svn+ssh://tlinnet@.../svn/relax/trunk
........
r27155 | tlinnet | 2015-01-07 20:25:09 +0100 (Wed, 07 Jan 2015) | 15 lines
Implemented ordinary_least_squares function the repeated auto-analysis.
Inspection of statistics books, shows that several authors does not recommend using regression through the origin (RTO).
From Joseph G. Eisenhauer: Regression through the Origin
- RTO residuals will usually have a nonzero mean, because forcing the regression line through the origin is generally inconsistent with the best fit.
- R Square measures (for RTO) the proportion of the variability in the dependent variable "about the origin" explained by regression.
This CANNOT be compared to R Square for models which include an intercept.
From "EXPERIMENTAL DESIGN AND DATA ANALYSIS FOR BIOLOGISTS", G. P. QUINN, MICHAEL J. KEOUGH
- minimum observed xi rarely extends to zero, and forcing our regression line through the origin not only involves extrapolating the regression line
outside our data range but also assuming the relationship is linear outside this range (Cade & Terrell 1997, Neter et al. 1996).
- We recommend that it is better to have a model that fits the observed data well than one that goes through the origin but provides a worse fit to the observed data.
- residuals from the no-intercept model no longer sum to zero.
- usual partition of SSTotal into SSRegression and SSResidual doesn’t work.
........
