Dear Carola,
I agree that inclusion of a LDV in the analysis of TSCS data is common
in our discipline. Note that including a LDV means to set up a
difference equation and the solution to a difference equation is a
time function. This means that the coefficients in your "LDV" model
have to be interpreted differently: They are merely one period
effects. If you want to know the "net" or cumulative effect of a
change in X_t, you have to take into account that this induces a
change in Y_t, which itself is used to "explain" changes in Y_t+1, and
so on. So you need to evaluate the effect of a change in X_t on Y over
time (in a VAR setting this is called the impulse response function).
It may well be that a variable which has no significant one period
effect turns out to have a significant and substantial cumulative
impact once you take the time dimension into account. Especially, if
the coefficient of the lagged dependent variable is substantial and in
almost any political science application I have seen this is indeed
the case.
Regarding your question on how to decide whether to include a LDV or
not, there are in my opinion two positions. The first requires a good
theoretical reason for why it should be included. From this point of
view, LDVs are used quite flippantly. The second emphasizes the
dynamics and that we can hardly be sure that the relationships we
study do not have a time dimension (i.e., a change in say X at t=1
causes only a change in Y_1 and this is inconsequential for all Y_t
with t>1). If you follow the first argument, you are probably quite
reluctant to include a LDV, I guess. If you follow the second
argument, you would be more likely to include a LDV.
Reading: A starting point is Keele/Kelly (2006 in Political Analysis
14: 186–205). They discuss the dynamic nature of this kind of
specification and assess bias due to introducing a LDV. There are
several other relevant articles in this issue of Political Analysis
(it was a special issue on analysis of TSCS data). Basic reading
related to the bias introduced by a LDV in a model with a fixed
effects Nickell's (1981) "Biases in Dynamic Models with Fixed
Effects" (Econometrica 49 (6): 1417-1426). Also, apart from
introductory chapters of standard time series textbooks, I think there
is a book forthcoming by John Freeman and co-authors on dynamic
analysis in the social sciences.
Hope that helps.
Yours
Michael Bechtel
--
ETH Zurich
Dr. Michael M. Bechtel
Center for Comparative and International Studies
WEC C 25
Weinbergstrasse 11
CH-8092 Zurich
[log in to unmask]
www.ib.ethz.ch/people/mbechtel
+41 44 632 62 68 phone
+41 44 632 12 89 fax
--------------------------------------------------------------------------------------------------------------------------------
Am 15.02.2009 um 20:28 schrieb Carola Herrera:
> Hi,
>
> I've run a country FE model and found that the coefficients
> previously found signficant in the pooled model are robust as long
> as the lagged dependent variable is included. I'm using a lagged dv
> because the dv is expenditure data and it seems that the previous
> year level of expenditure belongs in the model, plus (following Beck
> and Katz's and other's advice) it helps address the
> serialcorrelation problem.
> I know LDV may cause problems in FE models so I have run the FE
> model with the LDV and without it. As I said, the estimates are not
> robust to dropping the LDV in the FE model...So, what shall I
> conclude? Is it that the LDV really biases my FE estimates or is it
> that for some reason it belongs in the model and dropping it causes
> omitted variable bias?
>
> Thank you for your help.
>
> Carola
>
>
>
>
>
>
>
> **********************************************************
> Political Methodology E-Mail List
> Editors: Melanie Goodrich, <[log in to unmask]>
> Delia Bailey, <[log in to unmask]>
> **********************************************************
> Send messages to [log in to unmask]
> To join the list, cancel your subscription, or modify
> your subscription settings visit:
>
> http://polmeth.wustl.edu/polmeth.php
>
> **********************************************************
**********************************************************
Political Methodology E-Mail List
Editors: Melanie Goodrich, <[log in to unmask]>
Delia Bailey, <[log in to unmask]>
**********************************************************
Send messages to [log in to unmask]
To join the list, cancel your subscription, or modify
your subscription settings visit:
http://polmeth.wustl.edu/polmeth.php
**********************************************************
|