Bookmark and Share

Notice: On March 31, it was announced that Statalist is moving from an email list to a forum. The old list will shut down on April 23, and its replacement, statalist.org is already up and running.


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

Re: st: is gllamm appropriate? is it necessary?-more information


From   Partha Deb <partha.deb@hunter.cuny.edu>
To   statalist@hsphsun2.harvard.edu
Subject   Re: st: is gllamm appropriate? is it necessary?-more information
Date   Wed, 17 Mar 2010 11:18:04 -0400

I would definitely start with -mlogit- . -glamm- is not only overkill here, it may not be giving you what you want (I'm not particularly familiar with -glamm- syntax, however).

I would start with just the individual level characteristics - perhaps clustering standard errors at the county level. If you observe "noise" when you add county dummies, it's likely because some of the within-county cause of death frequencies are very small - so causing singularities. You might also consider grouping some small population counties.

Partha


Jessica Bishop-Royse wrote:
Cause of death is a nominal variable, and combining causes is
definitely on the list of things to do if I dont have enough
observations.

So I can, when modeling cause of death, put it in a model with only
the county level factors to determine which county level vars are
important, using mlogit?  what if I want to use individual level
variables like age and education?  I have tried mlogit before doing
this (combining individual and county level vars and I get a lot of
noise).

jcbr

On 3/17/10, Partha Deb <partha.deb@hunter.cuny.edu> wrote:
Jessica,

Is cause specific death just a multinomial (nominal) variable?  If it
is, and assuming that each of the 9 causes has a reasonable fraction of
observations (else consider combining some of the causes), why not
estimate a multinomial logit?

mlogit causeofdeath black, cluster(countynumber)

or

mlogit causeofdeath black i.countynumber

HTH

Partha



Jessica Bishop-Royse wrote:
you're right.  i am actually using counties in Florida which are 67.
Sorry for the confusion.

On 3/17/10, Nick Cox <n.j.cox@durham.ac.uk> wrote:

For extra context I guess wildly at what is not explicit here. Jessica
is using the counties of the US, which are about 3000 in number.

Please remember that this is an international list and that others may
not share your presumptions.

Nick
n.j.cox@durham.ac.uk

Jessica Bishop-Royse

Some more information about the project I am working on:

1.  I have two main research questions here.  One: What county-level
variables are associated with cause-specific death in 1980?  In 2000?
And Two: What are the net effects of individual characteristics and
county level  variables?  What are the changes from 1980 to 2000?

2.  I don't see county as a control variable- but rather as a cluster
variable.  In fact, I am not even really interested in county per but
rather the variables that I have for counties (like % minority, %
poverty, etc.)  Eventually I would like to make interaction effects
with county (ruralpoorminority counties versus ruralpoorwhite
counties, urbanpoorwhite counties, etc.)  Ideally, I would like to add
these variables to a model along with my individual level predictors.

3.  As of now, my cause of death variable is 10 categories, 9 causes
and survival. I would like the ability to model both ways (cause 1
versus all others and each cause versus survival).

Yesterday at about 2 pm, I set following command up to run.  It ran
all afternoon, and all night and still hadn't finished.  It went
through 80 iterations, most of which had the note "not concave" before
I finally canceled it this morning. I am sure that I am doing
something wrong and it makes me nervous because I haven't even added
all the predictors yet.

. gllamm causeofdeath black, i(countynumber)

Iteration 0:   log likelihood = -226514.74  (not concave)
Iteration 1:   log likelihood = -170460.09  (not concave)
Iteration 2:   log likelihood = -156165.69
Iteration 3:   log likelihood = -155236.77  (not concave)
Iteration 4:   log likelihood = -154882.46
Iteration 5:   log likelihood = -154869.14
Iteration 6:   log likelihood = -154848.88  (not concave)
Iteration 7:   log likelihood = -154814.85
Iteration 8:   log likelihood = -154814.29
Iteration 9:   log likelihood = -154814.23
Iteration 10:  log likelihood = -154814.23

number of level 1 units = 318493
number of level 2 units = 77

Condition Number = 2.4744641

gllamm model

log likelihood = -154814.23

------------------------------------------------------------------------
------
causeofdeath |      Coef.   Std. Err.      z    P>|z|     [95% Conf.
Interval]
-------------+----------------------------------------------------------
------
       black |    .024678   .0016439    15.01   0.000     .0214559
.0279
       _cons |    .024141    .001246    19.37   0.000     .0216988
.0265831
------------------------------------------------------------------------
------

Variance at level 1
------------------------------------------------------------------------
------

  .1547741 (.00038787)

Variances and covariances of random effects
------------------------------------------------------------------------
------


***level 2 (countynumber)

    var(1): .00002072 (8.730e-06)
------------------------------------------------------------------------
------



. gllamm causeofdeath black biryear, i(countynumber)

Iteration 0:   log likelihood = -226590.33  (not concave)
Iteration 1:   log likelihood = -170468.78  (not concave)
Iteration 2:   log likelihood = -154843.19  (not concave)
Iteration 3:   log likelihood = -154785.59  (not concave)
Iteration 4:   log likelihood = -154772.62  (not concave)
Iteration 5:   log likelihood = -154753.11
Iteration 6:   log likelihood = -154738.19
Iteration 7:   log likelihood = -154738.07  (not concave)
Iteration 8:   log likelihood = -154738.07  (not concave)
Iteration 9:   log likelihood = -154738.07  (not concave)
Iteration 10:  log likelihood = -154738.07  (not concave)
Iteration 11:  log likelihood = -154738.07  (not concave)
Iteration 12:  log likelihood = -154738.07  (not concave)
Iteration 13:  log likelihood = -154738.07  (not concave)
Iteration 14:  log likelihood = -154738.07  (not concave)
Iteration 15:  log likelihood = -154738.07  (not concave)
Iteration 16:  log likelihood = -154738.07  (not concave)
Iteration 17:  log likelihood = -154738.07  (not concave)
Iteration 18:  log likelihood = -154738.07  (not concave)
Iteration 19:  log likelihood = -154738.07  (not concave)
Iteration 20:  log likelihood = -154738.07  (not concave)
Iteration 21:  log likelihood = -154738.07  (not concave)
Iteration 22:  log likelihood = -154738.07  (not concave)
Iteration 23:  log likelihood = -154738.07  (not concave)
Iteration 24:  log likelihood = -154738.07  (not concave)
Iteration 25:  log likelihood = -154738.07  (not concave)
Iteration 26:  log likelihood = -154738.07  (not concave)
Iteration 27:  log likelihood = -154738.07  (not concave)
Iteration 28:  log likelihood = -154738.07  (not concave)
Iteration 29:  log likelihood = -154738.07  (not concave)
Iteration 30:  log likelihood = -154738.07  (not concave)
Iteration 31:  log likelihood = -154738.07  (not concave)
Iteration 32:  log likelihood = -154738.07  (not concave)
Iteration 33:  log likelihood = -154738.07  (not concave)
Iteration 34:  log likelihood = -154738.07  (not concave)
Iteration 35:  log likelihood = -154738.07  (not concave)
Iteration 36:  log likelihood = -154738.07  (not concave)
Iteration 37:  log likelihood = -154738.07  (not concave)
Iteration 38:  log likelihood = -154738.07  (not concave)
Iteration 39:  log likelihood = -154738.07  (not concave)
Iteration 40:  log likelihood = -154738.07  (not concave)
Iteration 41:  log likelihood = -154738.07  (not concave)
Iteration 42:  log likelihood = -154738.07  (not concave)
Iteration 43:  log likelihood = -154738.07  (not concave)
Iteration 44:  log likelihood = -154738.07  (not concave)
Iteration 45:  log likelihood = -154738.07  (not concave)
Iteration 46:  log likelihood = -154738.07  (not concave)
Iteration 47:  log likelihood = -154738.07  (not concave)
Iteration 48:  log likelihood = -154738.07  (not concave)
Iteration 49:  log likelihood = -154738.07  (not concave)
Iteration 50:  log likelihood = -154738.07  (not concave)
Iteration 51:  log likelihood = -154738.07  (not concave)
Iteration 52:  log likelihood = -154738.07  (not concave)
Iteration 53:  log likelihood = -154738.07  (not concave)
Iteration 54:  log likelihood = -154738.07  (not concave)
Iteration 55:  log likelihood = -154738.07  (not concave)
Iteration 56:  log likelihood = -154738.07  (not concave)
Iteration 57:  log likelihood = -154738.07  (not concave)
Iteration 58:  log likelihood = -154738.07  (not concave)
Iteration 59:  log likelihood = -154738.07  (not concave)
Iteration 60:  log likelihood = -154738.07  (not concave)
Iteration 61:  log likelihood = -154738.07  (not concave)
Iteration 62:  log likelihood = -154738.07  (not concave)
Iteration 63:  log likelihood = -154738.07  (not concave)
Iteration 64:  log likelihood = -154738.07  (not concave)
Iteration 65:  log likelihood = -154738.07  (not concave)
Iteration 66:  log likelihood = -154738.07  (not concave)
Iteration 67:  log likelihood = -154738.07  (not concave)
Iteration 68:  log likelihood = -154738.07  (not concave)
Iteration 69:  log likelihood = -154738.07  (not concave)
Iteration 70:  log likelihood = -154738.07  (not concave)
Iteration 71:  log likelihood = -154738.07  (not concave)
Iteration 72:  log likelihood = -154738.07  (not concave)
Iteration 73:  log likelihood = -154738.07  (not concave)
Iteration 74:  log likelihood = -154738.07  (not concave)
Iteration 75:  log likelihood = -154738.07  (not concave)
Iteration 76:  log likelihood = -154738.07  (not concave)
Iteration 77:  log likelihood = -154738.07  (not concave)
Iteration 78:  log likelihood = -154738.07  (not concave)
Iteration 79:  log likelihood = -154738.07  (not concave)
Iteration 80:  log likelihood = -154738.07  (not concave)
(Maximization aborted)

.
end of do-file

.
What do you think?

*
*   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/



--
Partha Deb
Professor of Economics
Hunter College
ph:  (212) 772-5435
fax: (212) 772-5398
http://urban.hunter.cuny.edu/~deb/

Emancipate yourselves from mental slavery
None but ourselves can free our minds.
	- Bob Marley

*
*   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/




--
Partha Deb
Professor of Economics
Hunter College
ph:  (212) 772-5435
fax: (212) 772-5398
http://urban.hunter.cuny.edu/~deb/

Emancipate yourselves from mental slavery
None but ourselves can free our minds.
	- Bob Marley

*
*   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   |   Site index