Re: st: Weights in Multi-Level Models

[Stas Kolenikov <skolenik@gmail.com>]

If producing a representative sample of the communities was not an
object of sampling to begin with, you will likely get a disbalanced
sample of these, at best; it will probably be representative, but this
would be a rather ugly sample.

You can supply either level-1 or level-2 or both level-1 and level-2
weights to -gllamm-. (You would have to supply both with -weight()-
option, but you can set one or the other to 1 if needed.)

Depending on how exactly the ethnic minority was oversampled, you may
or may not need the community level weights. If you sampled every 5h
ethnic community and every 20th majority community, and equal
proportion of people in both types, you need the community weights. If
you sampled communities at the same rate, but then sampled at a higher
rate within the ethnic ones compared to the majority ones, then you
only need the individual weights (and you will have very few degrees
of freedom/low accuracy on the estimates of the ethnic community
variables).

On Mon, Nov 1, 2010 at 5:37 PM, James Laurence
<james.laurence@sjc.ox.ac.uk> wrote:
> Dear Statalist,
>
> I was wondering if anyone could shed some light on a dilemna I am having. I've had a good search around but haven't found an answer and don't know anyone with any expertise in multi-level modelling.
>
> Basically, I have a data set that includes both a general population core sample and an ethnic booster sample. I also have neighbourhood identifiers in which respondents are clustered. However, the sampling was not done with multi-level modelling in mind.
>
> I want to run a multi-level analysis of neighbourhood effects.
>
> The problem is I need to include weights in these models because of the ethnic booster sample, but as far as I can tell you need two sets of weights to run multi-levels in STATA. One at the individual level and one at the community level. However, my dataset does not contain these.
>
> Is there any way of running a multi-level model with only one set of weights? Would this produce substantially biased results? Do you have to have two sets of weights to run a MLM with weights? Is there a method for calculating level 2 weights?



-- 
Stas Kolenikov, also found at http://stas.kolenikov.name
Small print: I use this email account for mailing lists only.

*
*   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/