[R] error when using multcomp and lm

Torsten Hothorn Torsten.Hothorn at stat.uni-muenchen.de
Fri Dec 18 17:06:52 CET 2009


On Fri, 18 Dec 2009, Achim Zeileis wrote:

> On Fri, 18 Dec 2009, chrischizinski wrote:
>
>> Just in case anyone ever has similar issues, it appears that 'glht' does 
>> not work without an intercept.
>
> That is not precise enough to be useful. I've had a look again and appears to 
> be the case that mcp() (or rather the internal function mcp2matrix) does not 
> work for models where there are multiple factors but no intercept.
>
> As the posting guide asks us to do, I have (1) provided a _reproducible_ 
> example below, and (2) included the maintainer of the package in the posting. 
> Torsten, it would be great if you could have a look.

Achim and Chris,

thanks for spotting this problem. The bug is now fixed in multcomp 
1.1-3 available from R-forge.

Torsten

>
> Best,
> Z
>
> ## package
> library("multcomp")
>
> ## various models with and without intercept
> m1a <- lm(breaks ~ tension, data = warpbreaks)
> m1b <- lm(breaks ~ 0 + tension, data = warpbreaks)
> m2a <- lm(breaks ~ wool + tension, data = warpbreaks)
> m2b <- lm(breaks ~ 0 + wool + tension, data = warpbreaks)
>
> ## these two are equivalent: one factor with/without intercept
> summary(glht(m1a, linfct = mcp(tension = "Tukey")))
> summary(glht(m1b, linfct = mcp(tension = "Tukey")))
>
> ## these two should be equivalent: two factors with/without intercept
> ## but the latter fails
> summary(glht(m2a, linfct = mcp(tension = "Tukey")))
> summary(glht(m2b, linfct = mcp(tension = "Tukey")))
>
>




More information about the R-help mailing list