Clustered SE & High amount of treatment obs.

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

Clustered SE & High amount of treatment obs.


I'm facing some problems in my analysis. I want to test a treatment-effect on various firms.

Panel data, unbalanced, 50 firms/15 years, about 600 obs, average treatment years: 4,37.

I'm currently using:
cgmreg Y DTreat SC_INCPT X1 X2 i.year i.firm, noconstant cluster(firm year)

where: DTreat is a treatment dummy, SC_INCPT Y X1 X2 are continuous (scaled) variables.

Y                       Coef.     Std. Err.      z      P>|z|     [95% Conf. Interval]
SC_INCPT        -6686.274   7841.345    -0.85   0.394    -22055.03     8682.48
DTreat           -.0217945   .0085074    -2.56   0.010    -.0384687   -.0051202
X1                  .0220784   .0250524     0.88   0.378    -.0270234    .0711801
X2                -.1035717   .0178838    -5.79   0.000    -.1386233   -.0685201
(year/firm dummies not reported, all variables scaled, both FE have highly significant F-Test)

1.) The treatment variable seems to be signifcant, but about 30% of the whole sample are treatment firm-year observations (4.37 on average). Could this cause problems? If yes, how can I adress them in Stata?

2.) X1 is unfortunately not significant. It is a change variable (xt-xt_1) which gets insignificant when clustered standard errors are used (highly significant without firm-clustered SE). How can I adress this issue?

Thank you!