[R-sig-ME] false positives?

[Ben Bolker]

   The test you show below (although solve(Hessian,grad) is slightly
better) suggests that the first warning is indeed a false positive.  I
would expect the second message to go away if you scale (centering is
probably a good idea too) your continuous predictors.  (The t-statistics
shouldn't change.)

  Ben Bolker

On 14-05-25 11:34 PM, John Kingston wrote:
> I received the warning messages below in response to running the second
> model displayed below. The data file is attached to this message.
> 
> Warning messages:
> 1: In checkConv(attr(opt, "derivs"), opt$par, ctrl = control$checkConv,  :
>   Model failed to converge with max|grad| = 0.212882 (tol = 1e-06)
> 2: In checkConv(attr(opt, "derivs"), opt$par, ctrl = control$checkConv,  :
>   Model is nearly unidentifiable: very large eigenvalue
>  - Rescale variables?
> 
> sfHIndR <- glmer(fRspRN ~ cStep +
>     h1Context +
>     h2Context +
>     cQRT +
>     (1 + cStep +
>     h1Context +
>     h2Context +
>     cQRT | participant),
>     family = "binomial",
>     control = glmerControl(optCtrl = list(maxfun=1000000)),
>     data = sfCompiled)
> 
> summary(sfHIndR)
> 
> # interactions with RT
> 
> sfHSxRtCxRtR <- glmer(fRspRN ~ cStep * cQRT +
>     h1Context * cQRT +
>     h2Context * cQRT +
>     (1 + cStep +
>     h1Context +
>     h2Context +
>     cQRT | participant),
>     family = "binomial",
>     control = glmerControl(optimizer = "bobyqa", optCtrl =
> list(maxfun=1000000)),
>     data = sfCompiled)
> 
> The test suggested in a earlier post produced the output below. These
> values look small.
> 
> gg <- sfHSxRtCxRtR at optinfo$derivs$grad
> hh <- sfHSxRtCxRtR at optinfo$derivs$Hessian
> vv <- sqrt(diag(solve(hh/2)))
> summary(abs(gg*vv))
> 
> #       Min.   1st Qu.    Median      Mean   3rd Qu.      Max.
> # 6.720e-07 3.108e-05 8.866e-05 2.542e-04 1.745e-04 3.345e-03
> 
> Any advice about how to proceed? Thanks.
> Best,
> John
> 
> 
> 
> John Kingston
> --------------------------------
> Professor and Head
> Linguistics Department
> University of Massachusetts
> 150 Hicks Way, 226 South College
> Amherst, MA 01003-9274
> 1-413-545-6837, fax -2792
> jkingston at linguist.umass.edu
> https://blogs.umass.edu/jkingstn/
> 
> 
> _______________________________________________
> R-sig-mixed-models at r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models
>