Kurt Beron posted on Thursday, June 11, 2009 - 6:22 pm
I have several dependent variables that are interval-ordered. That is I know their cutpoints, though they are censored at both ends. I would like to estimate a multilevel model, eventually adding in a longitudinal component.
One way to do this is to lose the information of the cutpoints and just run as an ordered probit. However there are econometric models that can handle interval-level regressions, even with one random effect (e.g., Stata has intreg and xtintreg).
I would like to run these models in Mplus to allow for the additional features your program has, but don't see this type of dependent variable available. I'm wondering if there is a way to do this with the present version.
The CENSORED option in Mplus can have centering from above or below but not both. The censoring point is taken from the data. It is not estimated.
Kurt Beron posted on Friday, June 12, 2009 - 1:44 am
Thanks, I believe you are confirming for me that an interval model, leaving aside censoring from both sides, is not now part of Mplus. The model I am asking about might have data where you know that, say, income is less than $5,000, $5,000 to $9,999, $10,000 to $14,999, etc (perhaps being topcoded as more than $100,000). The CENSORED (Tobit-type) option wouldn't work here without picking arbitrary midpoints, and an ordered probit wouldn't take advantage of the information embedded in the cutpoints.
Can you give a reference for the "interval-ordered" econometric modeling that you refer to?
Kurt Beron posted on Friday, June 12, 2009 - 1:56 pm
Cameron and Trivedi, Microeconometrics, 2005, Cambridge Press has a brief discussion and also gives the log likelihood. It is implemented within Stata and documented in its TOBIT procedure under the name INTREG. Given the type of data we often have it's a nice tool for the kit.