Statalist The Stata Listserver


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

RE: st: discrete time survival analysis and logit model


From   "Stephen P. Jenkins" <[email protected]>
To   <[email protected]>
Subject   RE: st: discrete time survival analysis and logit model
Date   Fri, 24 Feb 2006 10:21:02 -0000

> Date: Thu, 23 Feb 2006 11:22:48 -0600 (CST)
> From: Fen Lin <[email protected]>
> Subject: st: discrete time survival analysis and logit model
> 
> Dear all,
> 
> I am using a discrete time survivial model to analyze 
> diffusion a policy 
> among 30 countries. My data is in a person-period structure, 
> so i have 
> 30 sequences but 1228 data points. In stata, i use logit command:
> 
> logit y x1 x2 x3 ln(duration), cluster(country)
> 
> so the standard errors will be adjusted by cluster. i am wondering:

"ln(duration)" looks like an illegal variable name to me. And I bet x1
x2 x3 were not your actual variable names.

As the Statalist FAQ advises, show us what you actually typed. If it is
necessary to explain what your covariates are, then do so afterwards.

> 1) is the deviance adjusted by cluaster as well? in the 
> output, number of 
> observation is 1228, rather than 30. Is it correct?

The relevant # obs here is the number of country-period obs. (I presume
you mean that you have a country-period data structure, not
person-period, as you wrote.)

To understand why, look at the expression for the likelihood
contributions. (See e.g. the pdf book at my survival analysis web pages
cited below my signature.)  

If I recall correctly the deviance is a transformation of the
log-likelihood, and I don't think it is affected by cluster.  (Why
should it be? Please revisit the "theory", and documentation for the
command telling you what cluster options do.)   I also note that -logit-
does not return the deviance in output nor in e() results.

> 2) since my time interval is each year, i have lots of zero 
> in certain 
> year, and a tie in others (that is why i use discrete time 
> model). but i 
> cannot use "dhaz" command due to those zeros. there is always 
> a message 
> "no variation in event occurence in period 1".  in this case, 
> how can i 
> calculate hazard?

What is "dhaz" and what is it intended to do?  -findit dhaz- returns
nothing.

> 3) my predicted probablity is between 0.002 to 0.35. how come 
> the phat 
> never larger than 0.5? ( i think it might because i use 
> "country" as unit 
> of analysis, then i have omiting variable bias.??)

What is "phat"?   

Use of -predict- after -logit- returns the predicted probability for the
particular country-period, i.e. the interval hazard.  For examples of
how to derive (and graph) the discrete hazard against duration for a
country with a particular set of characteristics, see Lesson 6 at my
website. It also shows how to derive and graph the discrete survivor
function.


Stephen
-------------------------------------------------------------
Professor Stephen P. Jenkins <[email protected]>
Institute for Social and Economic Research
University of Essex, Colchester CO4 3SQ, U.K.
Tel: +44 1206 873374.  Fax: +44 1206 873151.
http://www.iser.essex.ac.uk  
Survival Analysis using Stata:
http://www.iser.essex.ac.uk/teaching/degree/stephenj/ec968/ 



*
*   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 1996–2024 StataCorp LLC   |   Terms of use   |   Privacy   |   Contact us   |   What's new   |   Site index