Notice: On April 23, 2014, Statalist moved from an email list to a forum, based at statalist.org.
From | Leny Mathew <lenymathewc@gmail.com> |
To | statalist@hsphsun2.harvard.edu |
Subject | st: Hurdle model using Gamma distribution |
Date | Mon, 26 Jul 2010 16:46:52 -0400 |
Hello All, I'm trying to use a hurdle model to model continuous data which has zeros due to the existence of a minimum detectable limit. Instead of the Poisson or negative binomial distribution which seem to be commonly used in hurdle models, I would like to to use the Gamma distribution to model the continuous data. Since the Gamma distribution is defined only for x >0, is it possible to develop this by estimating the binomial model separately from the parameters of a gamma regression model? I wrote out the Log likelihood for this model in the same way as described in http://www.stata-journal.com/sjpdf.html?articlenum=st0040 and the equation can be written out as the sum of the log-likelihood of the binary model and the log likelihood of the gamma model. However I'm not sure if this is the right way to proceed. On this data data set, I did run the Poisson hurdle model and the negative Binomial hurdle model and compared it to the output of the model I described above. The results of the three models are remarkably similar. Any comments would be greatly appreciated. Thanks, Leny * * 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/