# Linear mixed effects with covariates

I have a group of participants (Subj) all of whom have a covariate measure (Pars). Each participant saw a series of images (Image) that belonged to one of three categories (Stim). To each image, participants made a response (Resp) that was classified as a behavior (Beh). I have combined Stim and Beh (StimBeh), and for each of these Stimulus-Behaviors I have a parameter estimate (Est). I would like to see what the effect of the covariate is in the parameter estimate for the different types of behaviors associated with the different stimuli types, when treating subject and image as random effects. I believe StimBeh should be a fixed effect. I have been unable to make the model converge.

• You might do better on stat.ethz.ch/mailman/listinfo/r-sig-mixed-models if no replies here. Commented Sep 4, 2020 at 16:08
• Does the model with (1 | Subj) rather than (Pars | Subj) converge ? Also please don't post a picture of output, Post the output itself. How many subjects and images do you have ? And why do you appear to be subsetting the data based on values of the outcome ? Commented Sep 5, 2020 at 7:55
• I do not understand your nesting (or non-nesting) structure. You say that for "each of these Stimulus-Behaviors I have a parameter estimate (Est)." That implies that StimBehav represents the lowest level of your hierarchy, yet it is repeated, just as subjects are repeated. Can you be more specific about your data, perhaps posting a sample of it for us to look at? Commented Sep 5, 2020 at 15:45

Combining Stim and Beh is a strange approach. I may be missing something, but I would have incorporated Stim and Beh (or possibly Resp) as covariates. Combining them imposes an interaction without allowing each in the model as a main effect. Finally, if your parameter lies in $$[0,1]$$, you might want to transform the response.