Hello;
I have been trying to run amelia() including the ts and cs arguments.
Amelia accepts the ts argument, but keeps rejecting the cs argument. I keep
getting this error message:
> amelia(panel, ts=panel$period, cs=panel$country, outname="xt")
Amelia Error Code: 6
The 'cs' variable is out of the range of
possible column numbers or is not an integer.
$code
[1] 6
panel = panel data set
country is a number assigned to identify each country (i.e. 1 = Afghanistan,
2=Angola, etc.)
Does anyone have an idea what I am doing wrong?
Thanks
Joe
Hi all,
When you *match with replacement of the controls*, should you incorporate
any kind of correction to your standard errors? If you have 100 treated
units and 50 unique controls, you'll only have 150 real observations;
matching with replacement of controls will give you 200 total units. That
inflates your degrees of freedom with units that don't really exist. How
does this work? And does the correction depend on what percentage of units
are in fact unique?
Thanks!
John
**
Good morning everyone,
I had a bit of a rough night and need some help answering the questions
below:
#1 - Is anyone aware of a package or canned function in R that will
implement Prais-Winsten Regression?
#2 - Is OLS with PCSE (panel corrected standard error) just a variant of
FGLS (feasible generalized least squares)?
#3 - Does lagging the dependent variable in addition to correcting for first
order autocorrelation produce bias estimates?
Background-
#1 - Unable to exactly replicate my author's work, I downloaded Stata and
tried to replicate his work in that program. With a non-trivial amount of
help, I was finally able to exactly replicate the author's results. The
author claimed he used OLS with Panel Corrected Standard Errors. Stata
reports that it conducted a Prais-Winsten regression correcting for panel
specific first order autocorrelation.
#2 - I am a bit confused as to whether or not PW regression is the same as
OLS. OLS returns different point estimates than the PW regression (hence
the trouble I have had replicating the author's work). The many articles I
read last night seemed to indicate PW regression is a variant of FGLS. The
author cites Beck and Katz as the reason for conducting OLS with PCSE.
However, Beck and Katz specifically warn against using FGLS.
#3 - I have read many articles about the correctness of using a lagged
dependent variable and many other articles about the appropriateness of
correcting for autocorrelation. I have not read an article that advocated
or warned against doing both.
Thanks for your time; I appreciate any help light you can shed on my
intuitional problems..
Joe
Dear list,
We're trying to use multiply imputed data to estimate a random effects logit.
Has anybody had success doing this? We get the following error message:
summary(AM1t1c1)
Error in apply(coef1, 1, mean) : dim(X) must have a positive length
In addition: Warning messages:
1: In x$coef : $ operator is invalid for atomic vectors, returning NULL
2: In x$coef : $ operator is invalid for atomic vectors, returning NULL
3: In x$coef : $ operator is invalid for atomic vectors, returning NULL
4: In x$coef : $ operator is invalid for atomic vectors, returning NULL
But running a regular logit works fine.
Thanks,
Didi and Shahr
We're trying to run a linear mixed effects model with 3 random
effects: M1, M2 and Fam. M1 & M2 are dummies. We're actually
interested in the variances of the random coefficients (& the
remaining error). The variances of the random coefficients for M1 and
M2 need to be set as equal.
The following code produces equal variances for M1 and M2 together
with error, but we want an additional random coefficient at a Family
level.
test3 <- lme(g~SES, random = list(Fam = pdIdent(~M1 + M2 -1)), data
= ToyLong, method = "ML")
Any ideas how we should amend this code?
Jeremy
Dr Jeremy Hodgen
Senior Lecturer in Mathematics Education
King's College London
Department of Education and Professional Studies
Franklin-Wilkins Building
Waterloo Bridge Wing
150 Stamford Street
London SE1 9NH
Tel: 020 7848 3102
Fax: 020 7848 3182
E-mail: jeremy.hodgen at kcl.ac.uk
Has anyone run across an R package that does logit models for TSCS or BTSCS
data? I started writing my own code, but reinventing the wheel isn't
terribly efficient...
Thanks,
J
Hi everyone,
Quick question: I downloaded continuous data from the Penn World Tables, but
R treats the observations as "factorized" or categorical within a variable.
Consequently, I'm not able to include the variables in regressions; R
essentially treats each observation within these variables as dummies. I've
attempted to correct the problem with as.numeric, as.data.frame, and
as.matrix commands, but have not been successful.
Any thoughts? Thanks in advance,
Brett
In part 1b, the Zelig output for first differences bears no
resemblance to the difference in the expected values. Anyone have an
idea why this would be the case? I thought the FD process was just to
give us a standard error for the difference between the expected
values.
--
Jon Bischof
Graduate Student
Department of Government
Harvard University
Sorry to spam the list today -- coding issues are bugging me.
> coop1 <-
setx(op,COOP=1,fn=list(numeric=median,ordered=median,others=median))
*Error: C stack usage is too close to the limit
*I'm trying to set all values to the median -- what's my issue here?
Thanks,
J
Are people getting the *exact *answers that Zelig gives you from the
hand-coded poisson and negative binomial models? I've got it to within a few
thousandths, but I can never quite get the exact answer. Also (and I think
this could be the cause of the problem), I'm getting a number of warning
messages that the parameter of lgamma is out of its range; does anyone else
have this issue?
Thanks,
John