[R-sig-ME] r crossed nested random effects lme4

[Alday, Phillip]

lme4 doesn't force a hard nested-crossed distinction and can handle implicit nesting/crossing easily (assuming unique identifiers, which you have), so you don't have to worry about that.

The only "problem" I see with your model is that it is an "intercept-only" model. Given that there are only three items per condition, this makes sense for the by-item random effect. But you should consider whether the by-subject random effect should have a slope for condition. This is all assuming that you only sent us a screenshot of the top of the dataset and that you have more than three subjects ...

There are some more general issues about whether you should transform reaction time, but a quick search will yield lots of papers discussing the pros and cons of that.

Finally, please be kind to the list and be consistent in your names -- you swap back and forth between Group and Cond in your description.

Phillip

On 28/06/18 09:13, audusseau jean via R-sig-mixed-models wrote:

Dear all,I try to find the appropriate model for the following data set with lme4:
Each individual subject goes through 3 conditions ("Group", within-subject factor). These 3 conditions include 9 items (3 items in each condition, but the items differ in each condition). Score (4th column not represented above) is a continuous dependant variable (reaction time).I am interested in the fixed effect of the "Cond" variable and also would like to take into account the dependencies between my factors, but I have difficulties to know if my factors should be considered as nested or crossed.Does the following model seems correct to you ?Score~1+Cond+(1|Subject)+(1|item).Any help would be much appreciated.





_______________________________________________
R-sig-mixed-***@r-project.org<mailto:R-sig-mixed-***@r-project.org> mailing list
https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models



[[alternative HTML version deleted]]

[Phillip Alday]

Yes, use that model. Since Items are nested within conditions, it
doesn't make sense to have an by-item slope for condition.

Phillip

Phillip,
Many thanks for your answer. I apologize for such an unclear description
of my experiment. I wanted to make it simpler but it seems I produced
the opposite.
- _71_ subjects
- going through 3 conditions (so "condition" and "subject" are crossed)
- each condition comprises _30_ items (so items are nested in conditions)
Concerning the dependant variable, I use 1/RT.
As I understand your comments, I should acknowledge that the condition
1/RT~1+condition+(condition|subject) +(1| item)
Given that I have 30 items in each condition, would you advice to modify
the last term of this syntax ?
Thank you for your time.
Le jeudi 28 juin 2018 à 21:59:56 UTC+2, Alday, Phillip
lme4 doesn't force a hard nested-crossed distinction and can handle
implicit nesting/crossing easily (assuming unique identifiers, which you
have), so you don't have to worry about that.
The only "problem" I see with your model is that it is an
"intercept-only" model. Given that there are only three items per
condition, this makes sense for the by-item random effect. But you
should consider whether the by-subject random effect should have a slope
for condition. This is all assuming that you only sent us a screenshot
of the top of the dataset and that you have more than three subjects ...
There are some more general issues about whether you should transform
reaction time, but a quick search will yield lots of papers discussing
the pros and cons of that.
Finally, please be kind to the list and be consistent in your names --
you swap back and forth between Group and Cond in your description.
Phillip

Post by Alday, Phillip
Each individual subject goes through 3 conditions ("Group", within-subject factor). These 3 conditions include 9 items (3 items in each condition, but the items differ in each condition). Score (4th column not represented above) is a continuous dependant variable (reaction time).I am interested in the fixed effect of the "Cond" variable and also would like to take into account the dependencies between my factors, but I have difficulties to know if my factors should be considered as nested or crossed.Does the following model seems correct to you ?Score~1+Cond+(1|Subject)+(1|item).Any help would be much appreciated.
_______________________________________________
https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models