Hi guys. I'm doing criminology study with dependent variable data that is of the beset-with-zeroes, censored L-shaped distribution that plagues the field like a bad smell.
I'm using Tobit regression as one of my analyses, and have a couple of questions I hope someone might be able to help out on.
The first one is regarding the distinction between classic Tobit and heteroscedastic Tobit regression, as discussed by Sullivan et al (1995: http://www.springerlink.com/content/p1p6843r074g6504/). I'd rather use heteroskedastic Tobit if possible, and I've noticed the QLIM manual refers to the procedure as 'Tobit with heteroscedasticity'. I'm a little confused in my interpretation of the manual, though, as to whether I need to use a specific command to estimate the Tobit model with the heteroscedasticity assumption relaxed, or is this the default?
You'll need to learn a bit about the SAS macro language and perhaps ODS to use them. If QLIM will output the moments directly, then you can wrap the bootstrap macro around the procedure and get your estimates fairly directly. If it does not, then you will need use the more general ODS tool to output the moments and collect them for estimation (I've not done this, but know others who have).