Hi Scott,
It would help to know which error messages you find when using
AmeliaView, but without knowing that, a few observation.
First, you should be aware that the Amelia algorithm hinges on
assumptions about the missingness mechanism. If your data has
non-ignorable missingness (in the Rubin sense), then Amelia can be
problematic. Amelia is optimal when the missingness is predictable
given observed covariates, but does not depend on the missing values
themselves. This is called missing at random (MAR) by Rubin and
others. You can take a look at King, Honaker, Joseph and Scheve (2001)
for more details:
http://gking.harvard.edu/files/evil.pdf
For the organization of your data, note that Amelia is designed for
time-series cross-sectional data structures where there are quite a
few time periods for each unit. In this case the data is usually
arranged by unit-time. So your data would look like the following:
Unit Time Dep Neg
1 1 28 50
1 2 20 43
1 3 80 32
2 1 30 83
2 2 NA NA
2 3 NA NA
You could then set the cross-sectional variable to "Unit" and the time
variable to "Time."
Your data structure, though, should work with Amelia as well, as long
as you do not have that many variables. With either the TSCS format or
yours, with only 57 observations, you will quickly run up against a
boundary of 5-10 variables.
I hope that helps.
regards,
matt.
On Sat, Jan 3, 2009 at 1:44 AM, Scott T Gaynor <scott.gaynor(a)wmich.edu> wrote:
I could use some assistance in using AmeliaView. I
have data from a clinical trial where 57 participants were randomized to one of two
treatments. I am currently looking at data from measurements taken at three time points –
pretreatment, midtreatment, and posttreatment and would like to impute missing values from
those who dropped out or failed to complete assessments (i.e., non-random, non-ignorable
missing data). The input data format is SPSS. I have been getting a range of error
messages, but I think they stem from not organizing the data file correctly. Below is a
partial illustration of the current organization in SPSS (Dep=score on a depression
measure, Neg=score on a measure of negative thinking), which does not work:
Participant Treatment Dep-pre Dep-mid Dep-post Neg-pre Neg-mid Neg-post
1 1 28 20 15
80 65 50
2 2 25 23 20
85 80 80
3 1 30 25 80
70
4 2 20
70
.
.
.
Any thoughts on creating a data frame the corresponds to the elements in Amelia
View's _Variables Dialog_ and _Time Series Cross Sectional Dialog_ options would be
much appreciated.
Sincerely,
Scott
-
Amelia mailing list served by Harvard-MIT Data Center
[Un]Subscribe/View Archive:
http://lists.gking.harvard.edu/?info=amelia
More info about Amelia:
http://gking.harvard.edu/amelia
-
Amelia mailing list served by Harvard-MIT Data Center
[Un]Subscribe/View Archive:
http://lists.gking.harvard.edu/?info=amelia
More info about Amelia:
http://gking.harvard.edu/amelia