Thanks! I'll check out the material for the censoring and survival model you suggested. As I explain more below, other studies similar to my little project cite Maddala (1983), so maybe it's relevant too.
I want to model contract duration as a function of several firm and industry characteristics. I started out with a hurdle approach, where the probit portion addresses contract adoption (some firms don't use written contracts) and the truncated regression portion addresses contract duration.
I've since come across some literature that suggests a truncation issue that I'm not sure I'm addressing with this model. Crocker and Masten (1988, p. 339) and Joskow (1987, p. 178) cite Maddala's (1983, p. 165-170) discussion of maximum likelihood estimation for a truncated dependent variable for which observations above a certain value were eliminated from the dataset. (URLs to references below)
In my case, the contract data was recorded in 2006. Contract duration varies between 1 and 10 years. Consider the contract with a duration of 10 years. Potentially, other contracts of shorter duration may have been signed in the same year as this 10 year contract but I have no record of them. In this sense, the sample is truncated, but isn't this different than just truncating all zero observations (observations of no contract)?
Jason
References:
Crocker and Masten (1988) Mitigating Contractual Hazards: Unilateral Options and Contract Length
(http://www.jstor.org/stable/pdfplus/2555660.pdf)
Joskow (1987) Contract Duration and Relationship-Specific Investments: Empirical Evidence from Coal Markets
http://links.jstor.org/sici?sici=0002-8282%28198703%2977%3A1%3C168%3ACDARIE%3E2.0.CO%3B2-H
(http://www.jstor.org/stable/pdfplus/1806736.pdf)
Maddala (1983)
http://books.google.com/books?hl=en&lr=&id=-Ji1ZaUg7gcC&oi=fnd&pg=PR11&dq=Maddala+(1983)+limited+dependent+and+quali&ots=7d2o6EjTJM&sig=77sA4nJ7g5xQ6kzZMJHAkVjIVMI#PPA170,M1
-----Original Message-----
From: [email protected] [mailto:[email protected]] On Behalf Of Austin Nichols
Sent: Thursday, May 07, 2009 12:18 PM
To: [email protected]
Subject: Re: st: limited dependent variables
Franken, Jason R. <[email protected]>:
I think the reference does not apply to your case; I think you do not
want -truncreg- (or -heckman-? you don't specify whether the contract
length is the outcome or explanatory variable) because of the
normality assumption required. You have censoring in a survival
model--read the material at
http://www.iser.essex.ac.uk/iser/teaching/module-ec968 for starters.
For more help, you should be more specific about the model you hope to estimate.
On Thu, May 7, 2009 at 1:07 PM, Franken, Jason R. <[email protected]> wrote:
>
> Is there a user-written program appropriate for the type of truncated data discussed in Maddala (1983, p.165-170): "Limited-Dependent and Qualitative Variables in Econometrics"?
>
> I have data on contract duration for contracts observed in 2006 but signed at earlier dates. This means that I observe some contracts with a duration of ten years, but I do not observe a contract signed in the same year with a duration of 3 years (that is, it was binding for only 3 years). OLS estimates are therefore biased, and MLE is needed.
*
* 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/