding, linearity
- >>of data points. Does anybody know of any statistical
- >>test for linearity. I would appreciate any kind of
- >>help, references etc."
-
- >I would think that PLOTTING the data first would be necessary in order
- >to asses the rest of the test:
-
- >ie:if there is a pattern other than some kind of random scattering along some
- >invisible straight line (ie: curvilinear like a parabola) then he should fit
- >a polynomial model (possibly of second order), then the reduced model with
- >only the linear term and THEN asses linearity by means of the corresponding
- >tests.
- I agree. Actually testing the correlation is already assuming that
- linearity exists (a constant change in y per unit change in x) and the
- correlation test is a hypothesis that the slope (constant change in y)
- is equal to zero. Working with a model not assuming linearity (one with
- linear and quadratic and cubic components) and then testing a linearity
- restriction would seem much more in order since linearity technically
- can exist when the slope is equal to zero. Once you've then assumed
- linearity you may want to test that the slope is zero (where the
- restricted model has only an intercept term as a predictor). Accepting
- this model might suggest that the functional form of y is more complex
- than fitting any of the models thus far or that their is no relationship
- at all.
-
- ========================================================
- Richard Cole, Stat/Math Services Learning to
- Computation Center UT Austin TX, 78712 Paint by
- rcole@ccwf.cc.utexas.edu Numbers
- ========================================================
-