Minitab is the program that we have used here for years on
recommendation of our statisticians. It is excellent, provides what has
been discussed, easy access to residuals for plotting to visualize their
distribution and the Durbin-Watson which parametrizes the 'randomness' of
those residuals, among other features.
It used to be available in a student edition form perhaps from Wiley or
maybe McGraw-Hill for ~$50 that I use extensively and require in p. chem.
The student edition doesn't do matrices but the mainframe Unix version
that we have here does. The only thing it doesn't do is non-linear
regression .. I wish that Mathcad or Sigmaplot had the statistical utility
that Minitab does.
Of course with something so useful, our computer center is hepped up to
get rid of it. Go figure.
For scientific data sets R^2 is all but useless, and as with all
statistics that attempt to parameterize a variety of behaviors with
single numbers, you can get into trouble if you just buy the number
without examining the underlying behavior/distribution. To emphasize
this, there is quite a nice article in American Statistician, 1973, 27,
17-21 "Graphs in Statistical Analysis" that makes the basis of an excellent
illustrative exercise. Ken Loach of SUNY Plattsburg, Chemistry put me on
to this.
John Cooper, Chemistry, Bucknell University, Lewisburg PA 17837-2005
jcooper@bucknell.edu 717-524-3673