multilevlel panel data

classic Classic list List threaded Threaded
1 message Options
Reply | Threaded
Open this post in threaded view
Report Content as Inappropriate

multilevlel panel data

Hello Statalist

I have a panel dataset (12 Quarters, 300 car-models, 38 car-brands). The car models are obviously nested within brands. The dependent variable is a loyalty measure (y), and the main independent variable is the a dummy variable (x) for introduction of a new-generation car model (controlled for ad-spendings and other marketing mix measures).

As the loyalty measure is expected to be dependent on omitted variables and the mean loyalties are assumed to be model specific, i estimated the regression using the fixed effects model xtreg, fe. (hausman test confirms the use of fe over re)

More precisely: xtreg y x L(1/3).x F(1/3)x ads L.ads, fe vce(cluster model)

(1) First of all do you think this model (especially the use of the lags to model the development of y over time around the new introduction date) makes sense? Or is there maybe a better approach?

As the models are nested in brands i would like to also control for brand specific effects and error term correlation on brand level. I read a lot about the use of panel data and hierarchical models and think the use of -xtmixed- would be useful in this case. However i dont get how i can incorporate time-series and multilevel in the regression:

would something like that be the right way:

xtmixed y x L(1/3).x F(1/3).x ads L.ads || brand: || model:, mle

(2) or am i completely wrong here? Is the datevar (specified under "xtset model datevar") now still taken into account as the lowest level of the hierarchy? I dont really get this one.

Any help on this problem would be much appreciated. Thank you in advance for your support.


Tobias Friedli
Bachelor Student at the University of Z├╝rich,
Chair for Market Research