halfway between what you did and Brandon's suggestion might be to fixed
effects for something like regions, or groups of villages.
Gary
--
*Gary King* - Albert J. Weatherhead III University Professor - Director,
IQSS - Harvard University
<http://gking.harvard.edu/> - King(a)Harvard.edu -
@kinggary<http://twitter.com/kinggary>- 617-500-7570 - Asst 495-9271 -
Fax 812-8581
On Sat, Apr 23, 2011 at 7:20 PM, Brandon Stewart
<brandonmstewart(a)gmail.com>wrote;wrote:
Leslie,
Others on the list have more experience with large sets of fixed
effects than I do, but some observations:
1) Sure you can add 500 fixed effects, but you should expect those
with 6 observations to have much larger standard errors.
2) This is where many turn to hierarchical models (see Gelman and Hill
2007 book). The general idea is that we might believe these fixed
effects are related (for example drawn from a common distribution),
so we "borrow strength" from the other fixed effects to estimate the
villages with 6 observations. Gelman and Hill have a nice overview of
the applications and software available to fit such models.
3) You may find R slowing down considerably when you try to estimate
so many fixed effects. There are some clever ways to speed this up
(basically via de-meaning the data by village etc.) which you can look
up and implement if it starts to annoy you.
Brandon
On Sat, Apr 23, 2011 at 6:55 PM, Leslie Finger <lfinger(a)fas.harvard.edu>
wrote:
Hey class,
We're working on adding village-level fixed effects to our replication
model.
However, there are some villages that have as few
as 6 observations and
other
that have over one hundred. It seems strange to
add a dummy for a
village with
so few observations.
Should we just do it anyway? Additionally, there are as many as 500
villages in
our data set. Do people include fixed effects
dummies when there are so
many
like that?
-Leslie
_______________________________________________
gov2001-l mailing list
gov2001-l(a)lists.fas.harvard.edu
http://lists.fas.harvard.edu/mailman/listinfo/gov2001-l
_______________________________________________
gov2001-l mailing list
gov2001-l(a)lists.fas.harvard.edu
http://lists.fas.harvard.edu/mailman/listinfo/gov2001-l
_______________________________________________
gov2001-l mailing list
gov2001-l(a)lists.fas.harvard.edu