# singular variance-covariance warning in lmer

3 messages
Open this post in threaded view
|

## singular variance-covariance warning in lmer

 Dear R Users, I was hoping for some help with a recurrent error message in lmer. I am trying to model the effect of temperature on metabolic rate in animals (response = int.length) at different temperatures (mean.sst), with repeated measurements on the same individuals (random effect = female). Ideally I would make a random slope and intercept model where the rate can change differently with temperature for different individuals: model<-lmer(int.length~mean.sst+(mean.sst|female)) However, I get the following warning message: Warning message: Estimated variance-covariance for factor 'female' is singular in: `LMEoptimize<-`(`*tmp*`, value = list(maxIter = 200L, tolerance = 1.49011611938477e-08, summary(model) Linear mixed-effects model fit by REML Formula: int.length ~ mean.sst + (mean.sst | female)    AIC   BIC logLik MLdeviance REMLdeviance  155.4 164.5  -72.7      142.8        145.4 Random effects:  Groups   Name        Variance   Std.Dev.   Corr  female   (Intercept) 6.8459e-10 2.6165e-05           mean.sst    6.8169e-10 2.6109e-05 -0.065  Residual             1.3634e+00 1.1676e+00 number of obs: 46, groups: female, 18 Fixed effects:             Estimate Std. Error t value (Intercept)  48.8249     6.5895   7.409 mean.sst     -1.3609     0.2518  -5.406 Correlation of Fixed Effects:          (Intr) mean.sst -1.000 If I try and run just a random intercepts model I get similar problems: model2<-lmer(int.length~mean.sst+(1|female)) Warning message: Estimated variance for factor 'female' is effectively zero in: `LMEoptimize<-`(`*tmp*`, value = list(maxIter = 200L, tolerance = 1.49011611938477e-08, I have tried disabling PQL iterations  using control = list(usePQL = FALSE, msVerbose=TRUE), following Douglas Bates' recommendation on the mailing list archives but I still get a similar message. Does this mean that the variance among subjects is too close to zero for estimation of the random effects? I compared the random effects model to a linear model with just lm(int.length ~ mean.sst) using a likelihood ratio test and got p = 1.0 (which is always suspicious). It would actually make sense for there to be negligible variation among subjects in their response to temperature, however I am concerned that I am making a fundamental error somewhere along the line. I would greatly appreciate any suggestions you may have. Best regards Sam Weber University of Exeter, UK.         [[alternative HTML version deleted]] ______________________________________________ [hidden email] mailing list https://stat.ethz.ch/mailman/listinfo/r-helpPLEASE do read the posting guide http://www.R-project.org/posting-guide.htmland provide commented, minimal, self-contained, reproducible code.