Re: st: multilevel logistic model on aggregated data

[Joerg Luedicke <joerg.luedicke@gmail.com>]

So how exactly is your data structured? Do you have classes nested in
schools nested in countries? Or was the original data aggregated on
the class level and then aggregated again on the school level so that
you have only schools nested in countries? (Is it really impossible to
get the original data?) And: does "using weights for the numerosity of
the class" mean that you think about including class size as an
offset? If you only have your dependent measure in aggregated form,
i.e. the smoking rate, you would not need an offset as the class size
was already used when calculating the rate (e.g., 10 students in a
class of size 20 smoke => 10/20=0.5). You could use class size as a
regular covariate if you had a substantial hypothesis about it, for
example, that smoking rates increase with increasing class sizes. But
that would be a different thing and would require that you actually
have class level data.

Joerg

On Tue, Nov 22, 2011 at 9:10 AM, Maria Paola Caria
<Maria.Paola.Caria@ki.se> wrote:
> Thanks Joerg.
> I did not explain well.
> the original outcome is individual binary (say, current smoking  yes or not).
> but I do not have original data, only aggregated data. so the outcome is now the proportion of smoking students in the class. I think I should model this proportion with a logistic model (since the underlying distribution is binomial) using weights for the numerosity of the class. but then  I want to add the random intercept.
>
>
>
>
> Maria Paola Caria
> PhD student
> Dept. of Public Health Sciences
> Division of Public Health Epidemiology
> Karolinska Institutet
> SE-11891 Stockholm, Sweden
> E-mail: Maria.Paola.Caria@ki.se
> ________________________________________
> From: owner-statalist@hsphsun2.harvard.edu [owner-statalist@hsphsun2.harvard.edu] on behalf of Joerg Luedicke [joerg.luedicke@gmail.com]
> Sent: 22 November 2011 14:58
> To: statalist@hsphsun2.harvard.edu
> Subject: Re: st: multilevel logistic model on aggregated data
>
> I am not sure if I understand the problem correctly and maybe I am
> missing something, but if you do not have any individual level data
> you can only look at the variability of schools across countries as,
> for example, in:
>
> xtmelogit binaryoutcome x1  x2  || country:, or
>
>
> Joerg
>
>
> On Tue, Nov 22, 2011 at 4:58 AM, Maria Paola Caria
> <Maria.Paola.Caria@ki.se> wrote:
>> I need to fit logistic models on data aggregated at the school level.
>> Since there is a hierarchy in the data with schools clustered in regions,
>>  I need to add a random intercept  at the region level.
>> To clarify, If I had data at the individual level my model would be:
>>
>> xtmelogit binaryoutcome x1  x2  || country: || school: , or
>>
>> gllamm binaryoutcome x1 x2, i(school country) eform  family(binomial) link(logit)
>>
>> but data are aggregated at the school level.
>> Is there a command to fit multilevel logistic model on aggregated data in StataSE 12 ?
>> Any help is appreciated.
>> Thanks.
>> Maria Paola
>>
>>
>>
>>
>>
>>
>>
>>
>>
>> Maria Paola Caria
>>
>> PhD student
>>
>> Dept. of Public Health Sciences
>>
>> Division of Public Health Epidemiology
>>
>> Karolinska Institutet
>>
>> SE-11891 Stockholm, Sweden
>>
>> *
>> *   For searches and help try:
>> *   http://www.stata.com/help.cgi?search
>> *   http://www.stata.com/support/statalist/faq
>> *   http://www.ats.ucla.edu/stat/stata/
>>
>
> *
> *   For searches and help try:
> *   http://www.stata.com/help.cgi?search
> *   http://www.stata.com/support/statalist/faq
> *   http://www.ats.ucla.edu/stat/stata/
>
> *
> *   For searches and help try:
> *   http://www.stata.com/help.cgi?search
> *   http://www.stata.com/support/statalist/faq
> *   http://www.ats.ucla.edu/stat/stata/
>

*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/