# [R] contrasts in lm and lme

Peter Dalgaard BSA p.dalgaard at biostat.ku.dk
Fri Jun 15 21:27:11 CEST 2001

```"Peter B. Mandeville" <mandevip at deimos.tc.uaslp.mx> writes:

>          SAS        treatment    helmert    sum        sdif
> Pr(>F)   3.073e-05  3.073e-05    3.073e-05  3.073e-05  3.073e-05
>
> the probabilities of the different contrasts were
>
>          SAS        treatment    helmert    sum        sdif
> A-B                 0.072635     0.072635              0.072635
> A-C      8.69e-06   8.69e-06                0.000322
> B-C      0.00875                            0.637971   0.00875
>
> Why does the contrast contr.sum have distinct results from the other
> contrasts? Which ones are confiable?

Are you reading the output correctly? Sum contrasts are the ones that
sum to zero, so you get k-1 values and the k-th one is minus the sum
of the others. One result of this is that the test of each individual
coefficient is meaningless (whether a group level is equal to the
average of all group levels).

If the process is repeated with lme in place of lm with

> res <- lm(dbp~Treatment+Visit+dbp1,random=~1|Patient)

in place of

> res <- lm(dbp~Treatment+Visit+dbp1)

...
>          SAS        treatment    helmert    sum        sdif
> AIC      7501.212   7501.212     7511.151   7506.182   7501.212
> BIC      7546.116   7546.116     7556.055   7551.086   7546.116
> logLik   -3741.606  -3741.606    -3746.576  -3744.091  -3741.606
>
> Given that AIC and BIC are calculated logLik, it is reasonable that they
> differ given the different values of the logLik, but is it reasonable that
> the logLik's are different?

I think the model is contrast independent when specified like that.
(But you have a typo in the command, so that's not what you ran...)
Looks like there might a convergence problem with the sum and helmert
contrasts, although they are the ones normally considered numerically
more stable. Do the variance estimates look similar?

>          SAS        treatment    helmert    sum        sdif
> A-B                 0.2330       0.2330                0.2330
> A-C      0.0033     0.0033                  0.0168
> B-C      0.0850                             0.7553     0.0850
>
> Again, why does the contrast contr.sum have distinct results from the other
> contrasts? Which ones are confiable?

Same as before I think.

--
O__  ---- Peter Dalgaard             Blegdamsvej 3
c/ /'_ --- Dept. of Biostatistics     2200 Cph. N
(*) \(*) -- University of Copenhagen   Denmark      Ph: (+45) 35327918
~~~~~~~~~~ - (p.dalgaard at biostat.ku.dk)             FAX: (+45) 35327907
-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-
r-help mailing list -- Read http://www.ci.tuwien.ac.at/~hornik/R/R-FAQ.html
Send "info", "help", or "[un]subscribe"
(in the "body", not the subject !)  To: r-help-request at stat.math.ethz.ch
_._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._

```