[Date Prev][Date Next][Thread Prev][Thread Next][Date index][Thread index]
Re: st: Simple Question about Fixed Effects
Bob Rijkers wrote:
> I am using a fixed effects model to analyze the impact of training on
> industrial productivity. When I include both training and a lagged value
> of training, however, Stata refuses to execute fixed effects regression.
> Why should this be the case?
There are a number of ways in which one can run a fixed-effect regression
in Stata. Unfortunately, you don't say which one. My preferred option is
to use David Roodman's -newey2- routine, downloadable from SSC: this way,
you can obtain OLS parameter estimates controlling for both
autocorrelation and heteroscedasticity in the residuals. A dynamic panel
model option is another of David's user-written routines: -xtabond2-, also
available from SSC (however, suitable variables need to be found which can
be 'instrumented' with the lagged variable). If the only problem is
autocorrelation, use -prais, corc- (which does _not_ produce OLS
> Clearly, including lags of explanatory variables induces autocorrelation
> amonths the explanatory variables, but why is this problematic?
An unpublished paper by Christopher Achen (2000) makes a number of points
that amount to a very strong argument against using lagged dependent
variables (LDVs) under _any_ circumstances:
(1) including LDVs often bias the coefficients of other key predictor
variables toward negligible values that they would not otherwise take were
(2) because of (1), coefficient values of LDVs are often grossly inflated;
(3) conditions (1) and (2) often happen when serial correlation is high
and the exogenous variables are heavily trended, as is often the case in
panel data; and
(4) occasionally, other explanatory variables take on the wrong sign when
LDVs are included.
A .pdf file of the paper can be downloaded from the link at the bottom of
CLIVE NICHOLAS |t: 0(044)7903 397793
Politics |e: firstname.lastname@example.org
Newcastle University |http://www.ncl.ac.uk/geps
Achen CH (2000) "Why Lagged Dependent Variables Can Suppress the
Explanatory Power of Other Independent Variables", paper presented to the
Annual Meeting of the Political Methodology Section of the American
Political Science Association, University of California at Los Angeles,
July 20-22 (http://polmeth.wustl.edu/workingpapers.php?year=2000).
* For searches and help try: