[Date Prev][Date Next][Thread Prev][Thread Next][Date index][Thread index]
Re: st: xtfrontier fixed and random
Random effects is the default choice for the idiosyncratic error term (I have no idea about the fixed effects - i.e. if it is feasible to use dummies for the fixed effects or not). However, I suggest you to move to package Frontier 4.1 (or subsequent versions, if they exist) by Tim Coelli (you should find it at http://www.une.edu.au/econometrics/cepa.htm or something like that). It is a few hundred kilobytes, it is provided for free, it is easy-to-use, it has a small manual with some examples that are very useful and just a bit of the underlying statistics, and I regret to say that I found it to be better than Stata ('s official command for panels). In particular, Frontier 4.1 let you express the technical inefficiency term as a linear combination of some variables. INEXPLICABLY, Stata (just-updated, Intercooled, version 9.2) let you do this only for the cross-sectional command -frontier-, but not for the panel counterpart -xtfrontier-. In the latter command, the inefficienc!
y term can be only a random variable with truncated-normal distribution, or a random variable with truncated-normal distribution multiplied by a specific function of time (exponential decay).
Please remember that, if you need not to be bound to use only 1 output measure (dependent variable), or a specific functional form for the production frontier, you may use data envelopment analysis (not available on Stata to my best knowledge, but a program, DEAP, should be provided on the same webpage). However, a major weakness of data envelopment analysis is that it is deterministic, i.e. it does not distinguish between technical inefficiency and noise. Moreover two-stage estimation (i.e., calculation of efficiency scores and regression of these scores against exogenous variables) is problematic.
At 02.33 06/02/2007 -0500, you wrote:
>I'm beginning to work with xtfrontier and i was wondering to know in
>which way i can impose a fixed-effects or a random-effects panel model.
* For searches and help try: