Hi Pablo,
I'm not sure I completely follow what is happening in your data. If
both datasets contain information on the same units, the I would merge
the two datasets by unit, impute family income in dollars, and then
recreate the imputed intervals based on the categories. That is, if
the intervals are deterministic functions of the incomes, then the
goal should be to just impute the dollar incomes.
If you observe the interval but not the dollar value, you could use
your procedure in (b) along with the "bounds" argument in amelia to
produce imputations of the dollar income that are consistent with the
intervals.
Cheers,
matt.
On Sun, Jan 16, 2011 at 10:55 PM, Pablo Mitnik <pmitnik(a)stanford.edu> wrote:
Dear all,
I need to do multiple imputation in a context that, I believe, is at least
somewhat atypical.
I have two datasets. In one dataset, family income is measured in broad
dollar intervals (e.g., $0 - $4,999, $5,000 - $9,999, and so forth). In the
other, family income is measured in dollars. Many demographic variables are
common to both datasets. I want to do MI of income measured in dollars in
the first dataset using the information contained in the second and the
interval measure of income included in the first. I have two questions:
(a) Is there any methodological reason NOT to do MI using information in one
dataset to impute in the other?
(b) In order to use the information contained in the interval measure of
income, I am considering doing MI by interval. That is, I would create as
many auxiliary datasets as income intervals there are in the first dataset.
Each of these auxiliary datasets would have all observations in that income
interval from both original datasets. The idea is to do MI in each auxiliary
dataset separately, and then put together MI'ed complete datasets by
combining in the obvious way the original observations (including the
imputed variable) from all auxiliary datasets, but discarding the
observations contributed by the second original dataset. My question is
whether this would be a legit procedure, and whether there may be a better
way of achieving my goal.
I would appreciate any help with these issues enormously.
All best,
Pablo Mitnik
--
Pablo A. Mitnik
Postdoctoral Scholar
Stanford University (
http://www.stanford.edu)
Center for the Study of Poverty and Inequality
(
http://stanford.edu/group/scspi-dev/index.html)
450 Serra Mall, Building 80
Room 110
Stanford, CA 94305-2029
650 724 3889
pmitnik(a)stanford.edu
-
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