Statalist The Stata Listserver


[Date Prev][Date Next][Thread Prev][Thread Next][Date index][Thread index]

Re: st: gllamm output


From   Christian Ganser <christian.ganser@soziologie.uni-muenchen.de>
To   statalist@hsphsun2.harvard.edu
Subject   Re: st: gllamm output
Date   Tue, 13 Jun 2006 15:49:28 +0200

Dari,

if you just want a random intercept and count as an independent variable, type

gllamm volrel2 schmooze socialtrust civpartr group coll married healthr
relatendr count, i(countyid) family(bin) link(logit)

I'm sorry I can't tell much about the condition number, but to me 584 seems to be very high, indeed. However, the gllamm-manual states: "From our experience so far, large condition numbers do not necessarily imply poor identification; however, it is unlikely that a low condition number is obtained when the model is not identified." Perhaps someone else on the list can tell you more about this...

Dari Sylvester wrote:

Christian,

Thank you. Yes - you are correct that county id should be the level 2
identifier. If I want to have "count" as a level 2 independent
variable, would I have the following:

eqs count: cons
gllamm volrel2 schmooze socialtrust civpartr group coll married healthr
relatendr, i(countyid) family(bin) link(logit)
When I ran that, it seemed to converge, but with a condition number of
584! If I am correct, that's not necessarily a good sign.

Thanks for your help, Christian.

Best,
Dari
Dari E. Sylvester
Assistant Professor, Political Science
Senior Fellow, Jacoby Center for Public Service and Civic Leadership
University of the Pacific
Stockton, CA 95211
Phone: (209) 946-2007
Fax: (209) 946-2318

christian.ganser@soziologie.uni-muenchen.de 06/12/06 1:33 AM >>>
Only answering questions 1 and 2.

Are you sure you don't want to fit the following:

gllamm DV V1 V2 V3 V4 V5 V6 V7 V8 count, i(YOUR COUNTY-ID) family(bin)
link(logit)
The number of volunteer organizations in the county is an independent variable, isn't it? You used it as the county-identifier, so your model doesn't make much sense, I think.

Question 2: Using the latent-response formulation of the logistic regression model, y*i=b0+b1x1i+...+ei, it is assumed that ei has a logistic distribution with mean zero and variance pi^2/3=3.29.

Dari Sylvester wrote:

I used the gllamm command for a multilevel analysis that estimates the
dichotomous outcome of volunteering or not volunteering based on
individual characteristics (level 1) and the number of volunteer
organizations in the county in which one lives (level 2). There are explanatory variables at the level of individuals from V1
through V8.
I have no particular equation to estimate for the county level. I ran the following:

gllamm DV V1 V2 V3 V4 V5 V6 V7 V8, i(count) family(bin) link(logit) **Where DV is the dependent variable, V1-V8 are explanatory variables
for the individual level, COUNT is the count of organizations in the
county level. **
I received the following output:
number of level 1 units = 19395
number of level 2 units = 137
Condition Number = 481.48689
gllamm model
log likelihood = -8947.353

------------------------------------------------------------------------------

DV | Coef. Std. Err. z P>|z| [95% Conf.
Interval]

-------------+----------------------------------------------------------------

V1 | .0052359 .0003777 13.86 0.000 .0044956 .0059763
V2 | .6474295 .0736646 8.79 0.000 .5030495 .7918094
V3 | 1.544726 .0851626 18.14 0.000 1.37781 1.711642
V4 | .2398346 .0198328 12.09 0.000 .200963 .2787061
V5 | .4001724 .0403935 9.91 0.000 .3210025 .4793423
V6 | .386209 .0383842 10.06 0.000 .3109772 .4614407
V7 | .6155165 .0775643 7.94 0.000 .4634934 .7675397
V8 | 3.722145 .0636431 58.48 0.000 3.597407 3.846883
_cons | -5.60294 .1000996 -55.97 0.000 -5.799131 -5.406748

Variances and covariances of random effects

-----------------------------------------------------------------------

***level 2 (count)
var(1): .01434088 (.00613062)
Questions:
1. I'm not sure why the condition number is so high - leads me to
believe I've made an error in how I've set up the gllamm model.
2. Why isn't level 1 variance reported?
3. How does the interaction between levels (i.e. the individual nested
in numbers of county organizations) differ in this multilevel
estimation

from running a probit with interaction terms added for each of the
individual level variables of interest multiplied by the number of
county organizations? In other words, running: probit DV V1 V2 V3 V1*COUNT V2*COUNT V3*COUNT
Thank you kindly,
D. Sylvester

Dari E. Sylvester
Assistant Professor, Political Science
Senior Fellow, Jacoby Center for Public Service and Civic Leadership
University of the Pacific
Stockton, CA 95211
Phone: (209) 946-2007
Fax: (209) 946-2318
*
* For searches and help try:
* http://www.stata.com/support/faqs/res/findit.html
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/

*
* For searches and help try:
* http://www.stata.com/support/faqs/res/findit.html
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/

*
* For searches and help try:
* http://www.stata.com/support/faqs/res/findit.html
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/

*
*   For searches and help try:
*   http://www.stata.com/support/faqs/res/findit.html
*   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