Understand warning messages for mixed model in r lme4

r lme4 mixed-models
phish_researcher picture phish_researcher · Jun 13, 2014 · Viewed 9.9k times · Source

I have built and run a mixed effects logistic regression model in the lme4 package for r to estimate the probability of occupancy of fishes in different locations (cells/habitats). The data frame consists of 1,207,140 observations of 68 individual fish. For each individual (per day for ~ 1 year) it describes the the number of occurrences at each unique location relative to the total number of occurrences in all locations.

Here is the base model:

    m.base = glmer(cbind(N,t.move-N) ~ jdate + snSurface.Area + Restoration..P.A. +    
    Release.Location+ Sex + (1|Station) + (0 + jdate|ID), data=allfishdat, family=binomial)
where N=# unique positions, t.move=total positions, jdate=julian date, Station=locations, ID=fish ID

I get the following warning message:

Warning messages:
1: In checkConv(attr(opt, "derivs"), opt$par, ctrl = control$checkConv,  :
                Model failed to converge with max|grad| = 3349.26 (tol = 0.001)
2: In if (resHess$code != 0) { :
 the condition has length > 1 and only the first element will be used
3: In checkConv(attr(opt, "derivs"), opt$par, ctrl = control$checkConv,  :
 Model is nearly unidentifiable: very large eigenvalue
 - Rescale variables?;Model is nearly unidentifiable: large eigenvalue ratio
 - Rescale variables?

I done some searching to try to understand what these messages mean and their implications on the model, but have not yet understood the warnings.

Answer

It Figures picture It Figures · Jun 15, 2014

If the first problem has to do with R needing to go through more iterations to reach convergence, the code below may help. Replace '20000' with whatever max iteration number makes sense for your specific model. (Notice that your original model code has been modified at the end to include 'control = my.control'.) 

my.control=lmerControl(optCtrl=list(maxfun=20000); my.control

m.base = glmer(cbind(N,t.move-N) ~ jdate + snSurface.Area + Restoration..P.A. +  Release.Location+ Sex + (1|Station) + (0 + jdate|ID), data=allfishdat, family=binomial, control = my.control)

It may also be useful to you to review your current lmeControls with the following command:

str(lmerControl())

Further, this previous answer may be helpful to you: increase iterations for new version of lmer?

Related questions

How to get coefficients and their confidence intervals in mixed effects models?

In lm and glm models, I use functions coef and confint to achieve the goal: m = lm(resp ~ 0 + var1 + var1:var2) # var1 categorical, var2 continuous coef(m) confint(m) Now I added random effect to the model - used mixed …

Read More
lme4::lmer reports "fixed-effect model matrix is rank deficient", do I need a fix and how to?

I am trying to run a mixed-effects model that predicts F2_difference with the rest of the columns as predictors, but I get an error message that says fixed-effect model matrix is rank deficient so dropping 7 columns / coefficients. From this …

Read More
lmer error: grouping factor must be < number of observations

I am attempting to run a mixed effect model on some data but struggling with one of the fixed effects, I think primarily due to it a factor?! Sample data: data4<-structure(list(code = structure(1:10, .Label = c("10888", "10889", "10890", "10891", "10892", "10893", "10894", "10896", "10897", "10898", "10899", "10900", "10901", "10902", "10903", "10904", "10905", "10906", "10907", "10908", "10909", "10910", "10914", "10916", "10917", "10919", "10920", "10922", "10923", "10924", "10925", "10927"), class = "factor"), …

Read More