[R] Standardized beta-coefficients in regression

[Frank Harrell]

It is true that our lives are complex but we have to stick to principles. 
Plus, statisticians have the lowest unemployment rate in all of the sciences
(1% according to NSF from a poll taken 2 years ago) so we should be able to
capitalize on that by sticking to well-founded beliefs and facts and
choosing positions were respect for our expertise is a given.

Standardized coefficients cloud rather than clarify, and they only apply to
the trivial case where everything is linear. See
http://biostat.mc.vanderbilt.edu/ManuscriptChecklist for more information.
Frank

Jeroen Ooms wrote:
> 
> Unfortunately I found myself in the same position as outlined above, where
> I was requested to reproduce 'standardized regression coefficients' as
> reported by SPSS. Below an example that produces something very similar to
> the results table from an SPSS "Linear Regression" procedure, including
> the standardized regression coefficients:
> 
> mylm <- lm(Sepal.Width ~ ., data=iris, x=TRUE, y=TRUE)
> sd.x <- sd(mylm$x);
> sd.y <- sd(mylm$y);
> std.coef <- coef(mylm) * (sd.x / sd.y);
> coef.table <- as.data.frame(summary(mylm)$coefficients);
> coef.table <- cbind(coef.table, std.coef);
> print(coef.table);
> 
> I do agree with B.R. but unfortunately the life of an applied statistician
> is complex sometimes :-)
> 


-----
Frank Harrell
Department of Biostatistics, Vanderbilt University
--
View this message in context: http://r.789695.n4.nabble.com/Standardized-beta-coefficients-in-regression-tp791616p3843763.html
Sent from the R help mailing list archive at Nabble.com.