| ![](/includes/contimages/spacer.gif) |
[Date Prev][Date Next][Thread Prev][Thread Next][Date index][Thread index]
Re: st: An econometric question
As I understand, his original setup is to put "repeated observations" of
the averages for each country, but not directly putting z1, z2 (the time
varying variables) into the regression.
SamL wrote:
Why not?
I mean, if I understand correctly, the real data is 100 observations every
year, for 15 years. The averages are throwing away information.
Admittedly, one might want to do something to account for the clustering
of cases. Lots of options there--time series models, two-way fixed
effects, and so on, all of which would likely raise the standard errors of
Z1, Z2, and so on to some extent to reflect the loss of precision that
attends the inclusion of more terms in a model. But, none of those
approaches will just throw away the information that is in the 15 year
variation in the data, variation that could be useful (e.g., variation in
gas prices in the United States, 1975-1990)
Am I misunderstanding the original question?
Sam
On Sun, 8 Apr 2007, David Greenberg wrote:
No, it is not. David Greenberg, Sociology Department, New York University
----- Original Message -----
From: "Roy,Suryadipta" <[email protected]>
Date: Sunday, April 8, 2007 8:54 pm
Subject: Re: st: An econometric question
To: [email protected]
Maarten,
Thanks so much for your response! Now just to make sure
that I am understanding you right, there were initially
1500 cases (no. of year-countries) which was subsequently
reduced to 100 cases (no. of countries) by taking the
averages of the variables over 15 years. In this
situation, will it be alright to report the results I am
getting for the regression with 1500 observations (the
ones that are significant)?
Thanks again!
Suryadipta.
On Sun, 8 Apr 2007 20:25:18 +0100 (BST)
Maarten buis <[email protected]> wrote:
--- "Roy,Suryadipta" <[email protected]>
wrote:
I have some questions on the interpretation of my
results.
I have data on (say) 100 countries over a period of 15
years, i.e. a total of 1500 observations. Suppose, I am
running a model of the form:
y= a0 + a1x1 + a2x2 + a3x3, where y, x1 and x2 are the
mean values of the original variables (say, z, z1 and
z2)
over 15 years for all countries and x3 is a
time-constant
variable (say, a region dummy).
Now, when I am running the above regression in the
original dataset (with 1500 observations), I am getting
significant results for x1, x2 and x3. Of course, all
the
variables remain unchanged for any country over these 15
years.
On the other hand, if I "collapse" the dataset and run
the
regression with 100 observations (for 100 countries),
none
of the variables remain significant and the r-square
goes
down as well. The mean values for y, x1 and x2 (and x3)
are however, the same in both the regressions.
I would greatly appreciate some help if understanding as
to why I am not getting the same results as in the first
regression.
If you have more cases, you have more bits of
information going into
your estimate, so you are more confident about your
results, so you are
willing to say that smaller deviations from your null
hypothesis are
non-accidental, i.e. significant. So the question is: If
you want to
regress averages over 15 years of a 100 countries, do
you have a 100
cases (number of countries), or 1500 cases (number of
year-countries)?
The answer (unfortunately) is that you have only a 100
cases. By using
year-countries (1500 cases) you are just duplicating the
same
information, in other words, each year does not add any
new info.
Hope this helps,
Maarten
-----------------------------------------
Maarten L. Buis
Department of Social Research Methodology
Vrije Universiteit Amsterdam
Boelelaan 1081
1081 HV Amsterdam
The Netherlands
visiting address:
Buitenveldertselaan 3 (Metropolitan), room Z434
+31 20 5986715
http://home.fsw.vu.nl/m.buis/
-----------------------------------------
___________________________________________________________
Inbox full of unwanted email? Get leading protection and
1GB storage with All New Yahoo! Mail.
http://uk.docs.yahoo.com/nowyoucan.html
*
*
* For searches and help try:
* http://www.stata.com/support/faqs/res/findit.html
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/