Fractional response variable panel data

[Elisabeth Zeidler]

Dear Statalists,

I am looking for hints how to handle my data structure: my dependent variable is a fractional response variable (interval 0 to 1, 0 included) and my data has a panel structure with t(29). To account for the autoregressive order of the data structure, is it possible to extend the command used by Papke and Woolridge (200) by
xtgee dv ivs, fam(binomial) link(logit) vce(robust) i(policysector) t(year) corr(ar1)
?
I seek to find a way to model my data without using a lagged dependent variable, would this be the right way? Also I am not sure about the usage of lagged independent variables with this model. I am greatful for any help.

Best, Elisabeth

[joonsdad]

Hi,

I'm working with similarly structured data. I found Prof. Wooldridge's slides from the 2011 Stata Users' Meetings pretty helpful (http://www.stata.com/meeting/chicago...wooldridge.pdf). The slides also include examples of Stata code for a heteroskedastic fractional probit program, applied to panel data. The presentation pays a lot of attention to unobserved heterogeneity - perhaps that is a way of thinking about (part of) what your lagged DV would capture?

Prof. Wooldridge's unpublished 2010 paper "Correlated Random Effects Models with Unbalanced Panels" (http://econ.msu.edu/faculty/wooldridge/docs/cre1_r4.pdf) covers a lot of the same ground, in more detail, but without code examples.

Prof. Papke's webpage (http://www.econ.msu.edu/faculty/papke/) also includes a link to Stata .do files and the meap92_01 panel data set analyzed in Papke & Wooldridge (2008).

I have also assumed (perhaps incorrectly?) that, to the extent that the solutions for unbalanced panels that Prof. Wooldridge offers in those materials boil down to including constructed variables (panel averages, selection dummies, etc.) in the regression and variance estimation, they can also be applied to the Stata hetprobit command. My hope is that with those additions, and specifying vce(cluster id), Stata's hetprobit command can be used to analyze panel data. If I'm wrong about that, hopefully someone will let me know!

Mike Mulcahy

References:
Papke, Leslie E., and Jeffrey M. Wooldridge. "Panel data methods for fractional response variables with an application to test pass rates." Journal of Econometrics 145.1 (2008): 121-133.
Wooldridge, Jeffrey. 2010. "Correlated Random Effects Models with Unbalanced Panels". Unpublished (http://econ.msu.edu/faculty/wooldridge/docs/cre1_r4.pdf)
Wooldridge, Jeffrey. 2011. "Fractional Response Models with Endogenous Explanatory Variables and Heterogeneity" CHI11 Stata Conference, No. 12

[Panika Jain]

joonsdad I have gone through all the materials listed by you. In" https://www.stata.com/meeting/chicago11/materials/chi11_wooldridge.pdf", he has explained through meap94_98 dataset. could you tell me from where I can get this. I am also working with similar dataset ( unbalanced panel, n=119 t=18), but I am not sure about the inclusion of time dummies in the model.

[Dimitriy V. Masterov]

Richard Williams has meap94_98.dta on his web page.

[Nefer Titi]

I am desperately looking for meap92_01 panel data set analyzed in Papke & Wooldridge (2008) but Pr Papke's website ( (http://www.econ.msu.edu/faculty/papke/) has been down for weeks. I will appreciate indication on where to find the dataset. The stat do.file will be a bonus. Thanks