Dear Fellow Stata Users,
after reading various parts of the Stata documentation and some statistical texts from my discipline (political science), I am still unsure which method to use for the analysis of my data file. My hope is that someone on this list can help me with my problem. I already asked all colleagues and professors around here, but nobody is so deep into statistics that he or she knows a good answer.
Let me briefly describe the structure of the data:
The data file contains information about public budgets in 6 geographical units (let's call them countries). All cities do consist of some subunits (let's call them districts), 117 altogether, but unevenly distributed, so one country only has 9 districts, another 10, another 26 and so on. For each district, I have budget data for 15 consecutive years. This makes 15 years * 117 districts = 1,755 cases.
My dependent variable for the analysis is the so called "public expenditure quota" (that is: total expenditure of the district divided by the GDP in the district). There are are some control variables for economic development and context (unemployment, population etc.) Independent variables are for example a left-right judgement of the district administration (=government), district council party control, district council fragmentation and some more. The goal of the research is to explain variations in the size of the district budgets considering effects of the district and country.
My problem is to decide which statistical method I should use. I read instructions about General Least Squares equations, Multi-Level Mixed Models etc. As I understood, some of them allow seperating country specific and district specific effects from random effects and from the impact of the independent variables - with different levels or without and with hierarchy between the levels or without hierarchy. It would be nice to have that in order to seperate countries from each other and also districts within a country from each other.
On the other side, there are methods which do allow control for autocorrelation on the dependent variable. I need that too, because in fact I am running 117 time series parallely. I have used the Prais-Winsten regression on that and it works well for single districts, but all I can do with it is to search for general patterns through districts of the same country. So I would end up with 6 country models and would have to compare them qualitatively.
The problem is that I do not know which method does combine both approaches: I like to have a hierarchical model (countries on the top level, districts on the middle level, 15 years on the lower level), but on the lower level, autocorrelation must be controllable.
To come to the deciding point: Anybody knows how to deal with that problem?
Thank you for reading,
Bjoern
--
GRATIS für alle GMX-Mitglieder: Die maxdome Movie-FLAT!
Jetzt freischalten unter http://portal.gmx.net/de/go/maxdome01
*
* For searches and help try:
* http://www.stata.com/help.cgi?search
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/
