[R-sig-ME] Convergence Error: 0 Fixed Correlations and More

[Emmanuel Curis]

Speaking about this factor variables coding to have uncorrelated
random effects... Following Thierry Onkelinx's advice [thanks a lot
for this advice, by the way, I think I didn't answered yet, sorry], I
wrote down the models for a similar case, and it turned out that when
having uncorrelated random effects for all levels of a covariate, then
the Y variables were also uncorrelated, but it allowed different
variances for each level.

I thought one advantage of random effects was to introduce
correlations between observations taken for the same patient, so I
wonder if this trick really does what one expects (assuming I didn't
made mistakes, and I am not the only one to have this idea about
consequences of random effects), and that specifying special random
effects covariance matrix structures is less error-prone with nlme?

Best regards,

On Tue, Sep 22, 2015 at 09:40:26AM -0400, Ben Bolker wrote:
« On Tue, Sep 22, 2015 at 3:33 AM, Thierry Onkelinx
« <thierry.onkelinx at inbo.be> wrote:
« > Dear Chris,
« >
« > The correct syntax is (1 + FactorC | item) not (1 + FactorC || item).
« > Use a single |. I find the item.1 strange in the output. This might be
« > due to the syntax error.
« 
«    Chris might be trying to suppress the correlations between
« random-effect component:
« the double-bar notation expands to (1|item) + (0 + FactorC | item),
« but there's a problem here: there's not *really* a way to do this with the
« double-bar syntax.  If FactorC has two levels (B and S), then the
« right (tedious)
« way to do this is
« 
« ( 1|item)+(0+dummy(FactorC,"C")|item)
« 
« or maybe (?)
« 
« (0+dummy(FactorC,"C")|item)(0+dummy(FactorC,"C")|item)
« 
« 
« 
« (I think the current model is overparameterized)
« 
« >
« > The item random effect variances are quit high. You might have a
« > problem of quasi-complete separation. (1 + FactorC | item) might be
« > too complex for your data. Does (1 | item) converge?
« >
« > Best regards,
« >
« > ir. Thierry Onkelinx
« > Instituut voor natuur- en bosonderzoek / Research Institute for Nature
« > and Forest
« > team Biometrie & Kwaliteitszorg / team Biometrics & Quality Assurance
« > Kliniekstraat 25
« > 1070 Anderlecht
« > Belgium
« 
« 
«   [snip]
« 
« >
« >
« > 2015-09-21 18:37 GMT+02:00 Chris Heffner <heffner at umd.edu>:
« >> Hi,
« >>
« >> I'm running a psychology experiment with a few fixed effects and random
« >> factors, but for some of the models that I'm comparing I get an output that
« >> looks something like this:
« >>
« >> Generalized linear mixed model fit by maximum likelihood (Laplace
« >> Approximation) ['glmerMod']
« >>  Family: binomial  ( logit )
« >> Formula: FW ~ FactorA + FactorB + FactorC + FactorA:FactorC +
« >> FactorB:FactorC +      (1 | participant) + (1 + FactorC || item)
« >>    Data: east.acc1.subset
« >> Control: glmerControl(optCtrl = list(maxfun = 30000))
« >>
« >>      AIC      BIC   logLik deviance df.resid
« >>   1001.5   1066.9   -487.7    975.5     1120
« >>
« >> Scaled residuals:
« >>     Min      1Q  Median      3Q     Max
« >> -3.8335 -0.3041  0.1416  0.3566  2.8851
« >>
« >> Random effects:
« >>  Groups      Name        Variance  Std.Dev.  Corr
« >>  item     FactorCB       5.454e+00 2.3352985
« >>              FactorCS       3.097e+00 1.7597629 -0.81
« >>  item.1   (Intercept) 5.437e+00 2.3316731
« >>  participant (Intercept) 2.595e-08 0.0001611
« >> Number of obs: 1133, groups:  item, 55; participant, 23
« >>
« >> (Intercept)            0.1928833  0.0006222   310.0   <2e-16 ***
« >> FactorAInitial        1.8077886  0.0006222  2905.5   <2e-16 ***
« >> FactorB150        -0.4506653  0.0006220  -724.5   <2e-16 ***
« >> FactorB200        -0.5485114  0.0006220  -881.9   <2e-16 ***
« >> FactorCS                 -0.3923921  0.0006221  -630.8   <2e-16 ***
« >> FactorAInitial:FactorCS -0.0889474  0.0006221  -143.0   <2e-16 ***
« >> FactorB150:FactorCS   0.1347207  0.0006221   216.6   <2e-16 ***
« >> FactorB200:FactorCS   0.0682518  0.0006221   109.7   <2e-16 ***
« >> ---
« >> Signif. codes:  0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1
« >>
« >> Correlation of Fixed Effects:
« >>             (Intr) FAIn FB150 FB200 FCS FAI:FCS FB150:FCS
« >> FAIntl 0.000
« >> FB150 0.000  0.000
« >> FB200 0.000  0.000  0.000
« >> FCS       0.000  0.000  0.000  0.000
« >> FaInt:FCS 0.000  0.000  0.000  0.000  0.000
« >> FB150:FCS 0.000  0.000  0.000  0.000  0.000 0.000
« >> FB200:FCS 0.000  0.000  0.000  0.000  0.000 0.000  0.000
« >>
« >> convergence code: 0
« >> Model failed to converge with max|grad| = 0.113738 (tol = 0.001, component
« >> 1)
« >> Model is nearly unidentifiable: very large eigenvalue
« >>  - Rescale variables?
« >>
« >> I've tried look through my data, as my first thought was that data was
« >> somehow miscoded, but I can't see anything that would be the matter.  A
« >> more complicated version of the model had the same problem until I got rid
« >> of a single participant (who seemed otherwise entirely unexceptional).  The
« >> more complicated model now converges fine, but this simpler one now has
« >> these issues.  I have an almost identical dataset that I've been doing
« >> almost exactly the same models with that hasn't been giving me similar
« >> problems.
« >>
« >> Any thoughts?
« >>
« >> Thank you,
« >>
« >> Chris
« >>
« >>         [[alternative HTML version deleted]]
« >>
« >> _______________________________________________
« >> R-sig-mixed-models at r-project.org mailing list
« >> https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models
« >
« > _______________________________________________
« > R-sig-mixed-models at r-project.org mailing list
« > https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models
« 
« _______________________________________________
« R-sig-mixed-models at r-project.org mailing list
« https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models

-- 
                                Emmanuel CURIS
                                emmanuel.curis at parisdescartes.fr

Page WWW: http://emmanuel.curis.online.fr/index.html