Behram Wali posted on Tuesday, April 26, 2016 - 6:54 pm
I am new to MPlus. I am trying to estimate a bivariate ordered probit model (two equation ordered model with error correlation) with random parameters.
I shall briefly explain random parameters as researchers refer to it with different names. By random parameter model, i mean that some parameters can be fixed i.e. constant parameter estimate for all observations. While other can be random parameters e.g. normally or Weibull distributed random parameters. Thus for random parameters, the routine is expected to give me a mean and standard deviation for each random-held parameter. Limdep/NLOGIT allow random parameter models with simulated maximum likelihood procedures.
One of my professor recommended MPlus but i am not sure if MPlus has such a routine as i could not find any reference in past threads. Any guidance in this regard shall highly be appreciated.
Let me first ask if you are considering a multilevel model with observations nested within clusters.
Behram Wali posted on Wednesday, April 27, 2016 - 9:19 am
I do not aim at a multilevel model with clusters/groups as we do in random-effect models.
So if i have a two equation ordered probit model, with X and Y number of ind. variables, i would like to allow the ind. variables within each equation to vary across observations as per some pre-defined distribution. Said this, it will be a single-level model with no nested clusters. Thanks
Thanks a lot for your response. I agree that random parameter models (hierarchical, mixed, or multilevel models) are same as two-level models. I watched the exhaustive courses regarding multilevel modeling available at MPlus homepage. However, I have confusion that limit my ability to understand it correctly. In traditional literature, the applications of two or multi-level models are usually defined for data having hierarchies i.e. voters nested within countries, or workers nested within firms, motivating “between” and “within” analyses. In presence of hierarchies in data structure, I completely understand the mechanism behind two-level models. However, in my case, there exist no hierarchy in data i.e. there is only one level that is the cross section of road traffic crashes, with each row representing each crash and the dependent variables being injury severities (on ordinal scale) of two “crash involved drivers”. Said this, can random parameter bivariate ordered probit model be still estimated with no grouping/clusters variable?