Amelia March 2007

amelia@lists.gking.harvard.edu

14 participants
16 discussions

by Marco Antonio Morales-Barba

Dear all, I keep getting the following message whenever I add case priors to Amelia: 1 Error in solve.Lambda + solve.Sigma : non-conformable arrays I get the same message regardless of whether they are defined as ranges (5 col matrix) or distributions (4 col matrix). The imputation works fine without adding case priors. I'm using R 2.4.1, and Amelia II version 1.1-23. Any guidance would be appreciated. Best, Marco - Amelia mailing list served by Harvard-MIT Data Center [Un]Subscribe/View Archive: http://lists.gking.harvard.edu/?info=amelia

17 years

Acceptable imputation process with small case numbers

by Dale P Rodmell

Hi all, I'm new to Amelia and generally a statistics rookie and have been attempting to impute a dataset using the Amelia II GUI with a small number of cases (25) and a large number of variables (104). 25 of the variables have missing data of between 1 and 4 values and these are distributed fairly randomly across cases. Variables are mainly ordinal but with a few continuous and nominal. The only way I can get the programme to impute is by increasing the ridge/empirical prior to around 190. I've also used range priors for the majority of variables with maximum certainty. Is this an acceptable approach? An alternative would be to drastically cut the number of variables but even then it still requires a high empirical prior, much higher than the 5% of variables suggested as a moderate ridge prior level in the manual. It may also mean generating a number of separately imputed datasets containing a small number of variables for separate multivariate analyses. Is there any obvious answer about which approach is most acceptable, if any? I assume that either is better than an ad-hoc approach or forgetting multivariate analysis with the data. I'm hoping to employ the non-parametric multivariate approaches such as ANOSIM in the statistics package PRIMER with the imputed data set. I'd welcome any advice that can be offered. Regards, Dale Rodmell Scarborough Centre for Coastal Studies University of Hull ***************************************************************************************** To view the terms under which this email is distributed, please go to http://www.hull.ac.uk/legal/email_disclaimer.html *****************************************************************************************

17 years, 1 month

Re: [amelia] examples of work using Amelia

by Tim M�llenborn

Hallo, I think you can use the "save session" option for dokumentation purposes in the GUI. But in my version the "load" option in the gui does not work well. But you can paste the code which you have produced with the GUI to R (like in SPSS) To justify the imputation, i think you have to argue, that your estimates would be biased in a systematic way when you don't impute the data. Tim - Amelia mailing list served by Harvard-MIT Data Center [Un]Subscribe/View Archive: http://lists.gking.harvard.edu/?info=amelia

17 years, 1 month

examples of work using Amelia

by Paul Manna

Hello, I have a question about exposition and was seeking some examples. I'm using Amelia (and then Clarify for analysis) in a paper and am trying to decide how much description of the imputation procedure I should include in my discussion of methods. In the interest of transparency and replication, I always try to err on the side of giving more explanation of what I've done. But as we all know, page space can be tight; also, I don't want readers to get so hung up on the imputation aspect that they miss out on the substantive implications of the paper. Have you seen good examples of published work (or conference papers?) that used Ameila and did an admirable job of justifying and explaining its use? I wondered if people tended to integrate these things into the text, put them in footnotes or appendices, or had some other approach. I was hoping to read some examples to help me develop my own discussion of these things. Thanks, Paul Paul Manna Assistant Professor Department of Government Thomas Jefferson Program in Public Policy College of William and Mary http://pmanna.people.wm.edu/ tel: 757-221-3024 / fax: 757-221-1868 - Amelia mailing list served by Harvard-MIT Data Center [Un]Subscribe/View Archive: http://lists.gking.harvard.edu/?info=amelia

17 years, 1 month

deciphering error message

by tdavis7＠emory.edu

Hello all - I am trying to impute data for a particular data set. The variable of interest with missing data is an index that I compute using approximately 16 single indicators. I have successfully used Amelia to impute missing data for the calculated index in a separate data set excluding the single indicators. However, when I try to impute using a larger data set with each of the single indicators that comprise the index (to see if there are any differences between imputing the single indicators versus the full index), I get the following error message in Amelia: Error in newx[,j-1] <-ifelse (x[,i] = value[j], 1, 0): subscript out of bounds Can anyone help decipher this? I would appreciate any help. Thank you. - Amelia mailing list served by Harvard-MIT Data Center [Un]Subscribe/View Archive: http://lists.gking.harvard.edu/?info=amelia

17 years, 1 month

Another multilevel Q

by Levi Littvay (UNL)

Hello all. I tried this before with no luck. Maybe now... If I have a multilevel model and I let many of my level 1 coefficients vary across my groups (not just time, other predictors as well), how appropriate is it to use Amelia to impute level 1 missing. As far as I can tell in Amelia I can only let the intercept and the slopes of time vary across my groups. If I want to let other slopes vary in the analysis, don't I need an imputation technique that does the same. Any guidance is appreciated. Thanks L - Amelia mailing list served by Harvard-MIT Data Center [Un]Subscribe/View Archive: http://lists.gking.harvard.edu/?info=amelia

17 years, 1 month

simple Q about prior and tolerance

by gaoj

Dear all, i am new to Amelia. i have a question that may be really simple for you (but not for me). could anyone tell me anything about empirical prior? it does not make much sense to me. say if i have 50k cases (among them 3000 are missing). should i set the empirical prior to 50k*5%= 2500? or (50k-3k)*5% , what's the rationale here for it to shrink the covariance? must it be 5% ? many thanks gao

17 years, 1 month

Amelia and HLM

by tdavis7＠emory.edu

Hi - I have a data set that has missing data for 40% of cases on 1 computed variable (an index created from about 12 single variables). I would like to use HLM to analyze the data, since the data is hierarchical in structure. The variable of interest is at the second level of a 2-level HLM structure. The problem is that STATA assumes a common level-2 file. Therefore, if I impute data using Amelia, I must have a single level-2 file in order to run the analysis in HLM. Does anyone know if I can use Amelia and then some other program like Clarify to combine the imputed data sets before I run the HLM analyses? I would appreciate any help anyone could offer. Thanks. - Amelia mailing list served by Harvard-MIT Data Center [Un]Subscribe/View Archive: http://lists.gking.harvard.edu/?info=amelia

17 years, 1 month

Re: [amelia] Adding a range prior using AmeliaView

by Israel.Marques.II＠Dartmouth.EDU

Prof. Blackwell, Thank you for your assistance with saving priors. I've subsequently attempted to run Aemliaview and have recieved an error similar to Prof. Saunder's (the difference is which order's minor is not positive. Mine are generally 4 and 6). I've tried constructing my prior matrix in various different ways with different variables, but always end up with the same result. Not even my attempts at using a single prior at a time worked. I've attempted to run the program several times on the same data set using the same prior, and it appears to occur at different points during the first round of imputations. Generally, it seems to occur as the output hits 600 or so, although it has occurred as late as 2234. My data is a bit too varied and complex to be easily described, so I can e-mail a copy of my dataset if that would be useful. As always your help and dedication are greatly appreciated, -IM ---quote--- Matthew, Thank you for updating the software so quickly. I emailed before that I had the same problem as Heather. I just ran my imputations with range priors set and received the following error in the output window after the 4th imputation (of 5): There was an unexpected error in the execution of Amelia. Double check all inputs for errors and take note of the error message: Error in chol(copy.theta[c(FALSE, m[ss, ]), c(FALSE, m[ss, ])]) : the leading minor of order 28 is not positive definite This is the same error I received before except the order number given is not always the same (I tried several different variations). Repeating the imputations with a ridge prior results in the same error following the iterations of imputation 2: There was an unexpected error in the execution of Amelia. Double check all inputs for errors and take note of the error message: Error in chol(copy.theta[c(FALSE, m[ss, ]), c(FALSE, m[ss, ])]) : the leading minor of order 22 is not positive definite The variables for which I set range priors are continuous and ordinal, though I am considering the ordinal variables to be continuous. One ranges from -2.5 to 2.5 and only has 15 of 208 cases missing, and the others from 0 to 2 with a bit more missing cases (about 60 each). Let me know if you need more details. Thank you, Erin Saunders Portland State University On 3/12/07, Matthew Blackwell <blackwel@xxxxxxxxxxxxxxx> wrote: Hello Heather, Sorry for the late response, but I wanted to let you know that we've fixed the error in Amelia you described where the mean of the prior would be set to the maximum in the range. You can update your copy of Amelia by running the following line of code in R: install.packages("Amelia", repos="http://gking.harvard.edu";) Also, I wonder if you could see if this solves your other error message about the leading minor not being positive definite. If it doesn't, please send us another message with a bit more detail about your data and the conditions of the error message. thanks for helping us out, matt. On Mon, 5 Mar 2007, Heather Stoll wrote: > Dear all, > > I'm running version 1.1-20 of Amelia on a Windows system. In trying to help > a student add a range prior for an entire variable using AmeliaView, I > clicked on "Set observation priors"; then "add range prior"; and in the "add > prior" dialogue box that pops up, selected as "case" the whole variable, > selected the appropriate variable, set the minimum and maximum, entered an > appropriate confidence coefficient, and then clicked "OK". In the > "Observational Priors" window, what's now listed is a whole variable prior > with a mean equal to the maximum value that I fed to the range prior. This > doesn't seem right. I'm wondering either if I'm missing something or if this > is just a display issue. I haven't yet tried to multiply impute this data > set in R itself (and hence to build the desired prior matrix). > > Any thoughts would be much appreciated. > > Heather Stoll > - > Amelia mailing list served by Harvard-MIT Data Center > [Un]Subscribe/View Archive: http://lists.gking.harvard.edu/?info=amelia - Amelia mailing list served by Harvard-MIT Data Center [Un]Subscribe/View Archive: http://lists.gking.harvard.edu/?info=amelia ---quote--- - Amelia mailing list served by Harvard-MIT Data Center [Un]Subscribe/View Archive: http://lists.gking.harvard.edu/?info=amelia

17 years, 1 month

Adding a range prior using AmeliaView

by Heather Stoll

Dear all, I'm running version 1.1-20 of Amelia on a Windows system. In trying to help a student add a range prior for an entire variable using AmeliaView, I clicked on "Set observation priors"; then "add range prior"; and in the "add prior" dialogue box that pops up, selected as "case" the whole variable, selected the appropriate variable, set the minimum and maximum, entered an appropriate confidence coefficient, and then clicked "OK". In the "Observational Priors" window, what's now listed is a whole variable prior with a mean equal to the maximum value that I fed to the range prior. This doesn't seem right. I'm wondering either if I'm missing something or if this is just a display issue. I haven't yet tried to multiply impute this data set in R itself (and hence to build the desired prior matrix). Any thoughts would be much appreciated. Heather Stoll - Amelia mailing list served by Harvard-MIT Data Center [Un]Subscribe/View Archive: http://lists.gking.harvard.edu/?info=amelia

17 years, 1 month

2024

2023

2022

2021

2020

2019

2018

2017

2016

2015

2014

2013

2012

2011

2010

2009

2008

2007

2006

2005

2004

2003

2002

Amelia March 2007