[Limdep Nlogit List] can i pick your brains?

Eaton, Derek Derek.Eaton at wur.nl
Wed Mar 25 20:19:24 EST 2009


Having just dealt with another 'unconventional' panel data set (in my
case T=17) and a low signal-noise ratio with explanatory vars, I found
that I should have done more graphical exploratory analysis of my
dependent variable at the beginning before letting a barrage of models
loose on the data. When I finally did - and it took me a day just to get
the kind of plots I needed (in the end, I did this with R, not LIMDEP) -
a number of approaches were quickly eliminated.

In terms of panel data models, you may also want to experiment with
fixed effects models with both individual and time effects (2-way), and
even an interaction between them (see E11.12 in manuals). And your
problem seems to suggest a dynamic formulation, so the Arellano, Bond
and Bover GMM model (E11.15) is a possibility.

It sounds though like you have a more fundamental challenge. You wish to
explain changes in airline service over time at different airports but
do not have variables that exhibit this type of variation. Perhaps you
can look for logical interaction terms between an i-dominant variable
and a t-dominant one? You may also trying to see if you can successfully
explain total flights (aggregate across airports) with your t-dominant
variables first. If not, it seems unlikely you will get further at a
disaggregated level.

Derek Eaton
Derek Eaton 
Researcher, International Trade & Development 
Agricultural Economics Research Institute (LEI) 
part of Wageningen University & Research Centre 
t: +31-(0)70-335 82 43
p: P.O. Box 29703, The Hague, 2502LS, The Netherlands 

-----Original Message-----
From: limdep-bounces at limdep.itls.usyd.edu.au
[mailto:limdep-bounces at limdep.itls.usyd.edu.au] On Behalf Of Durham,
Sent: maandag 23 maart 2009 20:32
To: Limdep and Nlogit Mailing List
Subject: Re: [Limdep Nlogit List] can i pick your brains?

William you might want to try the TSCS panel estimator and use the
options that allow for autocorrelation over time and group=wise
heteroscedasticty etc. It's especially useful with more time periods
than cross-sections, but I think t=20 is enough to gain from the
autocorrelation options. Given your limited cross-section variation over
time in some variables you should be very thoughtful about your use of
cross-section dummy variables; they will collect much of the information
in those cross-section variables that don't change much. At least try it
with and without those dummies, before you accept findings of
insignificance for those cross-section variables. This is even more
critical if you are testing lagged dependent variables which can easily
be biased in this framework. If you are let me know, I have some
references I can recommend. Cathy

From: limdep-bounces at limdep.itls.usyd.edu.au
[mailto:limdep-bounces at limdep.itls.usyd.edu.au] On Behalf Of Bill Spitz
Sent: Monday, March 23, 2009 9:57 AM
To: listserv LIMDEP
Subject: [Limdep Nlogit List] can i pick your brains?

This may not be the best forum for this, but I'm trying to figure out a
reasonable modeling approach for the following:

I'm trying to explain changes in levels of airline service (measured as
flights per day) across many different airports over about 20 years. So
I have a panel set of observations y(it) for airport i and time t.

Here's the rub. Most of the candidates for explanatory variables are
-- "i-dominant" variables, ie., they vary significantly from one airport
to another, but hardly at all over time, eg. local real income,
population, distance to nearest hub, etc, (these serve to essentially
provide the scale of operations at the airport); OR --"t-dominant", they
vary significantly over time, but not by airport, eg. fuel prices,
airline costs, etc.

So I don't really have the traditional x(it)-type explanatory variables
that vary in both dimensions where one would employ a standard panel
estimator. I have quickly gotten lost in the LIMDEP manuals with all the
different possible panel models that are available.

Any insights into how best to approach this from a modeling standpoint
would be appreciated.

William Spitz

Privileged/Confidential information may be contained in this message.
If you are not the addressee indicated in this message (or responsible
for delivery of the message to such person), you may not copy or deliver
this message to anyone. In such case, you should destroy this message
and notify GRA, Inc. (postmaster at gra-inc.com) immediately. Please advise
immediately if you or your employer do not consent to Internet e-mail
for messages of this kind. Opinions, conclusions and other information
expressed in this message are not given or endorsed by GRA, Inc. unless
otherwise indicated by an authorized representative independent of this
Limdep site list
Limdep at limdep.itls.usyd.edu.au
Limdep site list
Limdep at limdep.itls.usyd.edu.au

More information about the Limdep mailing list