[Date Prev][Date Next][Thread Prev][Thread Next][Date index][Thread index]
st: xtabond - dropped observations
I'm trying to figure out a discrepany in the number of
observations reported by xtabond and xtabond2. For
comparison, I ran basic xtreg, fe. The number of
observations and groups in the same for the fixed
effects and xtabond2 (Blundell Bond). However, the
xtabond (Arellano-Bond) reports a different number of
Having surveyed the Statalist archives, I ran across a
posting that appears related but the solution posted
doesn't seem to apply to my problem.
I have an unbalanced panel with more than 30 years for
each group. Trying to analysis the impact of panel
length, I then subdivide each group into 3
non-overlapping subgroups. Here the number of groups
changes as well as the number of observations.
xi: xtreg mdr mdrl ebit_tal mbl dep_tal lntal fa_tal
rdl rd_tal ffindtargetm i.year, fe
xi: xtabond mdr ebit_tal mbl dep_tal lntal fa_tal rdl
rd_tal ffindtargetm i.year, robust
xi: xtabond2 mdr mdrl ebit_tal mbl dep_tal lntal
fa_tal rdl rd_tal ffindtargetm i.year, iv(i.year)
gmm(mdrl mbl ebit_tal lntal, c) robust
Number of Observations/Groups with Full Panel:
FE - 3115 obs, 117 groups
xtabond - 2955 obs, 117 groups
xtabond2 - 3115 obs, 117 groups
I believe that the difference here is related to the
number of groups, the number of independent variables,
and the year effects
(117 + 9 + 34 == 160 == 3115 - 2955).
I am uncertain how to relate the calculation above to
the numbers reported when I use subgroups (below).
Number of Observations/Groups with Subgroups:
FE - 3115 obs, 351 groups
xtabond - 2511 obs, 350 groups
xtabond2 - 3115 obs, 351 groups
If anyone can explain what is going on here, I would
Do You Yahoo!?
Tired of spam? Yahoo! Mail has the best spam protection around
* For searches and help try: