I have seen a model on a paper that can fit my research problem and I would like to implement it on Stata:
"we examine counts of citations in an 8-year (1988�1995) pooled cross-section using a negative binomial specification. We employ fixed university and year effects to control for unobserved heterogeneity across time and campuses, modeling the dependent variable as:
y(i,t) = alpha(i) + delta(t) + SUM [beta(j) * x(i,t,j)]+ e(i,t)
where alpha(i) is the effect of university i (i=1,...,N), delta(t) is the effect of year t (t=1,...,8), beta(j) is the within university slope for x(j) pooled over all universities and years."
I used the SUM to indicate that there is a sum for j= 1,...J of the terms between squared parentheses []. This is because there are J covariates. Can I translate "pooled cross-section" with:
nbreg y x alpha delta
where y is the dependent variable, x are the covariates, alpha is a variable from 1 to N, delta is a variable from 1 to 8???
Which advantages (or drawbacks) do I have in respect to the following???
tsset alpha delta
xtnbreg y x, fe
In the last question I am referring mostly to a general question - why should I treat panel data as crossectional data and can some tests help me with such choice --- none have been reported in the paper I want to copy, nor in other papers - and, to a lesser extent, to the issue of conditional vs. unconditional fixed effects (since it seems that Stata's conditional fixed effects for the negative binomial model are not really fixed, see e.g. the reference below --- surprisingly, I failed to find a proper discussion of this latter problem on the FAQ, despite its relevance, or in Statalist).
Thanks,
Nicola
Allison, Paul D. and Richard Waterman. 2002. �Fixed effects negative binomial regression models.� Ross Stolzenberg (ed.), Sociological Methodology 2002. Boston: Basil Blackwell.
*
* For searches and help try:
* http://www.stata.com/support/faqs/res/findit.html
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/
