13 Apr
2010
13 Apr
'10
6:52 p.m.
Hi all,
We're working on our replication paper which requires including fixed
effects in a probit regression. When we run the regression we get reported
coefficients for all of our independent variables and for each of the
possible fixed effect values except for one of them.
Before the regression we run as.factor to create industry.factor, the fixed
effects for industries and then include industry.factor in the probit
regression specification. After running summary on the output of the probit,
we get estimated coefficients for every fixed effect except for "Textiles".
Before the regression we can run summary(industry.factor) and we can see
that there are over 1200 observations for "Textiles" in industry.factor.
"Textiles" is the first industry listed in industry.factor, so I suspect
that R is somehow dropping it for being first in industry.factor. Any help
would be appreciated. We tried this with both the glm and zelig
implementations and found the exact same problem.
Thanks!
-Gabe & Tara
--
gc
13 Apr
13 Apr
7:14 p.m.
Hi Gabe,
For factor variables (like fixed effects) one category is usually used as a
reference category, as long as the model also contains an intercept. The
reason you need to do this is that the linear combination of the industry
dummies (which are your fixed effects) would be perfectly collinear with the
intercept term, leading to an identification problem. If you want the mean
outcome for your reference category, all the information you need is in the
intercept (and whatever values you wish to set the other covariates at).
Iain
On Tue, Apr 13, 2010 at 6:52 PM, Gabriel Chan
<gabe_chan at hksphd.harvard.edu>wrote:
Hi all,
We're working on our replication paper which requires including fixed
effects in a probit regression. When we run the regression we get reported
coefficients for all of our independent variables and for each of the
possible fixed effect values except for one of them.
Before the regression we run as.factor to create industry.factor, the fixed
effects for industries and then include industry.factor in the probit
regression specification. After running summary on the output of the probit,
we get estimated coefficients for every fixed effect except for "Textiles".
Before the regression we can run summary(industry.factor) and we can see
that there are over 1200 observations for "Textiles" in industry.factor.
"Textiles" is the first industry listed in industry.factor, so I suspect
that R is somehow dropping it for being first in industry.factor. Any help
would be appreciated. We tried this with both the glm and zelig
implementations and found the exact same problem.
Thanks!
-Gabe & Tara
--
gc
_______________________________________________
gov2001-l mailing list
gov2001-l at lists.fas.harvard.edu
http://lists.fas.harvard.edu/mailman/listinfo/gov2001-l
7:52 p.m.
another way to think about this is to consider the case with only 2
categories. would it make any sense to include a variable which is 1
if male and 0 if female and another vairable which is 0 if male and 1
if female?
Gary
---
http://gking.harvard.edu
On Tue, Apr 13, 2010 at 7:14 PM, Iain Osgood <osgood2 at fas.harvard.edu> wrote:
Hi Gabe,
For factor variables (like fixed effects) one category is usually used as a
reference category, as long as the model also contains an intercept. ?The
reason you need to do this is that the linear combination of the ?industry
dummies (which are your fixed effects) would be perfectly collinear with the
intercept term, leading to an identification problem. ?If you want the mean
outcome for your reference category, all the information you need is in the
intercept (and whatever values you wish to set the other covariates at).
Iain
On Tue, Apr 13, 2010 at 6:52 PM, Gabriel Chan <gabe_chan at hksphd.harvard.edu>
wrote:
Hi all,
We're working on our replication paper which requires including fixed
effects in a probit regression. When we run the regression we get reported
coefficients for all of our independent variables and for each of the
possible fixed effect values except for one of them.
Before the regression we run as.factor to create industry.factor, the
fixed effects for industries and then include industry.factor in the probit
regression specification. After running summary on the output of the probit,
we get estimated coefficients for every fixed effect except for "Textiles".
Before the regression we can run summary(industry.factor) and we can see
that there are over 1200 observations for "Textiles" in industry.factor.
"Textiles" is the first industry listed in industry.factor, so I suspect
that R is somehow dropping it for being first in industry.factor. Any help
would be appreciated. We tried this with both the glm and zelig
implementations and found the exact same problem.
Thanks!
-Gabe & Tara
--
gc
_______________________________________________
gov2001-l mailing list
gov2001-l at lists.fas.harvard.edu
http://lists.fas.harvard.edu/mailman/listinfo/gov2001-l
_______________________________________________
gov2001-l mailing list
gov2001-l at lists.fas.harvard.edu
http://lists.fas.harvard.edu/mailman/listinfo/gov2001-l
5150
days inactive
5150
days old
gov2001@lists.gking.harvard.edu
2 comments
3 participants
participants (3)
-
gabe_chan＠hksphd.harvard.edu -
king＠harvard.edu -
osgood2＠fas.harvard.edu