# Multi-level modeling using lmer in R

I have nested data as followed: 'Learning group sessions' in 'students' in 'learning groups' (each learning group consisted of different students and the learning groups chose to have between 3 and 10 sessions to learn).

Further, each learner had to state for each session whether he faced a motivational or a knowledge problem instead (but I guess this variable is no nested factor).

My dependent variable is the number of mentioned strategies ('Score') a learner used to tackle the specific problem he faced.

With my multilevel regression I would like to find out if students in the learning groups have (differently) increased the number of used strategies over time (with an increasing number of learning sessions). It can be assumed that both the learners in the groups and the groups themselves have different intercepts regarding the use of the strategy. But also that the use of strategies over the sessions for individual learners and also for the groups increases (or decreases) differently.

I am not sure if I have to model fixed and random effects for each level (which are likely here to cause convergence problems) and if my code whould therefore be correct, or if I should assume only random effects for all nested factors.

model <- lmer(Score ~ Problemtype + Session + Person + Group + (1+Person |Group) + (1+Session|Group:Person) + (1|Group:Person:Session), data=Data, REML = FALSE)

• So the predictor you are actually interested in is Session? And Problemtype is an additional predictor? Then these are fixed effects, the rest (Group and Person) are random. Score ~ Problemtype * Session + (Problemtype * Session | Group/Person). – amoeba Aug 15 '18 at 9:29
• Exactly, session should be a predictor. But I would also like to test to what extent group is a predictor. Yes, problem type should also be a predictor (but not nested, I guess) – Nadine M. Aug 15 '18 at 9:36