Re: st: Gllamm: Convergence not achieved: try with more quadrature points

[Kjetil Arne Van Der Wel <kjetil.wel@sam.hio.no>]

Dear Steve,
Many thanks for this. Sorry for describing the analysis so poorly. I did 
actually run the command with pweight (I tried both methods), but that 
did not change the situation. Also, each country within the survey 
provides a representative sample of its population. Together, the 
countries constitutes EU-25, and hence, do not represent a greater 
*population' of countries. Furthermore, the countries are not merely 
administrative borders, but political entities employing different 
policies that have profound impact on the lives of their inhabitants. 
Because of this, I feel it is correct to treat them as 'units' in which 
individuals (and regions) are nested (i.e. appropriate entities of 
variation that can be described with variables) rather than 'strata' 
(which I take to mean as constructed by the researcher, such as social 
class). Sorry if I completely misunderstood you, Steve and Stas.

To give you a better idea of the problem: I am fitting a multilevel 
model to look at the effects of country-level variables on 
individual-level outcomes, as well as interaction effects between the 
country-level variables and individual characteristics (cross-level 
interactions). In the medical sociological and epidemiological 
literature, it is not unusual to publish results from multilevel 
analysis with as little 15-25 level 2 units (which are often European 
countries). Most analysts within this tradition use MLwiN. I first tried 
xtlogit to do my analysis (which works well), but it does not allow 
pweights or other weights that vary within level 2 units. This is why I 
wanted to have a go with gllamm, which seemed to be more flexible in 
regard to the weighting problem.

Running this command:

gllamm depvar indepvar1-4, i(country) pweight(wt) link(logit)

fam(binom) nip(20) adapt

leads to the error message, while omitting the pweight works fine. (By 
the way; depvar and all indepvars are binary individual level variables, 
except age. in the next model, continous counry level variables will be 
added).

If it does not work, I could report unweighted results from xtlogit:

xtlogit depvar age indepvar1-4(i) indepvar5-7(j) interactions 
terms(i*j), i(country) quad(20)

N(i)=188 000
N(j)=25

Is it possible to change the dataset according to the weights in advance 
of analysis, which could be a way around the problem?


Best,
Kjetil

On 21.09.2010 16:11, Steve Samuels wrote:
> --
> The -weight()- option in -gllamm- is intended to indicate frequency
> weights. Use the -pweight()- option for probabilities. (It might make
> no difference). I'm not expert in -gllamm-, but n = 27 countries looks
> like a small number to me, especially if you have  country level
> covariates. Please, as the FAQ request, show the actual command that
> you issued.
>
> As Stas states, countries are strata;  but you treat them as random.
> You (probably) have enough observations to include country-specific
> intercepts and slopes, which you can summarize as you please without
> GLLAMM's strong assumptions of normal distributions and spatial
> independence. You state that the design is not "complex"; I take this
> to mean that individuals were selected within countries without
> intermediate stages. Is this correct? In any case, a survey analysis
> (-svyset-, followed by -svy: logistic- or -svy: logit-) might do all
> that you need.
>
>
> Steve
>
> Steven J. Samuels
> sjsamuels@gmail.com
> 18 Cantine's Island
> Saugerties NY 12477
> USA
> Voice: 845-246-0774
> Fax:    206-202-4783
>
> 2010/9/21 Kjetil Arne Van Der Wel<kjetil.wel@sam.hio.no>:
> > Dear all,
> > I am trying to run a logistic random intercept model with gllamm, using
> > sample weights. Whenever I include the pweight(wt) option I get the above
> > error message, but without the weights it works fine. I have tried
> > increasing nip (20, 30, 40), but it has been futile.
> >
> > Level 2 units are 25 countries in Europe. Because they are not a sample, wt2
> > is set equal to 1. The level 1 sample weight (wt1), that corrects for
> > different sampling probabilities for individuals within different countries
> > (as well as country specific attrition), is entered
> > as specified by data supplier (non-integer, four digits or more).  The
> > weight was not scaled (as is otherwise suggested by the literature on
> > multilevel modelling of complex survey data (e.g. Rabe-Hesketh&  Skrondal,
> > 2006)), because the sampling procedure was not ‘complex’,
> > i.e. level 2 units do not constitute a sample. Is this correct?
> >
> > This is the gllamm command used:
> > gllamm depvar indepvar1 indepvar2, i(country) weight(wt) link(logit)
> > fam(binom) nip(20) iterate(20) adapt
> >
> > My questions for you are:
> > 1. Why do I get the error message and how to fix it?
> > 2. Is my specification of the weights correct?
> >
> > Thanks,
> > Kjetil van der Wel, PhD-student, Oslo University College
> >
> >
> >
> > *
> > *   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/

--
Kjetil van der Wel
Stipendiat/PhD student
Gruppe for inkluderende velferd (GIV) / Research Group for Inclusive 
Social Welfare Policies,
Avdeling for samfunnsfag (SAM) / Faculty of social sciences
Høgskolen i Oslo / Oslo University College

Rom: SG304 (Stensberggata 29)
tlf/phone: +47 22 45 35 76 / +47 906 41 606
http://www.hio.no/giv

Postadresse/ postal address:
SAM
Høgskolen i Oslo
Postboks 4, St. Olavs plass
0130 Oslo


*
*   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/