Statalist


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

Re: st: SV: Latent variable DVs in gllamm


From   Stas Kolenikov <skolenik@gmail.com>
To   statalist@hsphsun2.harvard.edu
Subject   Re: st: SV: Latent variable DVs in gllamm
Date   Wed, 11 Feb 2009 16:52:56 -0600

If you have three levels, then you would need more equations -- one
for each level. That's where your error message comes from. How you
want to incorporate other leves is up to you though -- as random
effects, random slopes, whether they affect the lower levels latent
variables -- pretty much any configuration might be defensible, just
look at what makes sense for your problem and in your discipline.

On 2/11/09, Rahsaan Maxwell <rahsaan@polsci.umass.edu> wrote:
> Thanks, this definitely helps.
>
>  Does it look like I have set it up correctly?
>
>  Because I have two questions.
>
>  1) If I run the following code I get this error message, which I am not sure I
>  understand:
>
>  1 equations specified: response, need 2
>
>  2) Also, if I want to run a multilevel model with variables that analyze both
>  the individual and the neighborhood level, is it correct to include the
>  neighborhood grouping variable 'neighid' as I have done below?
>
>  Thanks,
>
>  -Rahsaan
>
>  sort council courts police
>  gen id=_n
>  expand 3
>  sort id
>  qui by id: gen response=_n
>  gen item1=response==council
>  gen item2=response==courts
>  gen item3=response==police
>
>
>  eq response : item1 item2 item3
>
> eq latent : x1...x4
>  gllamm response, i(id neighid) eq(response) geq(latent) link(ologit)
>
>
>
>
>
>
>  Quoting Stas Kolenikov <skolenik@gmail.com>:
>
>  > You expand the data to create a single variable, let's call it
>  > -response-. For each original observation identified by variable
>  > called -id-, then you will need to create three dummy variables, say
>  > -item1-, -item2- and -item3-, to keep track of where that response
>  > came from. Then your model will essentially look like
>  >
>  > eq response : item1 item2 item3
>  > eq latent : all regressors for the latent variable
>  >
>  > gllamm response, id( id ) eq( response ) geq( latent ) link() family()
>  >
>  > The -eq- options says what is the pattern of the loadings of the
>  > latent variable on the observed ones, and -geq- says what is the
>  > structural model for that latent variable -- in your case, your
>  > regression.
>  >
>  > On 2/11/09, Rahsaan Maxwell <rahsaan@polsci.umass.edu> wrote:
>  > > Thanks for the response.  I see that I need to expand the data but I'm not
>  > sure
>  > >  how to then combine the three observed variables into one multivariate
>  > >  response?
>  > >
>  > >  From what I can tell, section 8.4 in the gllamm manual is the closest to
>  > my
>  > >  situation but it looks like they are regressing y on the different items
>  > that
>  > >  comprise the latent variable, whereas I want to be regressing the items of
>  > the
>  > >  latent variable on other variables.  Or is that a necessary first step to
>  > get
>  > >  the factor loadings?  In which case how do I define y?  That doesn't seem
>  > to be
>  > >  addressed in the manual.
>  > >
>  > >  Thanks,
>  > >
>  > >
>  > >  -Rahsaan
>  > >
>  > >
>  > >  Quoting Mads Meier Jæger <Mads@sfi.dk>:
>  > >
>  > >  > Rahsaan,
>  > >  >
>  > >  > This can be done in gllamm by expanding the data so that the three
>  > observed
>  > >  > variables (trust in police, etc.) form a multivariate response for each
>  > >  > individual, and then by using a random effect to model the latent DV.
>  > You
>  > >  > need to expand the data to "trick" gllamm into treating the three
>  > observed
>  > >  > variables as a multivariate response, see chapter 4 and 8 in the gllamm
>  > >  > manual. It should be easy enough also to include more random effects to
>  > >  > account for additional multilevel (neighbourhood, etc.) clustering. You
>  > could
>  > >  > also use eqs() to model covariate effects on the latent DV.
>  > >  >
>  > >  > Mads
>  > >  >
>  > >  >
>  > >  >
>  > >  > -----Oprindelig meddelelse-----
>  > >  > Fra: owner-statalist@hsphsun2.harvard.edu
>  > >  > [mailto:owner-statalist@hsphsun2.harvard.edu] På vegne af Rahsaan
>  > Maxwell
>  > >  > Sendt: 11. februar 2009 06:49
>  > >  > Til: statalist@hsphsun2.harvard.edu
>  > >  > Emne: st: Latent variable DVs in gllamm
>  > >  >
>  > >  > Does anyone know if it is possible to build a multi-level mixed effects
>  > model
>  > >  > with a latent DV using glamm?
>  > >  >
>  > >  > I have a unobserved response variable (political trust) that is
>  > comprised of
>  > >  > three observed variables (trust in police, trust in government, trust in
>  > >  > politicians).  I am trying to run a multi-level model with fixed effects
>  > IVs
>  > >  > at
>  > >  > the individual and neighborhood level and a random intercept for the
>  > >  > neighborhood grouping level.
>  > >  >
>  > >  > However, I cannot figure out how to construct the DV as a latent
>  > variable.  I
>  > >  > have been trying to use the eqs function but that does not seem to work.
>  > >  >
>  > >  > Thanks,
>  > >  >
>  > >  > -Rahsaan
>  > >  >
>  > >  >
>  > >  > Rahsaan Maxwell, Ph.D.
>  > >  > Assistant Professor
>  > >  > Department of Political Science
>  > >  > University of Massachusetts, Amherst
>  > >  >
>  > >  > Postdoctoral Fellow
>  > >  > Transatlantic Academy
>  > >  > German Marshall Fund of the United States
>  > >  > http://rahsaanmaxwell.googlepages.com
>  > >  > *
>  > >  > *   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/
>  > >  >
>  > >
>  > >
>  > >  Rahsaan Maxwell, Ph.D.
>  > >  Assistant Professor
>  > >  Department of Political Science
>  > >  University of Massachusetts, Amherst
>  > >
>  > >  Postdoctoral Fellow
>  > >  Transatlantic Academy
>  > >  German Marshall Fund of the United States
>  > >  http://rahsaanmaxwell.googlepages.com
>  > >  *
>  > >  *   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/
>  > >
>  >
>  >
>  > --
>  > 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/
>  >
>
>
>  Rahsaan Maxwell, Ph.D.
>  Assistant Professor
>  Department of Political Science
>  University of Massachusetts, Amherst
>
>  Postdoctoral Fellow
>  Transatlantic Academy
>  German Marshall Fund of the United States
>  http://rahsaanmaxwell.googlepages.com
>  *
>  *   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/
>


-- 
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/



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