I am curious if some one who has played with Zelig for a while knows this.
We ran a linear model earlier today that used country fixed effect
variables that were collinear. We didn't realize this until running
it through stata and seeing that stata's linear regression dropped the
6 relevant independent variables.
Further inspection of our zelig output shows that it dropped six of
the fixed effect variables created by including as.factor(nation). It
then reported the point estimates and other goodies for all specified
independent variables and the remaining fixed effect variables.
What mechanism does R use to select the best submodel to run once a
singularity prevents it from doing its X'X thing. Not a
mission-critical question, just curious.
thanks,
/\llan
Andy's right. Include INC in the model specification.
----- Original Message -----
From: "Andrew Eggers" <aeggers(a)gmail.com>
To: "Olivia Lau" <olau(a)fas.harvard.edu>
Sent: Monday, November 15, 2004 12:45 AM
Subject: ps7q2
> Olivia,
>
> Did you mean for us to include "INC" in the model in question
> 2? The
> 2(c)1 scenarios talk about the incumbent being a Democrat, but
> the
> incumbent's party is not in the variables we used, as far as I
> can
> tell. I guess it doesn't make much of a difference, since you
> still
> have variation between scenarios.
>
> Andy
>
No, this is a good question, because the Zelig help page is wrong. 80
So remember that the distributional assumption for the probit model are
\tilde{pi} = EV = Phi(mu) where mu = x %*% beta
mu is just the parameterization of the systematic component. It's not the
mean of the distribution. Phi is the standard normal CDF with mean 0 and
sigma2 = 1. So you want to find out what's the probability of observing x
%*% beta, given that you think the distribution is Normal(0, 1).
I'll fix that page shortly...
Olivia.
On Sun, 14 Nov 2004, Kate Emans wrote:
> Hi Olivia,
> I don't think this is a great question, but I'm stumped on this one.
> For the log-likelihood of the probit, we know (e.g. from Zelig online help)
> that pie = CDF(xi'B) where this is the CDF of a variable mu, distributed
> normally.
>
> I don't understand what mu is. If we set mu=xi'B, then CDF(mu)=.5 all the
> time. . .
> If we set mu= 0, then this is clearly wrong.
>
> Help! Any hints on mu? Is it the mean of all the xi'B's? is it a
> parameter to estimate?
>
> Thank you!!!!
> Kate
>
>
hi guys,
Here is some code (not finished) for problem #1. I think most of it works,
but I'm sorry because I think there is some junk in here too. Who knows,
hopefully it will help. Maybe we should still try to meet on tuesday?
Mostly I hope it will help Allan, who i just heard has lost a family member
this weekend. Allan, I'm really sorry to hear about this and my deep
condolences and let us all know if we can help you somehow. Maybe they
will give you an extension on this one?
--Kate
ok, everything is posted now...thanks for being patient, people.
---------- Forwarded message ----------
Date: Thu, 11 Nov 2004 17:54:49 -0500
From: Kate Emans <emans(a)fas.harvard.edu>
To: Olivia Lau <olau(a)fas.harvard.edu>
Subject: uh oh, data not in the folder
Um, the data file for Prob. Set #7 is not in the "data" folder on the
course site. Am I really the first one to look for it this week? Is it
hiding somewhere else?
Hi,
I have been told that the presentation is changed to Littauer M-16. Still
Friday at noon. -- Ben
Ben Goodrich
Graduate Student at Harvard University
Ph.D. Program in Government and Social Policy
www.people.fas.harvard.edu/~goodrich/
goodrich(a)fas.harvard.edu
> -----Original Message-----
> From: Ben Goodrich [mailto:goodrich@fas.harvard.edu]
> Sent: Thursday, November 11, 2004 4:27 AM
> To: 'gov2001-l(a)lists.fas.harvard.edu'
> Subject: Methods Presentation Friday
>
> Hi everyone,
>
> I am presenting a methods paper this Friday at 12:00 (pizza) in Littauer
> M-17 (next to the Science Center). You will probably get something out of
> the presentation that you can apply to your Gov2001 paper if your
> replication involves "clustered" data (the same people/countries over time
> or something like that). The presentation is going to focus on least
> squares estimators but will address probit etc. briefly.
>
> Paper link: http://www.people.fas.harvard.edu/~goodrich/Files/Goodrich.pdf
>
> Hope to see you there,
> Ben
>
> Ben Goodrich
> Graduate Student at Harvard University
> Ph.D. Program in Government and Social Policy
> www.people.fas.harvard.edu/~goodrich/
> goodrich(a)fas.harvard.edu
It means that you have the wrong number of starting values in your optim
statement.
----- Original Message -----
From: "Kentaro Fukumoto" <fukumoto(a)dg8.so-net.ne.jp>
To: <gov2001-l(a)lists.fas.harvard.edu>
Sent: Thursday, November 11, 2004 12:10 AM
Subject: [gov2001-l] non-finite value supplied by optim
> Good evening all,
>
> I am now making a new model
> but get the following warning.
>
>> alogit(y ~ xn, ~ zn, n, data)
> Error in optim(c(start, rep(0, (k2 + 2))), llik.alogit, method
> = "BFGS", : non-finite value supplied by optim
>
> Does it mean MLEs become inifinity?
> And what can I do?
>
> Kentaro
>
> _______________________________________________
> gov2001-l mailing list
> gov2001-l(a)lists.fas.harvard.edu
> http://lists.fas.harvard.edu/mailman/listinfo/gov2001-l
>
Olivia.
Good evening all,
I am now making a new model
but get the following warning.
> alogit(y ~ xn, ~ zn, n, data)
Error in optim(c(start, rep(0, (k2 + 2))), llik.alogit, method = "BFGS", :
non-finite value supplied by optim
Does it mean MLEs become inifinity?
And what can I do?
Kentaro
Good question, Jason.
For #2: Optimze over s^2, but treat nu as fixed and optimize
iteratively by increasing nu in small increments, e.g. for nu =
1:1000.
----- Original Message -----
From: "Jason Morris Lakin" <jlakin(a)fas.harvard.edu>
To: <olau(a)fas.harvard.edu>
Sent: Sunday, November 07, 2004 11:30 AM
Subject: RE: #2
> hey olivia. i don't really understand problem 2. are you
> asking us not to
> maximize over v (as we do in prob 1 with sig2)? i was treating
> v and s as
> ancillary parameters like sig2 in prob 1. but then you can't
> maximize
> given different v's. so is just s an ancillary parameter?
> should i
> maximize across only 5 instead of 6 parameters?
>
> thanks
> j
>