Model failing to converge with LMER

Question

I want to predict the relationship between the number of groups a person belongs to and their overall well-being (totalwell). However, I would like to consider the possibility that a person could belong to numerous groups of low quality. To address this, I intend to use participant MapID and group_quality as random effects, while using the number of groups a person belongs to as the predictor and total well-being as the outcome variable.

I have this dataset:

 MapID CircleID TotalNumberofCircles totalwell group_quality
   <dbl>    <int>                <dbl>     <dbl>         <dbl>
 1  1381        1                    5      4.93          8.67
 2  1381        2                    5      4.93          2.67
 3  1381        3                    5      4.93         10   
 4  1381        4                    5      4.93          7.67
 5  1381        5                    5      4.93          5.33
 6  1391        1                    7      4.53          9   
 7  1391        2                    7      4.53          7   
 8  1391        3                    7      4.53          5   
 9  1391        4                    7      4.53          8   
10  1391        6                    7      4.53          7   
# ... with 190 more rows
> 

And this is the model I'd like to run:

ctrl <- lmerControl(optimizer = "Nelder_Mead", optCtrl = list(maxfun = 100))
MLM <- lmer(totalwell ~ CircleID + (1 | MapID:group_quality), data = df, control = ctrl)

And I'm getting the following convergence issues. Can anyone help me out here? Is there a problem with how my data is structured?

> ctrl <- lmerControl(optimizer = "Nelder_Mead", optCtrl = list(maxfun = 100))
> MLM <- lmer(totalwell ~ CircleID + (1 | MapID:group_quality), data = df, control = ctrl)
Warning messages:
1: In (function (fn, par, lower = rep.int(-Inf, n), upper = rep.int(Inf,  :
  failure to converge in 100 evaluations
2: In optwrap(optimizer, devfun, getStart(start, rho$pp), lower = rho$lower,  :
  convergence code 4 from Nelder_Mead: failure to converge in 100 evaluations
3: In checkConv(attr(opt, "derivs"), opt$par, ctrl = control$checkConv,  :
  Model failed to converge with max|grad| = 0.237603 (tol = 0.002, component 1)
4: In checkConv(attr(opt, "derivs"), opt$par, ctrl = control$checkConv,  :
  Model is nearly unidentifiable: very large eigenvalue
 - Rescale variables? 

EDIT!!

Okay so I have since restructured the model and this is what I'm trying to run:

MLM <- lmer(totalwell ~ CircleID + (1 + group_quality | MapID), 
            data = df)

and my fit is now singular

> MLM <- lmer(totalwell ~ CircleID + (1 + group_quality | MapID), 
+             data = df)
boundary (singular) fit: see help('isSingular')
> 

I've tried simplifying the model by taking out group_quality, and just holding Map ID as the only random effect. but this didn't work. In fact, it failed to converge when I took out group_quality

I'm wondering if perhaps I could run a bayesian analysis instead or if maybe I could use a different kind of model?

EDIT #2: I have since been trying this model instead, but still no luck with fixing the singular fit:

MLM <- lmer(totalwell ~ TotalNumberofCircles + (1 + group_quality | CircleID) + (1| MapID), 
        data = df)

Answer 1

The most obvious issue I can see here is that your random effects coding doesn't make a lot of sense. Remember that random effects are clusters which means they will be factors. Your group quality variable is numeric which is already a problem, but you have nested it as a random effect within MapID, which also doesn't make sense.

If group quality varies within subjects, it should be modeled like this:

MLM <- lmer(totalwell ~ CircleID + (1 + group_quality | MapID), 
            data = df)

Whether your include other controls into that model is up to how it converges, so I leave that up to you.