R help - Mar 2012 - Panel models: Fixed effects & random coefficients in plm

Hello,

I am using {plm} to estimate panel models. I want to estimate a model that
includes fixed effects for time and individual, but has a random individual
effect for the coefficient on the independent variable.

That is, I would like to estimate the model:
Y_it = a_i + a_t + B_i * X_it + e_it
Where i denotes individuals, t denotes time, X is my independent variable,
and B (beta) is the coefficient on that random variable. I want both a
coefficients to be estimated with fixed effects because I expect them to be
correlated with Y, and B to be estimated using a random effect.

I understand how to estimate two way fixed effects models in plm (which
have the fixed effects for time and individual, as I want) and how to
estimate random coefficient models (which have random effects for all
coefficients, including the intercept and the beta). I want to combine
these, though, and I cannot figure out how to do that. It seems like the
plm package is capable, but I can't figure it out.

Below is reproducible code (assuming you have plm installed) taken from the
vignette available on cran. It shows the two models I know how to estimate.
Any guidance on estimating the third model would be greatly appreciated.

-Mitch

# Data setup
library(plm)
data("EmplUK", package="plm")
names(EmplUK)
E <- pdata.frame(EmplUK, index = c("firm", "year"),
drop.index = TRUE,
row.names = TRUE)

# Two-way fixed effects model with constant beta

m1 <- plm(wage ~
output,data=E,model="within",effect="twoways")
summary(m1)

# Random (individual) effects for both intercept and beta

m2 <- pvcm(wage ~ output,data=E,model="random")
summary(m2)

Hello.
Such a procedure is not implemented in 'plm' but you can probably get
around to do what you want. One possible way (my preferred one) is to
demean the data by both dimensions as you would do for a two-way FE
model, and then estimate a random parameters model on the demeaned data.
This would get you rid of fixed effects both ways, although inducing
correlation in the residuals (something I don't think is particularly
harmful in your case, but this is just a quick opinion...). To do this
you can in fact use infrastructure from 'plm', e.g. using model.matrix()
and pmodel.response(); or you can do it by hand in base R.

Another way is to explicitly add time and individual dummies, including
'as.factor(<indvar>)' and 'as.factor(<timevar>)' to
the specification;
this way, though, you will be estimating the relative coefficients as
random parameters instead than fixed as well.

A third possibility, outside 'plm', is to check in the 'nlme'
package
(lme() function) whether such a specification is possible at all: I
think it is, putting the two sets of dummies as "fixed" parameters and
your beta at "random".

Best wishes,
Giovanni

Giovanni Millo, PhD
Research Dept.,
Assicurazioni Generali SpA
Via Machiavelli 4,
34132 Trieste (Italy)
tel. +39 040 671184
fax  +39 040 671160

###################### original message ##################
Message: 42
Date: Thu, 8 Mar 2012 10:28:40 -0500
From: "Downey, Patrick" <PDowney at urban.org>
To: <r-help at r-project.org>
Subject: [R] Panel models: Fixed effects & random coefficients in plm
Message-ID:
	<0F96478603980B46AAAFBA77069582ED18B39D0C at UIEXCH.urban.org>
Content-Type: text/plain;	charset="us-ascii"

Hello,

I am using {plm} to estimate panel models. I want to estimate a model
that
includes fixed effects for time and individual, but has a random
individual
effect for the coefficient on the independent variable.

That is, I would like to estimate the model:
Y_it = a_i + a_t + B_i * X_it + e_it
Where i denotes individuals, t denotes time, X is my independent
variable,
and B (beta) is the coefficient on that random variable. I want both a
coefficients to be estimated with fixed effects because I expect them to
be
correlated with Y, and B to be estimated using a random effect.

I understand how to estimate two way fixed effects models in plm (which
have the fixed effects for time and individual, as I want) and how to
estimate random coefficient models (which have random effects for all
coefficients, including the intercept and the beta). I want to combine
these, though, and I cannot figure out how to do that. It seems like the
plm package is capable, but I can't figure it out.

Below is reproducible code (assuming you have plm installed) taken from
the
vignette available on cran. It shows the two models I know how to
estimate.
Any guidance on estimating the third model would be greatly appreciated.

-Mitch

# Data setup
library(plm)
data("EmplUK", package="plm")
names(EmplUK)
E <- pdata.frame(EmplUK, index = c("firm", "year"),
drop.index = TRUE,
row.names = TRUE)

# Two-way fixed effects model with constant beta

m1 <- plm(wage ~
output,data=E,model="within",effect="twoways")
summary(m1)

# Random (individual) effects for both intercept and beta

m2 <- pvcm(wage ~ output,data=E,model="random")
summary(m2)
###################### end original message ##################

?
Ai sensi del D.Lgs. 196/2003 si precisa che le informazi...{{dropped:12}}