Bookmark and Share

Notice: On April 23, 2014, Statalist moved from an email list to a forum, based at

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

Re: st: providing raw weights for multivariate meta-analysis

From   Nick Darson <>
Subject   Re: st: providing raw weights for multivariate meta-analysis
Date   Mon, 11 Jul 2011 06:54:01 +1000

Thanks a lot for the info and the links.

I have not looked much into gllamm yet, but the description of Stata
12 xtmixed looks very promising for the weighting issue (I guess not
being able to provide specific weights for each level was the problem
to which Hox 2010 referred).

With regard to the "non-normalized" weighting issue, I found in the
mean-time the following link, which I wanted to share:

The site states that "pweight does not automatically normalize the
sample weight like aweight does.  Stata's survey commands do not allow

Note:  A normalized sample weight sums to the number of observations
in the data set and its mean is 1. "

Hence, it looks to me as if I can simply use "pweights", providing the
effect size variances to the first level as weights (and pweights
provides the inverse weight of this variable, non-normalized). Then,
restraining the first level variance to 1 and the multivariate model
should work (I hope).

I might try estimating such a model in Stata 12 and compare the
results with those of HLM using the V-known function (eventually, I
want to use Stata as I would like to add another level - HLM only
allows three levels when using the V-known function).

On Mon, Jul 11, 2011 at 4:02 AM, Austin Nichols <> wrote:
> Nick Darson <>:
> Have you checked out -gllamm- yet (on SSC and at
> Official commands in Stata 11 does not allow different weights at
> different levels, but -xtmixed- in Stata 12 will:
> My own inclination is not to impose normality using RE/xtmixed but to
> run a fixed effects model instead, as it requires weaker assumptions.
> You can absorb one dimension of fixed effects with -areg- if you don't
> want to include a large number of dummies.
> In that case, you can multiply your precision weights times any other
> weights, rescale, and specify the product as pweights.
> The point estimates should be identical whether specifying pweights or
> aweights; only SEs differ (pw=aw+_robust).
> See also
> and look up Sean Reardon's -hlm- package on SSC.
> On Sun, Jul 10, 2011 at 1:43 PM, Nick Darson <> wrote:
>> Thanks for your help, Stas.
>> The book I referred to is: Joop J. Hox 2010: Multilevel analysis:
>> techniques and applications, 2nd edition, Routledge
>> To my understanding, "aweights" normalizes the weights (such that the
>> sum of the weights =N, the number of observations).
>> However, it is important for the multivariate meta-analysis approach
>> that the weights are not normalized (so called precision weights or
>> raw weights).
>> This is one of the two mentioned requirements by Hox (2010). The
>> second requirement for multivariate meta-analyses is the constraint of
>> the lowest-level variance to 1 - what is given in Stata - please
>> correct me if I am wrong on this).
>> Therefore, I was wondering whether the weight issue is the reason that
>> Stata cannot be used for multivariate meta-analyses.
>> Cheers,
>> Nick
>> On Mon, Jul 11, 2011 at 3:12 AM, Stas Kolenikov <> wrote:
>>> I personally think that -help weights- explains it all. Variance
>>> weights are aweights: they state that the measurement error in a given
>>> observation is such and such. The pweights are probability weights,
>>> and when you invoke these, the calculation of not only the point
>>> estimates, but also the standard errors is different (according to
>>> probability sampling theory rather than the likelihood theory). There
>>> are also frequency weights, which is essentially the result of
>>> -collapse (count) ... , by(*)-. The iweights are "all other weights,
>>> whichever these may mean for you" -- there is no strict definition,
>>> and the program that allows them usually has its own idea what to do
>>> with them.
>>> Knowing very little about meta-analysis beyond the fancy name, I have
>>> a feeling that you should be looking towards -gllamm- rather than
>>> -xtmixed- to find the flexibility with the weights that you need. You
>>> might still want to run (the much faster) -xtmixed- first to get good
>>> starting values that you'd feed to -gllamm-. Joop Hox (a proper
>>> reference is in place, according to the rules of Statalist; I have a
>>> vague idea that this is a famous multilevel author) is a somewhat
>>> opinionated guy, to my impression (although I am much more opinionated
>>> than he is :)).
>>> On Sun, Jul 10, 2011 at 7:02 AM, Nick Darson <> wrote:
>>>> Dear Statalisters,
>>>> I attempt to carry out a multivariate meta-analysis using "xtmixed". I
>>>> have three outcome measures (the first level).
>>>> When using a multilevel approach for the multivariate meta-analysis, I
>>>> read that it is important to provide raw-weights (non-normalized) of
>>>> the inverse sampling variances for the first level (containing the
>>>> outcome measures).
>>>> I know that HLM6 has a special function for this ("V-known") for two
>>>> and three level models.
>>>> Now, I wanted to know whether "pweights" is an equivalent solution for
>>>> xtmixed in Stata?
>>>> I must admit that I am a bit confused by the various descriptions of
>>>> the weight functions I found online (aweights, iweights, etc - though
>>>> this are not allowed for xtmixed) .
>>>> Moreover, surprisingly,  I read in Hox (2010,p. 230) that public
>>>> domain software for multilevel analysis does not support the required
>>>> options for a multivariate model so far (i.e. providing raw weights
>>>> and being able to constrain the lowest-level variance to 1 - however,
>>>> both should be feasible).
> *
> *   For searches and help try:
> *
> *
> *

*   For searches and help try:

© Copyright 1996–2018 StataCorp LLC   |   Terms of use   |   Privacy   |   Contact us   |   Site index