[R-sig-ME] LMER vs MLwiN

[W Robert Long]

Thanks, but sadly I don't have access to Raudenbush and Bryk book.
I need to write about this, so a reference would be appreciated, but I'm 
also interested to see/show it in my data.


On 17/02/2012 7:55 PM, Doran, Harold wrote:
> Raudenbush and Bryk do discuss this in their book if you require a text. But, it is quite easy to show. At one point, there was an example of how to do this using mcmcsamp() in the lme4 package (I think). But, I don't see the lattice plots in that help page now.
>
>> -----Original Message-----
>> From: r-sig-mixed-models-bounces at r-project.org [mailto:r-sig-mixed-models-
>> bounces at r-project.org] On Behalf Of W Robert Long
>> Sent: Friday, February 17, 2012 2:33 PM
>> To: r-sig-mixed-models at r-project.org; Douglas Bates
>> Subject: Re: [R-sig-ME] LMER vs MLwiN
>>
>>
>> On 17/02/2012 4:37 PM, Douglas Bates wrote:
>> <snip>
>>>> The estimates look pretty close, but the standard errors for the REs are
>>>> quite different - I seem to remember the sampling variance of REs has a
>>>> skewed distribution, but I don't know if this has anything to do with it ?
>>>
>>> Those are not standard errors in the glmer output.  They are simply
>>> the variance estimates on the standard deviation scale (i.e. 0.15567 =
>>> sqrt(0.024233)).  The reason that glmer does not provide a standard
>>> error for an estimate of a variance component is because they don't
>>> make sense in most cases.  The distribution of the estimator is highly
>>> skewed.
>>>
>> <snip>
>>
>> Thank you for that. Could you provide a reference for this latter point
>> ? I have a copy of the Pinheiro and Bates (2000) book available in our
>> library, if it's in there ? Otherwise, a published paper would be also
>> be fine.
>>
>> _______________________________________________
>> R-sig-mixed-models at r-project.org mailing list
>> https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models