Bookmark and Share

Notice: On April 23, 2014, Statalist moved from an email list to a forum, based at statalist.org.


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

st: Latent normal distribution for count variables


From   [email protected]
To   [email protected]
Subject   st: Latent normal distribution for count variables
Date   Tue, 29 Mar 2011 18:19:46 +0200

Dear Statalisters,
I'm dealing with the issue I described here:

http://www.stata.com/statalist/archive/2011-03/msg01309.html

Basically, I can't find a way to perform a 3-level regression having a zero-truncated count variable as outcome.
Of course, I could group the outcomes in 2 or a few categories and perform binary or multinomial regression, but I was asked to keep the dependent variables as they are.
I started to take a look also outside Stata commands, to see whether similar issues have been faced so far. I found a paper on JRSS (Goldstein and Kounali, 2009), that you can find here:

http://onlinelibrary.wiley.com/doi/10.1111/j.1467-985X.2008.00576.x/full   

They propose a method to deal also with very large count variables. It seems an extension of what is done with multinomial distribution: they actually "treat count data as an order classification", using a smoothing function to prevent inefficiency and instability (due to the presence of many thresholds). Alternatively, they propose a latent normal characterisation of the Poisson distribution.
I think a similar approach would make sense to me. For example, one of my outcome is not properly a count variable, since it is "mean length of stay" and can assume also non-integer values (Stata performs Poisson regression as well, interpreting numbers as rates), while another one is total number of days in hospital in one year (therefore, there is an upper limit), so the Poisson distribution seems unfit (even if I found a way to control for non-zero truncation and assuming I can deal with over-dispersion).
Can I find in Stata something similar? Basically, I'd like use the "gllamm" command on data that are transformed in such a way to be normally distributed.
TIA,
Federico

Harvey Goldstein & Daphne Kounali, 2009.
"Multilevel multivariate modelling of childhood growth, numbers of growth measurements and adult characteristics," Journal Of The Royal Statistical Society Series A, Royal Statistical Society, vol. 172(3), pages 599-613.


*
*   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 1996–2018 StataCorp LLC   |   Terms of use   |   Privacy   |   Contact us   |   Site index