Hi Everyone,
A question has come up regarding the coding of MUSCATH in problem 4.
It is coded as
MUSCATH = 0 if Catholic
MUSCATH = 1 if Muslim
MUSCATH = 5 if missing
In the future I won't answer additional questions about the coding the
variables in the Fish dataset. Everything that I know know about it
can be figured out by reading the article and by maintaining the SPSS
information originally encoded in the file. To see this information
set use.value.labels=TRUE and to.data.frame=FALSE in the call to
read.spss() used to input the data and then look at what was read in.
Note that this will put the data in a format that can't directly be
used by R model fitting functions.
Best,
Kevin
------------------------------------------------------
Kevin Quinn
Assistant Professor
Department of Government and
Center for Basic Research in the Social Sciences
34 Kirkland Street
Harvard University
Cambridge, MA 02138
Hi Everyone,
Something that may be of use to those of you using emacs / Xemacs on
the ice machines is the emacs / Xemacs recover-file command. This can
be used to restore files you were working on at the time an emacs /
Xemacs session died. For more information see:
http://www-2.cs.cmu.edu/cgi-bin/info2www?(emacs)Recover
and
http://docsrv.sco.com:8457/cgi-bin/info2html?(xemacs.info.gz)Recover
Best,
Kevin
------------------------------------------------------
Kevin Quinn
Assistant Professor
Department of Government and
Center for Basic Research in the Social Sciences
34 Kirkland Street
Harvard University
Cambridge, MA 02138
Hi Everyone,
Someone recently asked:
> I am working on problem 4 of the final and I have a question, please
> let me know if you can answer it! I am getting only 19 observations
> coded as 1 for MUSLIM and 70 observations coded as 0 for MUSLIM - so
> only 89 countries are coded as being muslim or not. On Fish's table
> 1 he has a list of 47 islamic countries and he shows that there are
> 109 nonislamic. If you can't use MUSLIM to create the subsets, is
> there a reason that MUSLIM is even in his data set? Without the
> codebook I am not sure if this is the variable - but it seems like
> the only one that would be used to divide up the countries.
If you look at the commented out information at the top of the data
file (see
http://www.courses.fas.harvard.edu:80/~gov1000/Final_Exam/Fish_data.sav)
you can see that the description of the MUSLIM variable is:
MUSLIM /predominantly Muslim population, dummy variable
You might want to try reloading your data-- there should be plenty of
observations with MUSLIM==1.
Hope this helps.
Best,
Kevin
Hi Everyone,
Another question of clarification on problem 2 has come up. In an
earlier email about this problem I wrote:
> A question of clarification on final exam problem 2 has come up. Let
> epsilon_i and epsilon_j denote the disturbances for observations i
> and j. The phrase "independent and identically distributed normal
> disturbances" means that epsilon_i is statistically independent of
> epsilon_j for i not equal to j and that epsilon_i is a draw from a
> normal distribution with some constant mean and constant variance
> for i = 1,...,n. Here n is the number of observations. In other
> words, you can think of all the disturbances as independent draws
> from the same normal distribution.
The second half of this means that all of the disturbances have the
same mean and variance. They are identically distributed.
Hope this helps. Please let me know if this is not clear.
Best,
Kevin
------------------------------------------------------
Kevin Quinn
Assistant Professor
Department of Government and
Center for Basic Research in the Social Sciences
34 Kirkland Street
Harvard University
Cambridge, MA 02138
Hi Everyone,
There has been a question about final exam problem 4. This problem
asks you to consider several things-- one of which is whether you can
replicate Fish's results. What I'm looking for here is whether you can
use the same methods Fish used to get the same estimates that he got.
If you can't replicate his results I want you to discuss why you
can't.
Hope this helps.
Best,
Kevin
------------------------------------------------------
Kevin Quinn
Assistant Professor
Department of Government and
Center for Basic Research in the Social Sciences
34 Kirkland Street
Harvard University
Cambridge, MA 02138
Hi Everyone,
A question of clarification on final exam problem 2 has come up. Let
epsilon_i and epsilon_j denote the disturbances for observations i and
j. The phrase "independent and identically distributed normal
disturbances" means that epsilon_i is statistically independent of
epsilon_j for i not equal to j and that epsilon_i is a draw from a
normal distribution with some constant mean and constant variance for
i = 1,...,n. Here n is the number of observations. In other words,
you can think of all the disturbances as independent draws from the
same normal distribution.
Hope this helps.
Best,
Kevin
------------------------------------------------------
Kevin Quinn
Assistant Professor
Department of Government and
Center for Basic Research in the Social Sciences
34 Kirkland Street
Harvard University
Cambridge, MA 02138
Hi Everyone,
Some questions have come up on question 3. I don't want to say too
much here b/c part of the question is designed to make you think about
how to move from a verbal desciption of a null hypothesis to an actual
test of that null hypothesis.
That being said, I can say a few things now. First, the "after
adjusting for the other covariates in the model" language can be
intepreted to mean "holding the other covariates in the model fixed at
any arbitrary values". Second, you will need to think carefully about
what is being asked in parts b), c), and d). One or more of these
questions may be more complicated than they first appear.
Best,
Kevin
------------------------------------------------------
Kevin Quinn
Assistant Professor
Department of Government and
Center for Basic Research in the Social Sciences
34 Kirkland Street
Harvard University
Cambridge, MA 02138
Hi Everyone,
The final exam is now posted on the course website under the Final
Exam heading. If you have questions about any of the exam questions
you should email Alison, Ryan, and me directly-- please do not use the
course listserv to ask questions.
Good luck. Alison, Ryan, and I enjoyed having you in class.
Best,
Kevin
------------------------------------------------------
Kevin Quinn
Assistant Professor
Department of Government and
Center for Basic Research in the Social Sciences
34 Kirkland Street
Harvard University
Cambridge, MA 02138
Hi Michael,
Data points with large robust Mahal. distances and large residuals are
likely bad leverage points. These data points don't look like the bulk
of the (y, X) data and would likely have a large effect on the OLS
results. The MM estimator will usually automatically downweight these
observations so you can believe the MM estimates. On the other hand,
OLS results will be very sensitive to the inclusion of these data
points. This plot with robust residuals and robust Mahal. distance is
just a better way to identify highly influential data points for OLS
than the traditional diagnostics.
Hope this helps.
Best,
Kevin
On Tue, 4 Jan 2005, Michael William Nitsch wrote:
> Hi Professor Quinn,
>
> One question about the interpretation of the leverage plot we used in
> problem set 10 for the MM estimator (the one that plots robust studentized
> residuals again the chi sqare):
>
> How are we to interpret a cluster of
> points that appears in one of the "critical regions" of the plot, ie
> beyond the 97.5% confidence interval and with large robust
> studentized
> residuals? Are these points that the MM estimator will downweight when
> making it its coefficient estimates, or are they points that will present
> problems for the MM estimator (in the same way that points in the critical
> regions of traditional leverage plots present problems for the OLS
> estimator).
>
> Thanks,
> Michael
>