Error in Profile likelihood based confidence intervals in glmer()

Tue, Jul 22, 2014 11:55 AM

I should add this to the "troubleshooting" page, but:

* There's in principle no problem at all with profiling models with
multiple random effects (other than that it's likely to be slow)
* The error message indicates that during profiling, the optimizer found
a fitted value that was significantly better (as characterized by the
'devtol' parameter) than the supposed minimum-deviance solution returned
in the first place.  You can boost the 'devtol' parameter (which is
currently set at a conservative 1e-9 ...) if you want to ignore this --
however, the non-monotonic profiles are also warning you that something
may be wonky with the profile.  It should (???) be possible to capture
the new/improved parameters that were found (although I don't know if
this is implemented in profile.merMod; I may have done it for bbmle but
not for lme4).
* the 'slice2D' methods in the bbmle package (see e.g.
http://rpubs.com/bbolker/22607 ) may be useful for exploring the
likelihood surface.

On 14-07-22 11:40 AM, Ravi Varadhan wrote:

Hi,

I have longitudinal binary responses from a clinical trial. I am fitting
the following random effects model in lme4::glmer.  The model is
estimated without any problems.  However, I get an error when I try to
compute the confidence intervals using the profile likelihood.  Does not
the profiling approach work for more than one random effect?  Can
someone point  out the problem?

Thanks,

Ravi

 

summary(mod2 <- glmer(imps79b ~ tx + sweek + (sweek|id), data=schiz,
family=binomial))

 

Generalized linear mixed model fit by maximum likelihood (Laplace
Approximation) ['glmerMod']

Family: binomial  ( logit )

Formula: imps79b ~ tx + sweek + (sweek | id)

   Data: schiz

 

     AIC      BIC   logLik deviance df.resid

  1259.4   1291.7   -623.7   1247.4     1597

 

Scaled residuals:

     Min       1Q   Median       3Q      Max

-3.03973  0.00912  0.04957  0.21372  1.45395

 

Random effects:

Groups Name        Variance Std.Dev. Corr

 id     (Intercept) 13.993   3.741        

        sweek        4.093   2.023    -0.65

Number of obs: 1603, groups:  id, 437

 

Fixed effects:

            Estimate Std. Error z value Pr(>|z|)   

(Intercept)   9.1994     1.3105   7.020 2.22e-12 ***

tx           -2.5075     0.5996  -4.182 2.89e-05 ***

sweek        -3.1344     0.4503  -6.960 3.40e-12 ***

---

Signif. codes:  0 ?***? 0.001 ?**? 0.01 ?*? 0.05 ?.? 0.1 ? ? 1

 

Correlation of Fixed Effects:

      (Intr) tx    

tx    -0.786      

sweek -0.922  0.537

 

# profiled confidence intervals

confint(mod2, method="profile")

Computing profile confidence intervals ...

Error in zetafun(np, ns) : profiling detected new, lower deviance

In addition: Warning messages:

1: In profile.merMod(object, signames = oldNames, ...) :

  non-monotonic profile

2: In profile.merMod(object, signames = oldNames, ...) :

  non-monotonic profile

 

 

Ravi Varadhan, Ph.D. (Environmental Eng.), Ph.D. (Biostatistics)

Associate Professor,

Division of Geriatric Medicine & Gerontology

School of Medicine,

Johns Hopkins University

Ph: 410-502-2619

Email: ravi.varadhan at jhu.edu <mailto:ravi.varadhan at jhu.edu>

http://www.jhsph.edu/research/centers-and-institutes/johns-hopkins-center-on-aging-and-health/people/Faculty_personal_Pages/Varadhan.html

Error in Profile likelihood based confidence intervals in glmer()

Thread (2 messages)