[R-lang] [SPAM] Re: Re: comparisons in lmer

Tue Jan 18 09:11:42 PST 2011

Doing the Tukey's all-pairwise comparisons as in Andy's previous post  
works for me if I analyze my two groups of participants separately -  
then I can compare morpheme types within each group.
If I do the same thing in my original model that includes morpheme  
type and fluency as two interacting predictors, I get comparisons  
between morpheme types regardless of fluency - which is not  
informative. Is there a way to do comparisons of morpheme types for  
each fluency group within this full model? And, in addition to it, do  
comparisons of fluency groups for each morpheme type, all within one  
model?

Marina

Quoting Andy Fugard <andyfugard@gmail.com>:

> On Mon, Jan 17, 2011 at 5:46 PM, Marina Sherkina-Lieber <
> marina.cherkina@utoronto.ca> wrote:
>
>> Thank you, Andy!
>> What is the difference between lmer and glmer?
>>
>
> From ?glmer
>
> "The lmer and glmer functions are nearly interchangeable. If lmer is called
> with a non-default family argument the call is replaced by a call to
> glmerwith the current arguments. If
> glmer is called with the default family, namely the
> gaussian<http://127.0.0.1:21832/library/stats/html/family.html>family
> with the identity link, then the call is replaced by a call to
> lmer with the current arguments. (They are described as “nearly”
> interchangeable because the REML argument only applies to calls to lmer and
> the nAGQ argument only applies to calls to glmer.)"
>
>
>> And it would help if you decipher parts of the command for the Tukey test.
>>
>
> There's a bunch of examples in ?glht which might be of assistance.  Copied
> here (using aov but same syntax):
>
> ### multiple comparison procedures
>
>   ### set up a one-way ANOVA
>   amod <- aov(breaks ~ tension, data = warpbreaks)
>
>   ### set up all-pair comparisons for factor `tension'
>   ### using a symbolic description (`type' argument
>   ### to `contrMat()')
>   glht(amod, linfct = mcp(tension = "Tukey"))
>
>   ### alternatively, describe differences symbolically
>   glht(amod, linfct = mcp(tension = c("M - L = 0",
>                                       "H - L = 0",
>                                       "H - M = 0")))
>
>   ### alternatively, define contrast matrix directly
>   contr <- rbind("M - L" = c(-1, 1, 0),
>                  "H - L" = c(-1, 0, 1),
>                  "H - M" = c(0, -1, 1))
>   glht(amod, linfct = mcp(tension = contr))
>
>
> The same degrees of freedom problem for the t-tests which comes up in
> Gaussian lmer models also applies (multiply...) when doing pairwise
> comparisons...
>
> Now, does anyone know how to do this with 95% HPD intervals... expanding
> them as necessary...?
>
> Andy
>
>
>
>
>> Marina
>>
>>
>> Quoting Andy Fugard <andyfugard@gmail.com>:
>>
>>  On Wed, Jan 12, 2011 at 6:06 PM, Maureen Gillespie <
>>> gillespie.maureen@gmail.com> wrote:
>>>
>>>  What if you DO want to do a bunch of paired tests and your coding scheme
>>>> isn't going to get at them all in one go? What is the best way of doing
>>>> this? Run multiple models on subsets of the data, then adjust/correct for
>>>> multiple comparisons?
>>>>
>>>
>>>
>>>
>>> You could use Tukey's all-pairwise comparisons via glht in the multcomp
>>> package.  For instance:
>>>
>>>
>>>> require(languageR)
>>>> require(multcomp)
>>>>
>>>>
>>>> M1 = glmer(CaseMarking ~ WordOrder * AgeGroup +
>>>>
>>> +   AnimacyOfSubject + Text + (1|Speaker),
>>> +   family = "binomial", data = warlpiri)
>>>
>>>> M1
>>>>
>>> Generalized linear mixed model fit by the Laplace approximation
>>> Formula: CaseMarking ~ WordOrder * AgeGroup + AnimacyOfSubject + Text +
>>> (1 | Speaker)
>>>   Data: warlpiri
>>>   AIC BIC logLik deviance
>>>  296.2 327 -140.1    280.2
>>> Random effects:
>>>  Groups  Name        Variance Std.Dev.
>>>  Speaker (Intercept) 0.46023  0.6784
>>> Number of obs: 347, groups: Speaker, 27
>>>
>>> Fixed effects:
>>>                                     Estimate Std. Error z value Pr(>|z|)
>>>
>>> (Intercept)                           -2.7731     0.4659  -5.952 2.65e-09
>>> ***
>>> WordOrdersubNotInitial                 0.2731     0.5025   0.543   0.5868
>>>
>>> AgeGroupchild                          1.2059     0.4726   2.552   0.0107
>>> *
>>>
>>> AnimacyOfSubjectinanimate              0.6406     0.3822   1.676   0.0938
>>> .
>>>
>>> Texttextb                              0.2887     0.4833   0.597   0.5503
>>>
>>> Texttextc                              0.7376     0.4237   1.741   0.0817
>>> .
>>>
>>> WordOrdersubNotInitial:AgeGroupchild  -1.8233     0.7382  -2.470   0.0135
>>> *
>>>
>>> ---
>>> Signif. codes:  0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1
>>>
>>> Correlation of Fixed Effects:
>>>            (Intr) WrdONI AgGrpc AnmcOS Txttxtb Txttxtc
>>> WrdOrdrsbNI -0.341
>>> AgeGropchld -0.594  0.371
>>> AnmcyOfSbjc -0.031 -0.052  0.041
>>> Texttextb   -0.478 -0.040 -0.039 -0.137
>>> Texttextc   -0.568 -0.033 -0.009 -0.311  0.606
>>> WrdOrdNI:AG  0.249 -0.674 -0.434 -0.069  0.053   0.026
>>>
>>>>
>>>> summary(glht(M1, linfct=mcp(Text="Tukey")))
>>>>
>>>
>>>         Simultaneous Tests for General Linear Hypotheses
>>>
>>> Multiple Comparisons of Means: Tukey Contrasts
>>>
>>>
>>> Fit: glmer(formula = CaseMarking ~ WordOrder * AgeGroup + AnimacyOfSubject
>>> +
>>>
>>>    Text + (1 | Speaker), data = warlpiri, family = "binomial")
>>>
>>> Linear Hypotheses:
>>>                   Estimate Std. Error z value Pr(>|z|)
>>> textb - texta == 0   0.2887     0.4833   0.597    0.821
>>> textc - texta == 0   0.7376     0.4237   1.741    0.188
>>> textc - textb == 0   0.4490     0.4063   1.105    0.509
>>> (Adjusted p values reported -- single-step method)
>>>
>>> Cheers,
>>>
>>> Andy
>>>
>>> --
>>> Dr Andy Fugard
>>> http://www.andyfugard.info
>>>
>>>
>>
>>
>> --
>> Marina Sherkina-Lieber
>> Ph.D. candidate
>> Dept. of Linguistics
>> University of Toronto
>>
>>
>
>
> --
> Dr Andy Fugard
> http://www.andyfugard.info
>

-- 
Marina Sherkina-Lieber
Ph.D. candidate
Dept. of Linguistics
University of Toronto