Re: st: Please advise- Models using gllamm

[Steve Samuels <sjsamuels@gmail.com>]

I agree with the previous comments, but I do have some advice for
Daniel.

• There is a clear example in the -gllamm- help on how to specify random
slopes and intercepts in a two-level model.

• The proposed -gllamm- statements ignore level 2 ("agency") sampling
weights. Both -gllamm- and -xtmixed- have an option to scale the level 2
and level 1 weights. The scaling is discussed at length in the manual
entry for -xtmixed- (section on Survey Data). Failure to scale weights
can lead to biased estimates for standard deviations of agency effects
(Asparouhov, 2006). Exactly which weight to choose is, in the words of
the -xtmixed- manual, a "tough call". Given your choice, -xtmixed- will
do the scaling for you, but you must scale yourself if you use -gllamm-.
See the UNC page referenced below for a Stata command that will help.

• The -gllamm- models you showed can't be defended on even more basic
grounds: How do you know that the effects of continuous variables
are strictly linear; that there are no interactions; and that standard
deviations of random effects at each level are constant?

(On the last point there are some good suggestions in the UCLA and
Gutierrez references below.)

You haven't showed us your -xmixed- models, but I suspect that the same
criticisms apply. In short, your analyses are seriously flawed. If you
are planning to publish your findings, don't, at least not until you've done
a better job.


Steve
sjsamuels@gmail.com


References

Asparouhov T: General Multi-Level Modeling with Sampling
Weights. Communications in statistics. Theory and methods 2006,
35(3):439-460.

www.ats.ucla.edu/stat/stata/faq/hetero_resid_var.htm
www.stata.com/meeting/fnasug08/gutierrez.pdf
http://www.cpc.unc.edu/research/tools/data_analysis/ml_sampling_weights


On Mar 17, 2013, at 11:47 PM, Richard Williams wrote:

I didn't read the original post carefully. I do think that mere mortals might have a chance if they carefully read Stata's SEM Manual. But then again I worked with LISREL 30+ years ago when everything had to be written with matrices, so anything written in something that vaguely resembles English seems a bit more manageable.

At 08:52 PM 3/17/2013, Stas Kolenikov wrote:
> If a past president of the American Statistical Association says so,
> there should be some merit in his words. Models fitted by -gllamm- are
> surely fancy, but you need (i) a serious (matrix based) course in
> regression; (ii) a serious course in generalized linear models; (iii)
> a serious course in mixed models, and (iv) a serious (multivariate
> calculus based) course in nonlinear optimization -- before you can
> meaningfully work with -gllamm-. Or, for that matter, a lot of other
> advanced models like -sem- or -cmp- or -xtmixed- or -xtmelogit-. If
> all you had was a course in ANOVA, you might be able to find a
> consultant who could do the work for you for $150/hour (if you are
> lucky), and you won't own it. This may be OK for an urgent work if
> your student visa expires in two weeks, but I totally agree with Tony
> that it hardly makes for good academics.
> 
> -- Stas Kolenikov, PhD, PStat (SSC)
> -- Senior Survey Statistician, Abt SRBI
> -- Opinions stated in this email are mine only, and do not reflect the
> position of my employer
> -- http://stas.kolenikov.name
> 
> 
> 
> On Sun, Mar 17, 2013 at 5:38 PM, Lachenbruch, Peter
> <Peter.Lachenbruch@oregonstate.edu> wrote:
> > I was thinking more on the line of the accused in a murder trial of his parents asking for consideration of the fact that he was an orphan.
> >
> > This a hell of  a nerve.  It does not augur well for a future in academics.  As simply as i can put it, don't use tools you haven't studied.  Tell your advisor to either give you time to learn about gllamm or allow you to use other tools.
> >
> > Peter A. Lachenbruch,
> > Professor (retired)
> > ________________________________________
> > From: owner-statalist@hsphsun2.harvard.edu [owner-statalist@hsphsun2.harvard.edu] on behalf of JVerkuilen (Gmail) [jvverkuilen@gmail.com]
> > Sent: Sunday, March 17, 2013 1:17 PM
> > To: statalist@hsphsun2.harvard.edu
> > Subject: Re: st: Please advise- Models using gllamm
> >
> > On Sun, Mar 17, 2013 at 2:14 PM, Daniel Kaplan <dbk2006@gmail.com> wrote:
> >> Hi Statalisters-
> >>
> >> I am very quickly running out of time to finish this last step in my
> >> dissertation, and I am not only new to complex stats but have never
> >> used gllamm.  PLEASE someone take a look at my gllamm models and give
> >> me some insight as to how to proceed.
> >>
> >> This is a re-post of my earlier request, as I am desperate for advice
> >> and nobody responded as of yet.  I know there is a lot to read here,
> >> but I figure the more details I can offer the better.
> >>
> >
> > While I feel for you---and you need to realize that your thesis
> > advisor has utterly screwed you, too---there's no way that any of us
> > can meaningfully help. If I may make an analogy, your request is like
> > walking into an acute care center with a gunshot wound to the chest.
> >
> *
> *   For searches and help try:
> *   http://www.stata.com/help.cgi?search
> *   http://www.stata.com/support/faqs/resources/statalist-faq/
> *   http://www.ats.ucla.edu/stat/stata/

-------------------------------------------
Richard Williams, Notre Dame Dept of Sociology
OFFICE: (574)631-6668, (574)631-6463
HOME:   (574)289-5227
EMAIL:  Richard.A.Williams.5@ND.Edu
WWW:    http://www.nd.edu/~rwilliam

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