Hi.
I am trying to figure out two issues on Amelia II:
1) How to determine the q-order polynomials: The software allows one to specify
between 0-3. Since I am mostly interested in cross-section shifts, setting it
to 0 is not an option. But how would I determine to which order I'd set them
to? Also: in the Honacker and King (2006) article on missing values LOESS or
spline basis functions are recommended. Does Amelia II offer insights about
these two techniques as well?
2) How to set the case prior efficiently: I have 135 developing countries in my
data and would like to use a case prior to identify existing
similarities/differences between countries. Theoretically it makes sense to
group the countries by region/sub-region in terms of their similarity. Is it
possible to identify countries' similarities by regions/sub-regions rather than
making pairwise assessments? Also: the above-mentioned Honacker and King (2006)
article talks about observational priors but does not discuss the
advantages/drawbacks of case priors. Could you help me out here?
I look forward to your response.
Thanks very much!
Simone Dietrich
Ph.D. Candidate
The Pennsylvania State University
Thank you for the response, Gary.
I've successfully downloaded AmeliaView and I read through the documentation and What to do About Missing Values in Time Series Cross-Section Data.
My data are not time series in the sense of repeated measures within a subject across time. My data structure is people (level 1, n = 800) nested within organizations (level 2, g =90) at one time (cross-sectional). So people nested in the same organization all have the same value for a particular variable (let's call it cluster).
It's not clear to me how to communicate this to Amelia. My guess is that I need to make this clear in Step 2 Options (box for Time Series Index and/or box for Cross-Sectional Index). Can you point me in the right direction?
Thank you.
nick myers
-----------------------------------------------------------------------------------------------------------
Nicholas D. Myers, Ph.D.
University of Miami
School of Education
Department of Educational and Psychological Studies
Research, Measurement and Evaluation program
Merrick Building 311E
P.O. Box 248065
Coral Gables, FL 33124-2040
Tel: 305.284.9803
Fax: 305.284.3003
E-mail: nmyers(a)miami.edu
http://www.education.miami.edu/facultyStaff/Faculty_Bio.asp?ID=143
-----------------------------------------------------------------------------------------------------------
-
Amelia mailing list served by Harvard-MIT Data Center
[Un]Subscribe/View Archive: http://lists.gking.harvard.edu/?info=amelia
Hello.
A colleague has suggested that Amelia can be used to address the missing data problem described below. I just subscribed to Amelia list – so please let me know if there is a protocol in to posting that I am unaware of.
I am hoping that folks on the list are willing to
1. Let me know if they believe Amelia can be used to address this problem
2. Point me in the right direction as to where I can read about Amelia’s approach to this problem
I write looking for recommended resources/references on imputing for missing values of ordinal data at Level 1 (taking clustering into account). Responses on or off list will be appreciated.
The relevant data are nested (two levels). There are no missing data at Level 2 (number of groups = 32). The missing data at Level 1 are from ordinal variables (i.e., 32 Likert scale variables with 5 ordered categories). Within said variables there is not much missingness (i.e., no more than 5%). A multidimensional measurement model that accounts for the ordinal nature of the data (probit) will be imposed across the imputed datasets (all 32 variables).
Thank you.
nick myers
-----------------------------------------------------------------------------------------------------------
Nicholas D. Myers, Ph.D.
University of Miami
School of Education
Department of Educational and Psychological Studies
Research, Measurement and Evaluation program
Merrick Building 311E
P.O. Box 248065
Coral Gables, FL 33124-2040
Tel: 305.284.9803
Fax: 305.284.3003
E-mail: nmyers(a)miami.edu
http://www.education.miami.edu/facultyStaff/Faculty_Bio.asp?ID=143
-----------------------------------------------------------------------------------------------------------
-
Amelia mailing list served by Harvard-MIT Data Center
[Un]Subscribe/View Archive: http://lists.gking.harvard.edu/?info=amelia
Hello all,
I just wanted to inform you that a new version of Amelia (1.1-30) has
been updated to both the Amelia website and CRAN. The added features
mostly affect the AmeliaView GUI. They include:
-a "seed" option in AmeliaView for running Amelia with a specified
seed for easier replication
-the ability to save the Amelia output to a .RData file from within
the AmeliaView GUI
-the ability to hold the Amelia output in the R Global environment
from within the AmeliaView GUI, meaning that when you close the GUI,
the Amelia output will be in your workspace.
The seed feature is key for anyone trying to exactly replicate an
imputation procedure. The third feature allows you fire up R, load the
GUI, run Amelia from the GUI, close the GUI and interact with the
output of Amelia from within R. This may be convenient for those who
are learning R or have large datasets and are looking for a easy way
to configure options.
To get the latest version of Amelia, first make sure you have the
latest version of R, then run the following command in R:
install.packages("Amelia")
or
install.packages("Amelia", repos="http://gking.harvard.edu")
Please let us know if you have any problems with this updated version.
Furthermore, please let us know if there are any features that you
would like to see in future versions of Amelia. Thanks.
regards,
matt blackwell.
-
Amelia mailing list served by Harvard-MIT Data Center
[Un]Subscribe/View Archive: http://lists.gking.harvard.edu/?info=amelia
Hello -
My understanding of these methods is very basic, but I was hoping
someone could offer some clarification on the specifics of the
imputation procedure used in AMELIA II.
I understand that in single regression imputation variances are often
underestimated because the variable with missing data is a perfect
linear function of the other variables used to impute it (correct?).
However, I gather that AMELIA corrects for this in three ways: 1)
First, by using an EM algorithm which adds a residual term to the
variance 2) by adding a random component to reflect uncertainty in the
missing data 3) using multiple imputed data sets to reflect this
uncertainty as well.
Is this correct or is this too basic an understanding? I would
appreciate any help that could be offered.
Thank you.
-
Amelia mailing list served by Harvard-MIT Data Center
[Un]Subscribe/View Archive: http://lists.gking.harvard.edu/?info=amelia