0
$\begingroup$

My data has 264 subjects with their serial measurements. Event is death and sataset has 67 dead, 197 alive observations.

I'm trying to model repeated x and y measurements along with age and gender characteristics via Mixed - Effects Logistic Regression Model. My R code is below:

mel <- glmer(status ~ x + y + age + gender + (1 | time)+ (1|id), data = mydata, family = binomial, control = glmerControl(optimizer = "bobyqa"))

When I use this code, I encounter overfitting (I'm not sure this word is the best one to explain the situation). I mean, before modelling I have 67 events, after modelling, I have again 67 events, as predicted probabilities suggest.

On the other hand, when I try to model these repeated values via this other code below, it gave me an error like this:

mel <- glmer(status ~ x + y+age+gender+ (1 | time), data = mydata, family = binomial, control = glmerControl(optimizer = "bobyqa"))

Warning message: In checkConv(attr(opt, "derivs"), optpar,ctrl=controlpar,ctrl=controlcheckConv, : Model is nearly unidentifiable: very large eigenvalue - Rescale variables?

I'd like to know, if I'm missing something or doing it wrong.

$\endgroup$
  • $\begingroup$ Can you provide a sample of the data? Have you checked that the numeric variables are considered numeric by R, and the factors are factors? $\endgroup$ – Nakx Nov 20 '17 at 10:57
0
$\begingroup$

Part of my data looks like this:

id time age gender x y

1 0 61 1 3,814621015 1,961307217

1 1 61 1 3,63347795 2,044523568

1 2 61 1 3,421073664 2,037971697

1 5 61 1 3,148801558 2,359346485

2 0 60 1 4,180261483 3,695965805

2 3 60 1 3,421073664 1,758059625

2 4 60 1 4,363847995 0,827560091

2 5 60 1 5,335078126 2,799822309

3 0 61 0 2,988464864 -2,037793335

3 2 61 0 2,967471914 -1,722773861

3 3 61 0 2,410342633 -1,59529577

When I check the numeric and factor variables, R again gave another error:

Warning messages: 1: In checkConv(attr(opt, "derivs"), opt$par, ctrl = control$checkConv, : unable to evaluate scaled gradient 2: In checkConv(attr(opt, "derivs"), opt$par, ctrl = control$checkConv, : Model failed to converge: degenerate Hessian with 1 negative eigenvalues

I got confused. Any help would be greatly appreciated.

Thanks in advance.

$\endgroup$
  • $\begingroup$ Try to rewrite the data as.factor or as.numeric before you launch the glmer, by using: mydata$id <- as.factor(mydata$id) $\endgroup$ – Nakx Nov 21 '17 at 6:10
0
$\begingroup$

For similar issues, some suggest that the dataset is too small https://stackoverflow.com/questions/25205633/after-trying-various-optimzers-model-simplification-running-more-iterations-wh or that the issue is caused by the use of the BOBYQA optimiser "Model failed to converge" warning in lmer()

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.