[R-sig-ME] Interaction terms with random slopes

[Yashree Mehta]

Hi,

Thanks for your reply.

The model is as follows:

Production = seed + fertilizer + fertilizer : wheatlanduse + (1 +
fertilizer | Household)

The code implies the following: Production is the dependent variable and
the covariates are seed and fertilizer. Random intercept has been
specified at the Household level and the random slope variable is
fertilizer. Cross level interaction has been introduced with a wheat land
use dummy variable which interacts with fertilizer. As can be seen,
fertilizer has also been specified as a fixed effect.

My question is as follows: Is it necessary to also include wheatlanduse as
a fixed effect because it has been introduced as an interaction? I prefer
not to because it creates problems of multicollinearity with other
covariates in the model.

Thanks again.

Yashree






On Fri, Apr 3, 2020 at 11:13 PM Alday, Phillip <Phillip.Alday using mpi.nl> wrote:

> Perhaps because I never had to deal with software that relies on
> explicit nesting of levels, I find it very hard to follow the level-1
> and level-2 terminology. Can you provide an example of the model you're
> thinking of and how you would test the corresponding hypothesis (e.g.
> model comparison, etc.)?
>
> The usual rule of thumb is that for any random slope, you should have
> the corresponding fixed-effect slope in the model. There's a discussion
> of this over on CrossValidated:
> https://stats.stackexchange.com/a/339859/26743
>
> Now, I can think of exceptions to this rule, much like I can think of
> exceptions to the rule that you should always include the intercept in a
> linear model. But in both cases, if you have to ask, it usually means
> you shouldn't do it.
>
> On 11/3/20 9:00 pm, Yashree Mehta wrote:
> > Hi,
> >
> > I have the following question:
> >
> > I estimate a random intercept-random slope model. For my research
> question,
> > I want to interact the random slope variable with another level-1
> > variable(which is not necessary as a main effect in the model). I am
> > interesting in observing whether this level-1 variable moderates the
> random
> > slope on the dependent variable.
> >
> > Do I have to include the level-1 variable as a main effect in the model
> (as
> > is required by some linear modelling literature)? I would prefer not to
> > include this level-1 variable as a main effect due to problems of
> > multicollinearity.
> >
> > Thank you very much!
> >
> >       [[alternative HTML version deleted]]
> >
> > _______________________________________________
> > R-sig-mixed-models using r-project.org mailing list
> > https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models
>

	[[alternative HTML version deleted]]