Hi everyone,
I'm struggling with the negative binomial from part 2. Basically, I have a
function, which I'm pretty sure is okay (or at least close to okay). When I
run optim using the following specs, however, I get the following error
message:
> optim.negbin <- optim(par = rep(0, ncol(indep)), fn = negative.binary,
method = "BFGS", control = list(fnscale = -1, trace = 1, REPORT = 1), y =
pset7$groups, X = indep, hessian = TRUE)
Error in optim(par = rep(0, ncol(indep)), fn = negative.binary, method =
"BFGS", :
initial value in 'vmmin' is not finite
I've tried several combinations and nothing seems to work. The same
configuration works when I run it using my poisson function in part #1. Is
this basically telling me that my function is wrong?
thanks for any help in deciphering this error message,
Maya
Hi,
I'm confused about constructing the estimates and errors for the
counterfactual individuals, and the first differences. My estimates come
out negative (which doesn't make much sense), and I'm not sure how to
calculate the uncertainty intervals manually without using Zelig.
If anyone could help, I would appreciate.
Thanks!
Lucia
Hi all,
I had a quick question on parameterization of the negative binomial. We
have two ways of doing it, with the parameterization of sigmasq in Gary's
notes and the parameterization of theta in Zelig. It seems to me that
despite the different parameterizations and superficially different density
functions, the betas and the expected values should be the same for both.
Is that correct?
I've written both log likelihood functions and unless I made a coding error,
it seems I'm getting two very different sets of betas. The one I get
following the Zelig parameterization matches the output from Zelig.
Thanks
Patrick
hi all,
we've posted the next ps. It's due next Thursday.
Have a good weekend!
Holger
--
Holger Lutz Kern
Graduate Student
Department of Government
Cornell University
Institute for Quantitative Social Science
Harvard University
1737 Cambridge Street N350
Cambridge, MA 02138
www.people.cornell.edu/pages/hlk23
If you haven't already, I suggest you sign up for the political
methodology listserv. To subscribe go to
http://polmeth.wustl.edu/mailinglist/managesubscription.php There aren't a
lot of messages, but some are pretty informative. Its also useful just to
see what people are thinking about.
Gary
hi all,
I have to cancel my office hours this afternoon, but I will be available
via email or on the phone.
I'm sorry for any inconveniences.
Holger
___________________________________
Holger Lutz Kern
Graduate Student
Government Department
Cornell University
Graduate Associate
Harvard University
Institute for Quantitative Social Science
1737 Cambridge Street N350
Cambridge, MA 02138
(607) 227 6563
http://iq.harvard.edu/People/people.php?info=1254&sub=7
Hi all,
please do not forget to give me or Justin the following before the begin
of class today, and to stick around for a few minutes at the end.
For each group,
1) THREE CDs that contain all the data needed to replicate the paper,
the article you chose to replicate as a pdf, electronic codebooks
for the data, and your R code. If you should have several datasets,
please also include a readme file that describes the overall structure
of the data.
2) TWO hardcopies of the article you chose to replicate for Justin and me.
cheers,
Holger
--
Holger Lutz Kern
Graduate Student
Department of Government
Cornell University
Institute for Quantitative Social Science
Harvard University
1737 Cambridge Street N350
Cambridge, MA 02138
www.people.cornell.edu/pages/hlk23
Hi Justin and Holger,
I've been using the latex() command in the Hmisc library in R to make my
tables, which has been pretty simple. My output is very
professional-looking and should be easy for anyone reviewing this portion of
the project (i.e., the teaching team and other groups) to follow.
However, while the numbers match the numbers in our original article, the
table format is different. For example, the original article has a "Table
3" with 5 columns, text explaining how the columns differ, stars for
stargazing, and standard errors presented in parentheses beneath each
coefficient. I have a separate table for each column in Table 1, with
separate columns for coefficients, standard errors, and p-values. Instead
of paragraphs explaining what each column represents, I state in the title
or in a footnote to each table what variables are missing from each
regression and what each column represents.
My output is not presented in the best way to "minimize real estate" in
journals, nor is it presented in the most efficient way for an actual
publication. However, at this stage of this particular project, we have
found that looking at the output produced by the latex() command is easier
to read, and helps us to interpret our replication. We think that
presenting the data in this format will also make it much clearer to others
who do not have fluency in the original article text.
So my question is: at this stage of the project, Do I need to be concerned
with making actual publication-quality tables, or replicating the format of
the author's tables exactly? Or should I be primarily concerned with
presenting the data neatly and in a way that I think will be most useful to
those looking at this stage of the project?
Thanks,
Jill