[R-sig-ME] Unidentifiable model in lmer

[Phillip Alday]

I suspect that is a large part of the problem - the extra parameters are collinear by definition and that will cause issues with model fit.

Phillip


> On 18 Sep 2015, at 14:53, Ché Lucero <chelucero at uchicago.edu> wrote:
> 
> I don't know if this is causing the problem you're seeing, but A*B expands
> to A + B + A:B, so your model right now is R ~ A + B + A + B + A:B + (1|S).
> 
> On Thu, Sep 17, 2015 at 5:53 PM Takahiro Fushimi <taka.6765 at gmail.com>
> wrote:
> 
>> Hi everyone,
>> 
>> I have been working on a linear mixed effect model by using lmer()
>> function, but I got an error saying that the fitted model is not
>> identifiable.
>> 
>> The data set includes the following variables:
>> y = a numeric variable
>> factorA = a 3-level categorical variable
>> factorB = a 2-level categorical variable
>> subjectID = subject id number. 2 measurements of y for each subject
>> 
>> R code and output are as follows:
>>> (result <- lmer(y ~ factorA + factorB + factorA*factorB +
>> (1|subjectID)))
>> Linear mixed model fit by REML ['merModLmerTest']
>> Formula: y ~ factorA + factorB + factorA * factorB + (1 | subjectID)
>> REML criterion at convergence: 1928.966
>> Random effects:
>>  Groups    Name        Std.Dev.
>>  subjectID (Intercept) 4711
>>  Residual              7688
>> Number of obs: 97, groups:  subjectID, 51
>> Fixed Effects:
>>       (Intercept)           factorA1           factorA2 factorB1
>> factorA1:factorB1  factorA2:factorB1
>>             62411              -2700              -1124
>> -1037               1279               2482
>>> summary(result)
>> Model is not identifiable...
>> summary from lme4 is returned
>> some computational error has occurred in lmerTest
>> Linear mixed model fit by REML ['lmerMod']
>> Formula: y ~ factorA + factorB + factorA * factorB + (1 | subjectID)
>> 
>> REML criterion at convergence: 1929
>> 
>> Scaled residuals:
>>      Min       1Q   Median       3Q      Max
>> -1.93423 -0.62611  0.01837  0.48887  2.73380
>> 
>> Random effects:
>>  Groups    Name        Variance Std.Dev.
>>  subjectID (Intercept) 22194074 4711
>>  Residual              59108495 7688
>> Number of obs: 97, groups:  subjectID, 51
>> 
>> Fixed effects:
>>                   Estimate Std. Error t value
>> (Intercept)          62411       2065  30.227
>> factorA1             -2700       3238  -0.834
>> factorA2             -1124       3008  -0.374
>> factorB1             -1037       2512  -0.413
>> factorA1:factorB1     1279       4013   0.319
>> factorA2:factorB1     2482       3642   0.681
>> 
>> Correlation of Fixed Effects:
>>             (Intr) fctrA1 fctrA2 fctrB1 fA1:B1
>> factorA1    -0.638
>> factorA2    -0.687  0.438
>> factorB1    -0.608  0.388  0.418
>> fctrA1:fcB1  0.381 -0.601 -0.262 -0.626
>> fctrA2:fcB1  0.420 -0.268 -0.606 -0.690  0.432
>>> 
>> 
>> Could anyone give me some idea of why this unidentifiability problem
>> happens and how to fix it?
>> Any help would be appreciated.
>> 
>> Best regards,
>> Takahiro
>> 
>> _______________________________________________
>> R-sig-mixed-models at r-project.org mailing list
>> https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models
>> 
> 
> 	[[alternative HTML version deleted]]
> 
> _______________________________________________
> R-sig-mixed-models at r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models