I would like to run a lagged random effects regression.
The data is from an experiment in which participants were assigned to groups of five and participated in an interactive game for 20 rounds.
Participants could exchange something during the experiment, which is the dependent variable.
Now I would like to predict/explain, how much participant received from other participants based on the behaviour of previous rounds.
Since the data is clustered on three levels: subject, group and time (rounds), I am a little bit lost how to correctly formulate the model.
I am currently using the lme4 package in R. I transformed the dependent variable to a 0/1 (nothing received/something received) variable, due to high skewness, so I would need to specify a multilevel logistic model.
So far, I specified and ran the following models:
glmer(DV ~ predictors* + (1 + round * subject | group), family = binomial)
and:
glmer(DV ~ predictors* + (1 + round * group | subject), family = binomial)
*predictors are on subject-level.
I get similar (although not the same) estimates for both models, however in model1, z-values are much higher (and therefore p-values much lower).
Can someone help me on that?
What I want to know is; Can previous behaviour (that is behaviour from round x-1 etc.) predict how much a participant received in round x. But control/acknowledge that participants are clustered in groups and that behaviour is correlated over time (rounds).
