Dear list & Matt,
After running a successful imputation in AmeliaView I wish to produce a
further few sets of (5) imputations and view the corresponding Diagnostics
for each set. I wish to do this without exiting & restarting AmeliaView.
If I do restart AmeliaView and run an (apparantly) identically specified set
of imputations on identical data I notice that for each set of imputations
the corresponding diagnostic plots vary slightly. I assume this is normal?
However after running a second and subsequent set of imputations the Output
log does not seem to change. Without restarting AmeliaView I change the
output directory before each fresh set of imputations; AmeliaView behaves
well, producing fresh sets of (5) imputations and saving them to the
allocated directory (csv files). However the log does not appear to change
and the diagnostic plots appear identical to those produced for the first
set of imputations. Am I missing something please?
many thanks
Simon UK
Dear members of the Amelia list,
I tried to conduct multiple overimputation in Amelia. More than one variable in the dataset should contain measurement error. I looked in the manual whether such a case is covered, but I have not found any example. I was also not successful in creating an appropriate "mo.list" object which combines the information obtained from several moPrep calls.
Could you provide an example of overimputation with more than one mismeasured variable?
Many thanks and kind regards,
Alexander
...........................................
Alexander Robitzsch
Bundesinstitut für Bildungsforschung, Innovation &
Entwicklung des österreichischen Schulwesens (BIFIE)
Department Bildungsstandards & Internationale Assessments (BISTA)
Alpenstraße 121, 5020 Salzburg
Österreich (Austria)
Tel. +43 (0)662 620088-5019 Fax +43 (0)662 620088-5900
https://sites.google.com/site/alexanderrobitzschhttp://www.bifie.at
...........................................
Hello,
I have a clustered longitudinal dataset consisting of four waves of data on
individuals who are nested in neighborhoods. I have missing data on
several time-varying variables at both the individual- and
neighborhood-levels. There are also some time-invariant individual
variables.
>From what I gather in the documentation, I should do the following things
to account for this data structure:
-make sure the data is structured in long format, with a time indicator
and 4 rows for each individual.
-identify time series variable with ts = ____ option
-identify individual ID with idvars = ____ option
-identify neighborhood ID with cs = ____ option
-use to intercs = TRUE option to allow variation time trends
Here are my questions:
1. Am I specifying the cross section variable properly? Do I really have
two cross section variables (individual ID and neighborhood ID) or does the
idvars and ts options account for the clustering within person, over time?
2. How do I tell Amelia that some variables are time varying and some are
time invariant? Is there a way to have polytime=0 for some variables to
force a constant time trend, but then have polytime=1 for others?
Thanks,
-WRJ
--
William Johnston
Doctoral Candidate
Harvard Graduate School of Education
Hello,
Is it possible to tell Amelia to not impute in situations where information
is missing by design? For example, if the youngest cohort of subjects has
missing data for a variable that the older cohorts are expected to have, is
it possible to have Amelia only impute for the older cohorts?
As an alternative I figure I can just impute the data that is missing by
design and then simply ignore it at the analysis stage, but I worry that
the imputation might be getting thrown off somehow.
Thanks,
WRJ
--
William R. Johnston
Postdoctoral Fellow | Harvard University
HiI have installed amelia 1.7.3 under R version 3.2.1 and trying to just run tcode with africa datalibrary(amelia)data(africa)
a.out <- amelia(x = africa, cs = "country", ts = "year", logs = "gdp_pc")
but receive the following error:-- Imputation 1 --
Error in outer(X, Y, SE, l) : argument "Y" is missing, with no default
Can anybody help me with this problem? Regards,
Email Address: saeedh_1999(a)yahoo.com
I have a question about combining imputed data from Amelia. I understand the rationale for running your end analysis on each imputed data set separately, and then combining the model results. However, what if your analysis is more complicated than a simple LM? For example, for my analysis, I am using imputed data sets (5) of time series variables (12 independent water quality variables, all time series, and 1 dependent time series). I am decomposing each series using loess and extracting the trend only. Then, I am using prewhitening and cross correlation to identify lags of variables that may be useful predictors. Finally, I am differencing each series and creating and comparing ARIMA models with external regressors to find the best model. I am having a hard time understanding how going through each of these steps with each imputed data set separately (and trying to combine the best models) is not going to create more variability and decrease the confidence of the model compared to averaging the imputed data sets before doing any analysis.
In short, if my imputed data sets are not "that" different, and the range of values for each of my predictors is relatively small, could it possibly be better to average the data first instead of trying to combine the best model from each?
I would greatly appreciate any comments or suggestions.
Thank you for your help.