Gov2001 April 2008

gov2001@lists.gking.harvard.edu

31 participants
45 discussions

[gov2001-l] zelig and lm gives me different results with mi

by keith.schnakenberg＠gmail.com

This isn't directly related to class, but I thought someone might be able to give me some idea why this is happening. I've run the same model with zelig and with lm. I'm getting differences that are substantially different. The output and calls are below. Theories? Imputation: mi3 <- amelia(data=m3sample, m=10, p2s=0, frontend=FALSE, idvars="WTSTRAT", startvals=0, tolerance=0.0001, outname="m3outdata", write.out=TRUE) m3.1 <- read.csv("m3outdata1.csv") m3.2 <- read.csv("m3outdata2.csv") m3.3 <- read.csv("m3outdata3.csv") m3.4 <- read.csv("m3outdata4.csv") m3.5 <- read.csv("m3outdata5.csv") m3.6 <- read.csv("m3outdata6.csv") m3.7 <- read.csv("m3outdata7.csv") m3.8 <- read.csv("m3outdata8.csv") m3.9 <- read.csv("m3outdata9.csv") m3.10 <- read.csv("m3outdata10.csv") ZELIG: Model: ls Number of multiply imputed data sets: 10 Combined results: Call: zelig(formula = ACCULT ~ TIME + PCTH2 + CONTACT + SEND + VISIT2 + LIVED + MEXICAN + MALE + EDU + INCOME + AGE, model = "ls", data = mi(m3.1, m3.2, m3.3, m3.4, m3.5, m3.6, m3.7, m3.8, m3.9, m3.10), weights = "WTSTRAT") Coefficients: Value Std. Error t-stat p-value (Intercept) 4.32459 2.38034 1.81679 7.004e-02 TIME 0.25307 0.03718 6.80730 1.874e-10 PCTH2 -5.02070 1.65078 -3.04141 2.533e-03 CONTACT 0.73485 0.14803 4.96431 1.127e-06 SEND -2.00209 0.75850 -2.63953 8.410e-03 VISIT2 0.17439 0.21159 0.82417 4.100e-01 LIVED -0.51087 0.85970 -0.59424 5.526e-01 MEXICAN -1.02115 1.14343 -0.89306 3.730e-01 MALE 0.06435 0.69695 0.09233 9.265e-01 EDU 2.11346 0.35462 5.95983 6.801e-09 INCOME 0.99713 0.25999 3.83522 1.713e-04 AGE -0.19368 0.03041 -6.36831 2.627e-10 For combined results from datasets i to j, use summary(x, subset = i:j). For separate results, use print(summary(x), subset = i:j). LM: Call: lm(formula = ACCULT ~ TIME + PCTH2 + CONTACT + SEND + VISIT2 + LIVED + MEXICAN + MALE + EDU + INCOME + AGE, data = mi(m3.1, m3.2, m3.3, m3.4, m3.5, m3.6, m3.7, m3.8, m3.9, m3.10), weights = WTSTRAT) Residuals: Min 1Q Median 3Q Max -11.8463 -3.6531 -0.0422 2.9975 15.3804 Coefficients: Estimate Std. Error t value Pr(>|t|) (Intercept) 2.8595 2.1757 1.31 0.1899 TIME 0.2472 0.0321 7.69 3.3e-13 *** PCTH2 -4.4872 1.5051 -2.98 0.0031 ** CONTACT 0.8163 0.1340 6.09 4.1e-09 *** SEND -2.0498 0.7187 -2.85 0.0047 ** VISIT2 0.1259 0.1999 0.63 0.5294 LIVED -0.3668 0.7874 -0.47 0.6417 MEXICAN -0.9188 0.9988 -0.92 0.3585 MALE 0.1749 0.6460 0.27 0.7868 EDU 2.2613 0.3196 7.08 1.5e-11 *** INCOME 1.0732 0.2264 4.74 3.6e-06 *** AGE -0.1902 0.0290 -6.57 2.9e-10 *** --- Signif. codes: 0 ?***? 0.001 ?**? 0.01 ?*? 0.05 ?.? 0.1 ? ? 1 Residual standard error: 5 on 253 degrees of freedom Multiple R-Squared: 0.637, Adjusted R-squared: 0.621 F-statistic: 40.3 on 11 and 253 DF, p-value: <2e-16

16 years

[gov2001-l] loading ussu

by ltai＠post.harvard.edu

Hi all, I am having trouble loading ussu.rdata. When I say ussu<-load("ussu.rdata"), typing in ussu yields ".Traceback" "ussu" . I've looked at the file in a text editor and in Excel, and I can't see the table in the column of numbers. If anyone else has successfully loaded the file, I would appreciate his or her suggestions. Thanks, Laurence

16 years

[gov2001-l] 1C table

by jdkuo＠fas.harvard.edu

Dear List, In problem 1 part C when it says "in a table, present the following," am I right in thinking that the question just asks for one number? (like, one percentage?) Or is this table supposed to have 78 rows with True/False statements for each observation? Thanks Didi

16 years

[gov2001-l] negative binomial troubles

by keith.schnakenberg＠gmail.com

Hello classmates, I am running into a snag with my negative binomial log likelihood function that I cannot seem to fix. My code and error are below: > #NEGATIVE BINOMIAL LOG-LIKELIHOOD FUNCTION > ll.negbin <- function(par, X, Y){ + theta <- X%*%par[1:ncol(X)] + gamma <- par[(ncol(X)+1)] + sigma2 <- exp(gamma) + out <- sum(lgamma((theta/(sigma2 - 1)) + 1) - lgamma(Y+1) - lgamma (theta/(sigma2 -1)) + Y*log((sigma2-1)/sigma2) - (theta/(sigma2 - 1)) *log(sigma2)) + } > > #OPTIMIZE > opt <- optim(c(0,0,0,0,0), ll.negbin, X = X, Y = Y, method = "BFGS", control = list(fnscale = -1), hessian = TRUE) Error in optim(c(0, 0, 0, 0, 0), ll.negbin, X = X, Y = Y, method = "BFGS", : initial value in 'vmmin' is not finite So there's some problem with my log likelihood function, but I'm not sure what it is. To elaborate on the problem, here are the results when I evaluate the function outside of optim: > test <- ll.negbin(par=c(0,0,0,0,0), X=X, Y=Y);test [1] NaN What is my problem here? Thank you, Keith

16 years

[gov2001-l] rank-deficient

by keith.schnakenberg＠gmail.com

Did anyone else get this warning? > snctmodel <- zelig(RES ~ IMPORT + COST + TARGET + COOP + TARGET*COOP, model="oprobit", data=snct) Warning message: In function (formula, data, weights, start, ..., subset, na.action, : design appears to be rank-deficient, so dropping some coefs What causes this? Does it mean the estimates are wrong?

16 years

[gov2001-l] Shade regions

by vrios＠fas.harvard.edu

Hi all, Quick question: How can I ask R to shade a region in a graph? Best, Viridiana R?os 617-997-2471

16 years

[gov2001-l] Powerpoint for Presentations on THursday?

by jhainmueller＠gmail.com

Yes you can use PP IFF (so that's if and ONLY if) you send us your slides by Wednesday midnight so we can fit them into the slideshow with the abstracts. But remember that you only have three minutes! j. From: Jose Luis Romo Cruz [mailto:jose_luis_romo at ksg08.harvard.edu] Sent: Tuesday, April 15, 2008 8:00 PM To: jhainmueller at gmail.com; jmlarson at fas.harvard.edu Cc: viridianarios at gmail.com Subject: Abstract and Title Hi, Attached you will find our abstract and title of our article. We have some results that we would like to present to the group. Can we use Power Point? Thanks a lot Jose Luis Romo Cruz Harvard University | Kennedy School of Government Master in Public Policy 2008 (857) 233 7289 jose_luis_romo at ksg08.harvard.edu

16 years

[gov2001-l] Clarification

by jhainmueller＠gmail.com

Ok I think I understand now (I misremembered the number in the PS and so I thought you were referring to part A) so ignore my previous comment. For part C you should do exactly what you had in mind. Switch Import for each observation and then check if these N counterfactuals are inside the convex hull of the actually observed data. Hth, Jens From: Marcy McCullaugh [mailto:marcy52080 at gmail.com] Sent: Tuesday, April 15, 2008 12:47 PM To: jhainmueller at gmail.com Subject: Re: [gov2001-l] Clarification Hi Jens, Yes, I understood that for Part B and did so accordingly, but I am still confused as to how to set up Part C, which is what my e-mail was asking about. It says to create a factual dataset "as the world was observed," which I thought referred to just the original sanctions dataset, and a counterfactual dataset "where IMPORT is switched for all observations." I assumed this meant we're supposed to switch the 1's to 0's and the 0's to 1's, but I just wanted a clarification. If we are supposed to do a treatment vs. control set-up for the convex hull question a la the section notes and Part B, then the question was worded quite unclearly. Thank you, Marcy On Tue, Apr 15, 2008 at 9:31 AM, Jens Hainmueller <jhainmueller at gmail.com> wrote: The questions ask for the difference in the expected probabilities between a state of the world where import is 0 and import is 1, so you should set the X accordingly (it says nothing about imports varying by person). Hth, jens From: gov2001-l-bounces at lists.fas.harvard.edu [mailto:gov2001-l-bounces at lists.fas.harvard.edu] On Behalf Of Marcy McCullaugh Sent: Tuesday, April 15, 2008 12:26 PM To: gov2001-l at lists.fas.harvard.edu Subject: [gov2001-l] Clarification Hi, I am a little confused about the wording in Q1, Part C: For the counterfactual dataset, are we supposed to switch all import observations, where a 1 becomes 0 and a 0 becomes 1, or are we just supposed to switch all the 0's to 1's per #2 in Part B? Thanks, Marcy -- Marcy E. McCullaugh Ph.D. student Department of Political Science University of California, Berkeley 210 Barrows Hall Berkeley, CA 94720 Exchange Scholar 2007-08 Department of Government and Davis Center for Russian and Eurasian Studies Harvard University 1730 Cambridge Street, 3rd Floor Cambridge, MA 02138 _______________________________________________ gov2001-l mailing list gov2001-l at lists.fas.harvard.edu http://lists.fas.harvard.edu/mailman/listinfo/gov2001-l -- Marcy E. McCullaugh Ph.D. student Department of Political Science University of California, Berkeley 210 Barrows Hall Berkeley, CA 94720 Exchange Scholar 2007-08 Department of Government and Davis Center for Russian and Eurasian Studies Harvard University 1730 Cambridge Street, 3rd Floor Cambridge, MA 02138

16 years

[gov2001-l] bootstrap

by jbischof＠fas.harvard.edu

Hi all, I understand that a bootstrap (like we did in an early problem set) is one way to estimate the variability of a point estimate. However, I am not sure how one should interpreted bootstrapped standard errors: it seems like they should almost always be larger than the original (say, OLS) errors because we're always using a subset of the data and so less degrees of freedom with each simulation. How can one implement the bootstrap to show that the standard errors for an estimation technique used on the whole dataset underestimates the variability in the point estimates? Does the difference between the regular and bootstrapped standard errors just have to be huge? -- Jon Bischof Graduate Student Department of Government Harvard University

16 years

[gov2001-l] poisson log-likelihood

by jbischof＠fas.harvard.edu

Hey all, I'm having trouble with the log-likelihood for the poisson distribution. Gary's book says that the LL function is: lambda <- X%*%par out <- sum((lambda%*%t(y))-exp(lambda)) However, lambda*y is a nxn matrix and exp(lamda) is a nx1 matrix, so they cannot be subtracted. Furthermore, we want in the end a kx1 matrix of betas, and there is no k length dimension in either of these matrices. Clearly I have somehow misinterpreted the slides, but I'm not sure how. Any ideas for getting out the betas? -- Jon Bischof Graduate Student Department of Government Harvard University

16 years

2024

2023

2022

2021

2020

2019

2018

2017

2016

2015

2014

2013

2012

2011

2010

2009

2008

2007

2006

2005

2004

2003

Gov2001 April 2008