[Rd] Different results from 'lm' for different 32-bit Windows versions

Andreas Fromkorth @ndre@@@|romkorth @end|ng |rom @tegm@nn@y@tem@@com
Mon Nov 23 08:38:05 CET 2020


I recently stumbled upon the fact that different 32-bit R Windows versions produce slightly different results when calling 'lm'. More precisely, this seems to be caused by a change in the transition from R 3.2.x to R 3.3.0.

A minimal example, for reproducing this is "lm(y~x, data.frame(x=c(1,2,3), y=c(1,2,3)))".

In R 3.3.0 the result is
lm(formula = y ~ x, data = data.frame(x = c(1, 2, 3), y = c(1,
    2, 3)))

(Intercept)            x

0                     1

In R 3.2.4 Revised on gets the result
lm(formula = y ~ x, data = data.frame(x = c(1, 2, 3), y = c(1,
    2, 3)))

(Intercept)            x
-4.441e-16    1.000e+00

Both versions behave identically until the call C_Cdqrls in the fit.lm function. What causes this changed behavior?

Best regards,

Andreas Fromkorth.
Next PLA 3.0 Trainings (Business classes): www.bioassay.de/training <http://www.bioassay.de/training>
Dr. Andreas Fromkorth
Project Manager
andreas.fromkorth using stegmannsystems.com

Stegmann Systems GmbH, Raiffeisenstr. 2, 63110 Rodgau, Germany
Fon: +49 (6106) 77010-0, Direct: +49 (6106) 77010-213, Fax: +49 (6106) 77010-190
Handelsregister/Commercial Register Offenbach HRB 43033
Gesch?ftsf?hrer/CEO: Dr. Ralf Stegmann

Pick a product: APS 2.10<https://www.aps-info.de/?utm_source=signatur&utm_medium=email&utm_campaign=E-Mail-Signatur> | PLA 3.0<https://www.bioassay.de/?utm_source=signatur&utm_medium=email&utm_campaign=E-Mail-Signatur>
Follow us on LinkedIn<https://www.linkedin.com/company/stegmann-systems-gmbh>

	[[alternative HTML version deleted]]

More information about the R-devel mailing list