Zelig February 2008

zelig@lists.gking.harvard.edu

11 participants
16 discussions

predicted probabilities with zelig and imputed data

by Amber Wichowsky

Hello,I have estimated a logistic regression model with Zelig using the mi command to combine multiple imputed datasets. I understand that the summary output is the combined result and that I can also print out individual results by dataset. However, in calculating predicted probabilities, I am wondering which coefficients are being used by Zelig - would these be from the combined results? In addition, when setting the independent variables to their means (setx command) - would this be some weighted mean of that variable across those five datasets? Finally, could someone clarify the difference between the expected values and the predicted values that are given in the output. Thank you, Amber Wichowsky University of Wisconsin-Madison

16 years, 1 month

Zelig_3.1-1 released

by Ferdinand Alimadhi

Hi everyone, The first bugfix release of Zelig version 3.1 (Zelig_3.1-1) is online. What's new? - Improved the code and documentation for *.mixed models(thanks to Gregor Gorjanc) - Fixed sur, twosls, threesls models which had problems due to some api changes in the last version of "systemfit" (see http://www.systemfit.org/) - Added some other models (including *.mixed models) in plot.ci - etc. To update/install this release: - *nix / windows users: install.packages("Zelig", repos="http://gking.harvard.edu" <http://gking.harvard.edu%22>) - mac users: install.packages("Zelig", repos="http://gking.harvard.edu" <http://gking.harvard.edu%22>, type="source") Please let us know if you run into problems. thanks, Ferdi

16 years, 2 months

help on ARIMA

by GSB E-Mail

All, I am new to Zelig. I am trying to build a ARIMA(2,0,2) model for my analysis. The command I used is: ngccret_arima <- zelig(Diff(ngccret, 0)~lag.y(2)+lag.eps(2), model="arima", data=ngccret) I got the error message: Error in eval(parse(text = unlist(strsplit(deparse(mf[[2]][[3]], width.cutoff = 500), : numeric 'envir' arg not of length one Here ngccret is a vector of 2010X1. I searched the mailing-list and did not find anything related postings. Your help will be greatly appreciated. Thank you. -- John Li

16 years, 2 months

Re: [zelig] problem with "sim" when using the "by" command

by Kosuke Imai

Hi Casey, It looks like that the current infrastructure of Zelig does not allow the simultaneous use of "by" command with multiple imputation... We hope to figure that out in the future though. Sorry! Kosuke On Wed, 13 Feb 2008, Casey Klofstad wrote: > Ok, here is the code and what happens: > > z.out <- zelig(civtotw2 ~ treat + civtotw1 + dcode1 + dcode3 + dcode4 > + dcode5 + dcode6 + dcode7 + dcode8 + dcode9 + dcode10 + dcode11 + > dcode12 + dcode13 + dcode14 + dcode15 + dcode16, weights="weights", > data = mi(ds1, ds2, ds3, ds4, ds5), model = "ls", by = "civw1_q") > > #this ran successfully; note that the "civw1_q" factor is an ordinal > four-point scale > > x.untreat <- setx(z.out, treat = 0) > > #this also ran successfully > > untreat.out <- sim(z.out, x = x.untreat) > > #this failed; here is the traceback: > > 9: stop("incompatible arguments") > 8: mvrnorm(num, mu = coef(object), Sigma = vcov(object)) > 7: param.lm(object[[i]], num = numM, bootstrap = bootstrap) > 6: param(object[[i]], num = numM, bootstrap = bootstrap) > 5: MIsimulation(object, num, prev, bootstrap, bootfn = bootfn, x = x, > x1 = x1, ...) > 4: sim.setx.MI(object[[i]], x = x[[i]], x1 = x1[[i]], num = numN, > bootstrap = bootstrap, ...) > 3: sim(object[[i]], x = x[[i]], x1 = x1[[i]], num = numN, bootstrap = > bootstrap, ...) > 2: sim.setx.strata(z.out, x = x.untreat) > 1: sim(z.out, x = x.untreat) > > I get the same problem even if I don't weight the data. > > Another complication (or bit of evidence) is after I ran the above, I > ran a piece of code that should work. It is the same model, onyl > without the "by" command. The model runs fine, but when I get to > "setx" I get this: > > Error in value[[jj]][ri] <- if (is.factor(xij)) as.vector(xij) else > xij : nothing to replace with >> traceback() > 4: rbind(deparse.level, ...) > 3: rbind(dta, tmp) > 2: setx.MI(z.out, treat = 0) > 1: setx(z.out, treat = 0) > > If I close R and rerun the above, it works. > > Thoughts? Thanks! > > -c > > On Feb 13, 2008 2:03 PM, Olivia Lau <olau(a)fas.harvard.edu> wrote: >> Casey, after the sim command (when you get the error), >> >> can you tell me the output you get when you do: >> >> traceback() >> >> ? >> >> Also, try it without weights. Thx, O >> >> >> >> On Feb 13, 2008 11:05 AM, Casey Klofstad <klofstad(a)gmail.com> wrote: >> >>> Yeah, the demo works for me too. >>> >>> Here is what I am trying to run: >>> >>> #START CODE >>> >>> z.out <- zelig(civtotw2 ~ treat + civtotw1 + dcode1 + dcode3 + dcode4 >>> + dcode5 + dcode6 + dcode7 + dcode8 + dcode9 + dcode10 + dcode11 + >>> dcode12 + dcode13 + dcode14 + dcode15 + dcode16, weights="weights", >>> data = mi(ds1, ds2, ds3, ds4, ds5), model = "ls", by = "civw1_q") >>> >>> x.untreat <- setx(z.out, treat = 0) >>> >>> x.treat <- setx(z.out, treat = 1) >>> >>> #note: the code appears to work fine until I get to the "sim" stage; >>> that is where I get the error "Error in mvrnorm(num, mu = >>> >>> coef(object), #Sigma = vcov(object)) : incompatible arguments" >>> >>> untreat.out <- sim(z.out, x = x.untreat) >>> >>> treat.out <- sim(z.out, x = x.treat) >>> >>> FD.out <-sim(z.out, x = x.untreat, x1 = x.treat) >>> >>> #END CODE >>> >>> I'm sure I'm missing something obvious here, so thanks very much for >>> your indulgence. >>> >>> Best, >>> >>> -c >>> >>> >>> >>> >>> On Feb 13, 2008 10:56 AM, Kosuke Imai <kimai(a)princeton.edu> wrote: >>>> demo(strata) works for us and so maybe you can send us the code so that >> we >>>> can see what the problem is? >>>> >>>> Thanks, >>>> Kosuke >>>> >>>> >>>> On Wed, 13 Feb 2008, Casey Klofstad wrote: >>>> >>>>> I'm using simple OLS. Also, I just updated from 3.0-6 to 3.1-0. That >>>>> does not appear to fix the issue. >>>>> >>>>> Is it just the case that I can't run "sim" on a series of models that >>>>> were generated with the "by" command? Or am I missing something here? >>>>> >>>>> Thanks much, >>>>> >>>>> -c >>>>> >>>>> On Feb 12, 2008 11:07 PM, Kosuke Imai <kimai(a)princeton.edu> wrote: >>>>>> Which model are you using? Also, did you try the latest release? >>>>>> >>>>>> Kosuke >>>>>> >>>>>> >>>>>> On Tue, 12 Feb 2008, Casey Klofstad wrote: >>>>>> >>>>>>> I am able to estimate my model using the "by" command in zelig to >> run >>>>>>> a model n times for n different values of a variable of interests. I >>>>>>> am also able to then run "setx" without any errors. >>>>>>> >>>>>>> The problem comes when I try to run "sim." I get the error: "Error >> in >>>>>>> mvrnorm(num, mu = coef(object), Sigma = vcov(object)) : incompatible >>>>>>> arguments." >>>>>>> >>>>>>> Maybe I am missing something obvious here. Thoughts? >>>>>>> >>>>>>> Thanks much! >>>>>>> >>>>>>> -c >>>>>>> >>>>>>> >>>>>> >>>>> >>>>> >>>>> >>>>> >>>> >>> >>> >>> >>> >>> >>> >>> -- >>> Casey A. Klofstad >>> University of Miami >>> Department of Political Science >>> Coral Gables, FL >>> >>> klofstad(a)gmail.com >>> http://moya.bus.miami.edu/~cklofstad >>> >> >> > > > > - Zelig Mailing List, served by Harvard-MIT Data Center Send messages: zelig(a)lists.gking.harvard.edu [un]subscribe Options: http://lists.gking.harvard.edu/?info=zelig Zelig program information: http://gking.harvard.edu/zelig/

16 years, 2 months

Offset variable in "poisson.gam"

by 宋时歌

I have a quick question: how to specify offset (exposure) variable in a " poisson.gam model"? Thanks. Shige

16 years, 2 months

Bayesian Model Averaging as an Approach to Estimating the Treatment Effect

by Andrew Stokes

Hi, Does anyone have any thoughts on Bayesian Model Averaging as an approach to estimating a summary treatment effect? Or a different approach? I would like to incorporate treatment effects obtained through models from the survival and logit classes and from a variety of specifications within those classes (both from before and after matching). Thanks. Sincerely, Andy Stokes Institute for Health Metrics and Evaluation - Zelig Mailing List, served by Harvard-MIT Data Center Send messages: zelig(a)lists.gking.harvard.edu [un]subscribe Options: http://lists.gking.harvard.edu/?info=zelig Zelig program information: http://gking.harvard.edu/zelig/

16 years, 2 months

problem with "sim" when using the "by" command

by Casey Klofstad

I am able to estimate my model using the "by" command in zelig to run a model n times for n different values of a variable of interests. I am also able to then run "setx" without any errors. The problem comes when I try to run "sim." I get the error: "Error in mvrnorm(num, mu = coef(object), Sigma = vcov(object)) : incompatible arguments." Maybe I am missing something obvious here. Thoughts? Thanks much! -c -- Casey A. Klofstad University of Miami Department of Political Science Coral Gables, FL klofstad(a)gmail.com http://moya.bus.miami.edu/~cklofstad - Zelig Mailing List, served by Harvard-MIT Data Center Send messages: zelig(a)lists.gking.harvard.edu [un]subscribe Options: http://lists.gking.harvard.edu/?info=zelig Zelig program information: http://gking.harvard.edu/zelig/

16 years, 2 months

Hazard Ratios, RR, and the exponential survival model in Zelig

by Andrew Stokes

Hi there, My ultimate goal is to produce a plot of effect sizes associated with the treatment across a range of models. I am translating all results into relative risks with 95% confidence intervals. When you run a duration dependent model using the exponential survival distribution and obtain hazard ratios, have you thus obtained the quantity of interest? Using Zelig in the Logit framework, for example, I set values of X and ran simulation to obtain relative risks. But in the survival framework, it appears that simply through regression, you obtain the analogous quantity. Is there any benefit in using Zelig to run simulations in this case? Best Regards, Andy Stokes Institute for Health Metrics and Evaluation - Zelig Mailing List, served by Harvard-MIT Data Center Send messages: zelig(a)lists.gking.harvard.edu [un]subscribe Options: http://lists.gking.harvard.edu/?info=zelig Zelig program information: http://gking.harvard.edu/zelig/

16 years, 2 months

None

by E. Michael Foster

f6f- " <shigesong(a)gmail.com> CC: zelig(a)lists.gking.harvard.edu Subject: Re: [zelig] After nonparametric preprocessing... References: <5abc11d80802010453m714cbcf9u3e5dd1a79ef95e7(a)mail.gmail.com> In-Reply-To: <5abc11d80802010453m714cbcf9u3e5dd1a79ef95e7(a)mail.gmail.com> X-Spam-HMDC: YES Sender: owner-zelig_at_lists_gking_harvard_edu(a)mail.hmdc.harvard.edu Precedence: bulk List-Unsubscribe: <mailto:majordomo@mail.hmdc.harvard.edu?body=unsubscribe%20zelig_at_lists_gking_harvard_edu> X-MajorDomo: zelig_at_lists_gking_harvard_edu X-Mime-Autoconverted: from 8bit to 7bit by courier 0.55 good morning, if what you're modeling varies across siblings, there are better ways to use information on siblings. whatis the "treatment"/ thanks, michael e.f6f- wrote: > Dear Colleagues, > > I am studying women's fertility behaviours in China. In my data, each > women can have up to nine births. A multilevel model with children > nested within women seems to be an ideal solution. In order to reduce > model dependence, I preprocessed the data using nearest neighbour > method. Sine the number of the "treated" case is relatively small > (10%), even with the ratio set to 2, the sample size of the matched > data reduced drastically and damaged the multilevel structure. Now > most of women in the matched sample has only one child left. My > question is: does it still make sense to estimate a multilevel or > mixed model instead of a plain OLS regression or logistic regression? > Further, are there good references on propensity score matching in a > multilevel situation? > > Best, > Shige -- E. Michael Foster getmike.info download vcard: www dot unc dot edu/~emfoster/Foster.vcf Professor of Maternal and Child Health of Health Policy and Administration of Biostatistics (adjunct) Associate Chair for Faculty Advancement (W) 919-966-3773 (F) 919-966-0458 Office: 407C Rosenau Hall << use for FEDEX School of Public Health University of North Carolina, Chapel Hill Rosenau Hall, Campus Box# 7445 Chapel Hill, NC 27599-7445 [the http is throwing off some spam filters) visit the new FAQ section of my blog: fosterfaq dot blogspot dot com 27) [Mollie Ivins] "What do you mean cancer is a gift. You're as a dumb as Shrub. Cancer just killed me. Some gift." "I'm not quite dead yet, sir!" -- Concorde, Monty Python and the Holy Grail 29) Feeling middle-aged? I've got the answer. 8-) - Zelig Mailing List, served by Harvard-MIT Data Center Send messages: zelig(a)lists.gking.harvard.edu [un]subscribe Options: http://lists.gking.harvard.edu/?info=zelig Zelig program information: http://gking.harvard.edu/zelig/

16 years, 2 months

Compare Predicted value from sim() vs training and tester set

by Aungsutorn Homtientong

To Whom this may Concern Hi there, I have tried using the multinomial logistic regression (mlogit) to estimate the model in zelig from the imputed data m=5 from Amelia. I actually have to compare the predicted value from the mlogit model vs the training set from the data i have.(Im using the data from data mining cup 2004 for my bacholor project). I know that sim() would generate the predicted value by inputting the parameter of x.out and z.out. Then can i actually compare the predicted value from sim() directly to the training set i have to check the accuracy? since the parameter did not require any data sets, or did it acquire the data from either x.out or z.out? then Later on i have to use the model from the zelig to apply to the tester set, while predict() that can input the model and the data won't work on zelig. how exactly can i use the model from zelig to apply to the tester set here and find accuracy? Many thanks Thank You A. Homtientong

16 years, 3 months

2024

2023

2022

2021

2020

2019

2018

2017

2016

2015

2014

2013

2012

2011

2010

2009

2008

2007

2006

2005

2004

2003

Zelig February 2008