Hello,
I am switching to R from Stata and I am having particular trouble with
the transition from Stata's 'xtabond' and 'ivreg' commands to the
"plm" package. I am trying to replicate some of the dynamic panel data
work using the UK Employment data in Arellano and Bond (1991) and
available as 'EmplUK' under the 'plm' package.
I have been reading "Panel Data Econometrics in R: The plm Package" by
Croissant and Millo available at
http://cran.r-project.org/web/packages/plm/vignettes/plm.pdf and "How
to Do xtabond2: An Introduction to 'Difference' and 'System' GMM in
Stata" by David Roodman available at http://www.cgdev.org/content/publications/detail/11619
. Roodman provides a very clear exposition of how to use Stata to
analyze the UK Employment Data. I am trying to replicate Roodman's
results for the UK Employment data using R instead of Stata but I am
having limited success.
Using:
>library('plm')
>data("EmplUK", package = "plm")
>emp.plm <- plm(dynformula(emp ~ wage + capital + output, lag =
list(2, 1, 2, 2), log = TRUE), EmplUK, effect = "time")
>summary(emp.plm)
I am able to perfectly replicate Roodman's "naive model" (on page 17)
regressing Log(Employment) on its own first and second lags as well as
current and first lags of log(wages) and current/first/second lags of
capital and output. Roodman uses the Stata command "regress n nL1 nL2
w wL1 k kL1 kL2 ys ysL1 ysL2 yr*" (n=employment, w=wages, k=capital,
ys=output, yr*=year dummy variables, and nL1=first Lag of employment).
I am unable to replicate other results. Specifically, I cannot even
replicate the Least Squares Dummy Variable model with effects for both
time and firm (in Stata: xi: regress n nL1 nL2 w wL1 k kL1 kL2 ys ysL1
ysL2 yr* i.id)
In R I tried:
>emp.lsdv <- plm(dynformula(emp ~ wage + capital + output, lag =
list(2, 1, 2, 2), log = TRUE), EmplUK, model="within", effect =
"twoways")
>summary(emp.lsdv)
but the coefficients do not match up with results shown on p 18 of
Roodman. Can someone help point out what I am doing incorrectly?
Can anyone help me implement a First Differences model that also
includes Year specific effects? First Differencing eliminates the
individual effects, but I should still be able to add year specific
effects, no? When I run the commands:
>emp.fd <- plm(dynformula(emp ~ wage + capital + output, lag =
list(2, 1, 2, 2), log = TRUE), EmplUK, model="fd", effect = "time")
>summary(emp.fd)
the output says it is running a "time" effect First-Difference Model,
but I am unable to extract any time effects, nor can I find any
differences with the output from:
>emp.fdid <- plm(dynformula(emp ~ wage + capital + output, lag =
list(2, 1, 2, 2), log = TRUE), EmplUK, model="fd", effect =
"individual")
>summary(emp.fdid)
What am I missing? Even the degrees of freedom appear the same to me.
Eventually, I would like to understand how to implement instrumental
variables in the dynamic panel setting using General Method of Moments
using R rather than Stata, but it seems I have quite a ways to go to
better understand how 'plm' works. Any other resources anyone could
point me to would be appreciated.
Thanks,
Aaron
______________________________________________
R-help@r-project.org mailing list
https://stat.ethz.ch/mailman/listinfo/r-help
PLEASE do read the posting guide http://www.R-project.org/posting-guide.html
and provide commented, minimal, self-contained, reproducible code.
