I'm struggling to determine the most appropriate way to estimate a
hazard model for the following data:
Dependent Variable
Yitj=1,0 where i=istitutions, t=periods from t=0, and j=firm level
(Different institutions (i) are invested in firm (j) and then at some
point in time the firms choose to sell their investment)
Independent Variables
Xi = vector of institutional characteristics that don't vary over time
Zit= vector of institutional variables that do change over time
Ptj= vector of firm variables that change over time
The panel of data is such that t is measured in discrete chunks
(quarters) and ranges from 1 to 16 until the observations (ij) leave the
sample (are either censored or experience the event). The events are
measure discretely, but the actual theoretical event is continuous.
The number of total observations is approximately 115,000 with 230 firms
(j) and 1 to 2,000 institutions (i) in each firm (j).
I'm trying to estimate a hazard model where Yitj=f(Xi,Zit,Ptj). I've
run both a maximum likelihood logit estimation and a partial likelihood
estimation, but I'm concerned because I likely have clustering of my
error terms at the firm level. Does anyone have any suggestions for
dealing with this? Should I run a fixed or random effects model? Is
there a method for simply correcting my standard errors in stata? Can
you point me to some literature? Thanks for the help.
Chad
*
* 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/
