just imagine each month is a year and its all the same.
on your other point, the more missingness your data have, the more
model-dependent your results will be. you can improve things if you add
variables to predict the missingness. you can also add constraints. one
way is to transform data prior to Amelia (such as bounded variables to an
unbounded scale) and the untransform afterwards. if you use Amelia for
Gauss, then you can impose more complicated constraints, such as
the compositional data constraints you mention.
Best of luck with your research,
Gary King
: Gary King, King(a)Harvard.Edu
http://GKing.Harvard.Edu :
: Center for Basic Research Direct (617) 495-2027 :
: in the Social Sciences Assistant (617) 495-9271 :
: 34 Kirkland Street, Rm. 2 HU-MIT DC (617) 495-4734 :
: Harvard U, Cambridge, MA 02138 eFax (928) 832-7022 :
On Fri, 10 Jan 2003, Akunda, Eric wrote:
Dear Prof. King,
I have recently begun using Amelia to do data imputation.
I attempted to use the time series cross-sectional function in Amelia
for Windows but could not. The reason is my data are observed monthly,
and the example in the Amelia document states how to perform imputation
for yearly data. What should I do to include monthly observations (in
either the windows/gauss version)?
My second question regards rejection sampling. I have examined the
imputed data and find that for the variable with the highest level of
missingness (60% missing), the imputations are rather unreasonable. I
intend to use rejection sampling by specifying a range from publicly
available data sources (I can get annual data for the variables from
some published sources). Since the imputed values "should" sum up to the
reported annual values, is there a way I can take this into account
(e.g. use the mean for the 12 months)?
Thank you very much for taking the time to answer my questions.
Eric A. Akunda
Doctoral Student (Marketing)
Kenan-Flagler Business School
University of North Carolina - Chapel Hill
CB # 3490 McColl
Chapel Hill, NC 27599-3490
Tel: 919-962-0783
Voice: 919-918-1222
Fax: 919-962-7186
Email: eric_akunda(a)unc.edu <mailto:eric_akunda@unc.edu>