Bookmark and Share

Notice: On March 31, it was announced that Statalist is moving from an email list to a forum. The old list will shut down at the end of May, and its replacement, statalist.org is already up and running.


[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: st: 3-level model for crime victimization


From   "JVerkuilen (Gmail)" <jvverkuilen@gmail.com>
To   statalist@hsphsun2.harvard.edu
Subject   Re: st: 3-level model for crime victimization
Date   Tue, 25 Sep 2012 20:16:12 -0400

Oh, whoops, I thought that the neighborhood was a level 3 unit with
some sub-district being a level 2! Yes if there isn't a level 3
because it's one city, there's nothing to fit.

Yes it's definitely a small sample for a multilevel model, but not out
of bounds of reality assuming the sample sizes are reasonably equal.
I've run fully Bayesian models and ML models with modest higher level
DF and have had quite reasonable results for random intercepts.
However it's quite likely to fail with many level 2 covariates.

Clustered robust or clustered bootstrapping is probably the most
straightforward thing to do, with a dummy for each district.



On Tue, Sep 25, 2012 at 11:59 AM, Joerg Luedicke
<joerg.luedicke@gmail.com> wrote:
> If there are 8 neighborhoods in 1 city then the level-3 sample size
> would be 1! As a result you would throw a bunch of constants in your
> model which does not make sense. Besides, a level-2 sample size of 8
> is very small (at least for the standard likelihood based methods) and
> is usually problematic with respect to point and variance estimates of
> the fixed effects as well as the varying intercept/coefficient
> variation. I believe this is discussed in any multilevel textbook.
>
> Joerg
>
> On Tue, Sep 25, 2012 at 10:45 AM, JVerkuilen (Gmail)
> <jvverkuilen@gmail.com> wrote:
>> On Tue, Sep 25, 2012 at 9:41 AM, Erik Alda <erik.alda@gmail.com> wrote:
>>> Dear Statalisters
>>>
>>> I am running a, so far, 2 level model on crime victimization using a survey
>>> of 8 neighborhoods in a large city in Brazil. The firs level is individual
>>> level predictors and the second level is neighborhood level predictors.
>>> Since the survey is not representative of the whole municipality, would I be
>>> able to include a 3rd level with citywide variables to estimate the effect
>>> of policy?
>>
>> I don't see why you can't include it though your level 3 N is somewhat
>> small. As long as the sample sizes per neighborhood aren't reasonably
>> even it should run OK, though. The fact that the survey isn't
>> representative is a study limitation and affects what conclusions you
>> can draw from it, but that's separate from whether there's level 3
>> heterogeneity in the sample you do have. In some cases having the
>> external generalizability of a a representative sample is mandatory.
>> In other case it's not.
>>
>>
>> --
>> JVVerkuilen, PhD
>> jvverkuilen@gmail.com
>>
>> "Out beyond ideas of wrong-doing and right-doing there is a field.
>> I'll meet you there. When the soul lies down in that grass the world
>> is too full to talk about." ---Rumi
>> *
>> *   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/



-- 
JVVerkuilen, PhD
jvverkuilen@gmail.com

"Out beyond ideas of wrong-doing and right-doing there is a field.
I'll meet you there. When the soul lies down in that grass the world
is too full to talk about." ---Rumi
*
*   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/


© Copyright 1996–2014 StataCorp LP   |   Terms of use   |   Privacy   |   Contact us   |   Site index