Hi
I try to implement a simple 2-factorial repeated-measure anova in the
lme framework and would be grateful for a short feedback
-my dependent var is a reaction-time (rt),
-as dependent var I have
-the age-group (0/1) the subject belongs to (so this is a
between-subject factor), and
-two WITHIN experimental conditions, one (angle) having 5, the other
3 (hands) factor-levels; means each subjects performs on 3 * 5 = 15
different task diffiulties
Am I right in this lme implementation, when I want to investigate the
influence of the age.group, and the two conditions on the rt:
my.lme <- lme(rt ~ age.group + angles * hands, data = my.data, random ~ 1
|subject)
then I think I would have to compare the model above with a more
elaborated one, including more interactions:
my.lme2 <- lme(rt ~ age.group * angles * hands, data = my.data, random
= ~ 1 |subject)
and comparing them by performing a likelhood-ratio test, yes?
I think, if I would like to generalize the influence of the experimental
conditions on the rt I should define angles and hands as a random
effect, yes?
?
thanks for a short feedback. It seems, repeated-measures anova's aren't
a trivial topic in R :)
Cheers!
Christoph
--
Christoph Lehmann <christoph.lehmann at gmx.ch>
On 11 Jun 04, at 20:12, Christoph Lehmann wrote:> Hi > I try to implement a simple 2-factorial repeated-measure anova in the > lme framework and would be grateful for a short feedback > > -my dependent var is a reaction-time (rt), > -as dependent var I have > -the age-group (0/1) the subject belongs to (so this is a > between-subject factor), and > -two WITHIN experimental conditions, one (angle) having 5, the other > 3 (hands) factor-levels; means each subjects performs on 3 * 5 = 15 > different task diffiulties > > Am I right in this lme implementation, when I want to investigate the > influence of the age.group, and the two conditions on the rt: > > my.lme <- lme(rt ~ age.group + angles * hands, data = my.data, random > ~ 1 |subject) > > then I think I would have to compare the model above with a more > elaborated one, including more interactions: > > my.lme2 <- lme(rt ~ age.group * angles * hands, data = my.data, random > = ~ 1 |subject) > > and comparing them by performing a likelhood-ratio test, yes? > > I think, if I would like to generalize the influence of the experimental > conditions on the rt I should define angles and hands as a random > effect, yes? >Perhaps I've missed something here, but wouldn't your ability to generalize about the experimental conditions depend, in part, on how their levels were selected? Were the angles randomly sampled? Were the hands randomly sampled (not sure what that would mean)? If not, how does defining these conditions to be random effects in a model enable valid generalization? ---JRG John R. Gleason Syracuse University 430 Huntington Hall Voice: 315-443-3107 Syracuse, NY 13244-2340 USA FAX: 315-443-4085 PGP public key at keyservers
Hi Christoph,> Am I right in this lme implementation, when I want to investigate the > influence of the age.group, and the two conditions on the rt: > > my.lme <- lme(rt ~ age.group + angles * hands, data = > my.data, random = ~ 1 |subject) > > then I think I would have to compare the model above with a more > elaborated one, including more interactions: > > my.lme2 <- lme(rt ~ age.group * angles * hands, data = > my.data, random = ~ 1 |subject) > > and comparing them by performing a likelhood-ratio test, yes?If you compare these two models, you test whether the interactions of age.group with angles and hands and the three-way interaction all together make for a significant improvement of the model. Is that what you want? Also note: if you do this, you need to use the method ML and not the default REML. Or you start with the second model, use anova (my.lme2) and reduce the model stepwise. You can also ask whether there is a subject to subject variability in variables other than the intercept (i.e. interactions between your random and the fixed variables) and e.g. try things like random = ~ age.group + angles * hands | subject> I think, if I would like to generalize the influence of the experimental > conditions on the rt I should define angles and hands as a random effect, > yes?I do not see exactly what you aim at, here. Possibly, the second part of my answer above is an answer to this as well?> thanks for a short feedback. It seems, repeated-measures > anova's aren't a trivial topic in R :)They never are. But, after having read most of Pinheiro and Bates' book 'Mixed effects modelling in S and S-PLUS' (Springer), it seems easier to me than ever, because they use a consistent, integrated and concise approach. Regards, Lorenz - Lorenz Gygax, Dr. sc. nat. Tel: +41 (0)52 368 33 84 / lorenz.gygax at fat.admin.ch Center for proper housing of ruminants and pigs Swiss Veterinary Office agroscope FAT T??nikon, CH-8356 Ettenhausen / Switzerland Fax : +41 (0)52 365 11 90 / Tel: +41 (0)52 368 31 31