COUNT MODEL USING PANEL DATA

[wanfairos]

Hi everybody!

I'm Wan Fairos from Malaysia. Currently, I am in the midst of doing my Phd and my thesis deal with a balance panel data focussing on modelling the non linear count model. I plan to used E-views 6 to analyze the data. As I am dealing with road accident death count data type (the non-negative integer value nature of data), the type of model that will be employed in my thesis is the nonlinear count model that involved exponential family distribution such as the Poisson Regression/Negative Binomial /Quasi Max.Likelh Model. I might overlooked, but as I go through the EViews user guide, the estimation procedure to develop the Poisson/NB/QML model available only for individual time series or individual cross section data. When I looked into the Pooled Time series and Cross Section data chapters, the procedure to estimate the pooled model available only for methods based on linear assumptions.
Thus, my questions are;
1) Using the pooled time series and cross section data, (panel count data), can the Fixed Effects (FE) and Random Effects (RE) of Panel Poisson or Negative Binomial Model developed by Hausman et.al (1984) be analyze using EViews 6?
2) Can Eviews 6 estimate the parameter models of FE and RE of Panel Poisson/NB model using the maximum likelihood of the conditional joint distribution?
I would very much appreciate for any answers and feedbacks regards to my questions.

Thank you.
Regards,
Wan Fairos

[EViews Glenn]

There are no in-built tools for this, but in principle you should be able to use the logl object.

[wanfairos]

Hi..where can I get the Deviance value in Poisson count model? Which statistics that it refers to?
TQ

[EViews Glenn]

The deviance isn't reported as part of the standard output, but it is easily computed for the Poisson model. Suppose that you have an equation object that has an estimated Poisson count model. From my notes (you should double check the formulas below)...first make the residuals and the fitted mean

Code: Select all

eq01.makeresid yresid
eq01.fitted ymean


Then compute the deviance contributions for each individual

Code: Select all

series dev = @recode(y<>0, y*log(y)-log(ymean)-yresid, -2*res)


You'll want the sum of this
scalar devscalar = @sum(dev)

[wanfairos]

Ok then. Thank you very much... . Really appreciate it. I'll will try to culculate the value based on your instruction. If there's any problem I will get back to you. Thanx!!!

Fairos

[wanfairos]

I've manage to calculate the residual (yresid) and the fitted (ymean) value. But still do not able to get the deviance series. There's a syntax error that I need to look into. Where can I refer the formula to generate the deviance series? TQ