Skip to content

glmer warning messages using version 1.1.7

2 messages · Sophia Kyriakou, Ben Bolker

#
The bad news is that when I run the profList function I get an error
message instead
+                    bobyqa=g0.bobyqa,
+                    NM=g0.NM,
+                    nlminb=g0.nlminb,
+                    LBFGSB=g0.LBFGSB),profile,dev.tol=1e-3,
+                    delta=0.1)
There were 14 warnings (use warnings() to see them)
Error in diag(vcov(object, use.hessian = use.hessian)) :
  error in evaluating the argument 'x' in selecting a method for function
'diag': Error in solve.default(h) : system is computationally singular:
reciprocal condition number = 3.04588e-17

warnings()
Warning messages:
1: In zeta(shiftpar, start = opt[seqpar1][-w]) :
  slightly lower deviances (diff=-3.55271e-15) detected
2: In nextpar(mat, cc, i, delta, lowcut, upcut) :
  Last two rows have identical or NA .zeta values: using minstep
3: In nextpar(mat, cc, i, delta, lowcut, upcut) :
  Last two rows have identical or NA .zeta values: using minstep
4: In profile.merMod(X[[1L]], ...) : non-monotonic profile
5: In zeta(shiftpar, start = opt[seqpar1][-w]) :
  slightly lower deviances (diff=-3.55271e-15) detected
6: In nextpar(mat, cc, i, delta, lowcut, upcut) :
  Last two rows have identical or NA .zeta values: using minstep
7: In nextpar(mat, cc, i, delta, lowcut, upcut) :
  Last two rows have identical or NA .zeta values: using minstep
8: In profile.merMod(X[[2L]], ...) : non-monotonic profile
9: In nextpar(mat, cc, i, delta, lowcut, upcut) :
  Last two rows have identical or NA .zeta values: using minstep
10: In profile.merMod(X[[3L]], ...) : non-monotonic profile
11: In zetafun(np, ns) : slightly lower deviances (diff=-3.55271e-15)
detected
12: In zetafun(np, ns) : slightly lower deviances (diff=-3.55271e-15)
detected
13: In nextpar(mat, cc, i, delta, lowcut, upcut) :
  Last two rows have identical or NA .zeta values: using minstep
14: In profile.merMod(X[[4L]], ...) : non-monotonic profile


 Many thanks,
Sophia

Message: 5

  
  
1 day later
#
Sophia Kyriakou <sophia.kyriakou17 at ...> writes:
I'm snipping most of the context from this thread, but just coming
back here to say I've posted a fairly thorough analysis of this issue
(along with pictures) at

http://rpubs.com/bbolker/proftest

  The answer there about the original question, "is it OK to 
ignore thse warnings?", is

tl;dr; yes. These are singular fits (due to a small number of levels of the
grouping variable), but that?s not actually the proximal problem. It turns
out that the second derivative of the random-effects std dev parameter is
quite small near zero (the MLE), making the estimated uncertainty in this
parameter very large and triggering warnings about scaling, etc etc, from
lme4. Plotting the profiles demonstrates (for the most part) that the level
of uncertainty is actually reasonable (upper 95% confidence interval around
2), just not well characterized by the local curvature at the MLE.