Hi,
Sorry to bother you again.
I tried doing regressions using lme (because i want p-values) and ran
across two issues.
*(1) about model specification:*
this is a mixed model. i tried two model specifications:
(a) specify that subjectID ("subj") is nested within the between-subject
variable ("age"):
* summary(lme(bias ~ dprime * emo * race * age, random =~ 1|age/subj, data
= data)) *
=> gives NaN p-values for "age" :-(
(b) omit to specify the nested relationship between the variables:
* summary(lme(bias ~ dprime * emo * race * age, random =~ 1|subj, data =
data))*
=> gives numerical p-values for "age"
should i go with the specification (b), even if the variables are indeed
nested?
is the formula in (a) wrong?
*(2) about anova(model):*
because several of the factors have more than 2 levels, i wanted to use *
anova()* on the model to compute p-values for the whole effects.
however, the results from *summary(model)* and *anova(model)* are
inconsistent.
sometimes significant variables in *summary(model)* are not significant in *
anova(model)*, sometimes the reverse occurs. this also happens with factors
that only have 2 levels (so not due to dummy variables).
is this normal, is it a bad sign? what does it mean?
Thank you for any hint about those issues.
Best,
L.
2013/10/9 laurie bayet <laurieba...@gmail.com>
> Hi Peter, and thank you for your quick and helpful reply !
>
> "Do you want to know whether the predictors affect the marginal
> distributions of Y1, Y2,... or are you interested in conditional effects
> given other DVs (aka test for additional information)?"
> Hmmm I think that, yes, i am looking for that additional information
> (although i don't know what "marginal distributions" means). *So multiple
> regression it is, thank you !*
>
> Yes *i do have a random effect* to include (subject's number). Is it ok
> to do that ? Can i do this multiple regression with, say, lmer or glmer,
> even if i am not sure if the relation between the two DVs is actually
> linear (can i run a multiple regression on ranks instead, should i test for
> a linear correlation beforehand ?) ?
>
> Thank you again, your answer was very helpful :-)
>
> Best,
> L.
>
>
> 2013/10/9 peter dalgaard <pda...@gmail.com>
>
>> As a matter of principle, yes, multivariate mixed models do exist, look
>> at the last bit of example(manova) (in reasonably recent versions of R).
>>
>> In practice, it often doesn't really buy you much. It just gives a joint
>> test for all the DVs, the estimates are the same as in separate analyses.
>>
>> The tricky bit is usually to define precisely what the research question
>> is: Do you want to know whether the predictors affect the marginal
>> distributions of Y1, Y2,... or are you interested in conditional effects
>> given other DVs (aka test for additional information)? The latter case
>> leads to regression models where other DVs are entered as covariates.
>>
>> There's no issue with having categorical variables as predictors in
>> multiple regression in R, dummy variables are created internally. But if
>> you are considering mixed models, presumably you have a random effect that
>> needs to be included?
>>
>> -pd
>>
>> On Oct 9, 2013, at 10:23 , laurie bayet wrote:
>>
>> > Hi,
>> >
>> > Sorry to bother you again.
>> >
>> > I would like to estimate the effect of several categorical factors (two
>> > between subjects and one within subjects) on two continuous dependent
>> > variables that probably covary, with subjects as a random effect. *I
>> want
>> > to control for the covariance between those two DVs when estimating the
>> > effects of the categorical predictors** on those two DVs*. The thing
>> is, i
>> > know the predictors have an effect on DV1, and i know DV2 covaries with
>> > DV1, so it would be "cheating" to simply estimate the effect of the
>> > predictors on DV2 because those effects could be indirect (via DV1),
>> right ?
>> >
>> > I see two solutions :
>> >
>> > *One solution would be a mixed model MANOVA (if that even exists)*. But
>> i
>> > don't know how to run a mixed model MANOVA, i tried to do it with
>> > Statistica but couldn't find the right module (I know how to declare two
>> > DVs and run a GLM, but *I don't know if the covariance between my two
>> DVs
>> > is automatically controlled for*). Same thing with R. I tried to ask a
>> > question on Statistica's forum with no answer, tried looking around in
>> the
>> > manuals with no improvement.
>> >
>> > *A backup solution would be a multiple regression* (regressing DV2
>> against
>> > DV1 with the categorical predictors). But i am not sure how to
>> implement a
>> > mixed model, which function i should use and besides, it would be *much
>> > less convenient because one of my categorical predictors has three
>> > levels*(so i would have to split it and make it two predictors,
>> > right?).
>> >
>> > Thank you for any help at all !
>> >
>> > Cheers,
>> >
>> > L.
>> >
>> > [[alternative HTML version deleted]]
>> >
>> > ______________________________________________
>> > R-help@r-project.org mailing list
>> > https://stat.ethz.ch/mailman/listinfo/r-help
>> > PLEASE do read the posting guide
>> http://www.R-project.org/posting-guide.html
>> > and provide commented, minimal, self-contained, reproducible code.
>>
>> --
>> Peter Dalgaard, Professor
>> Center for Statistics, Copenhagen Business School
>> Solbjerg Plads 3, 2000 Frederiksberg, Denmark
>> Phone: (+45)38153501
>> Email: pd....@cbs.dk Priv: pda...@gmail.com
>>
>>
>
[[alternative HTML version deleted]]
______________________________________________
R-help@r-project.org mailing list
https://stat.ethz.ch/mailman/listinfo/r-help
PLEASE do read the posting guide http://www.R-project.org/posting-guide.html
and provide commented, minimal, self-contained, reproducible code.
