11 Apr
2011
11 Apr
'11
10:19 a.m.
So if the problem is that what you observe, when you observe something, is
income if income is less than (say) $100K, and $100K otherwise. And then
some of the income items are missing. So the key here is that the top
category is only partially observed since its censored at $100K. Setting
aside the problem of the fully missing observations, you need to impute this
censored category too, but under the constraint that its really > $100K.
You might be able to deal with that via cell priors. If you do that, then
it ought to help you with the fully missing observations too.
Gary
--
*Gary King* - Albert J. Weatherhead III University Professor - Director,
IQSS - Harvard University
GKing.Harvard.edu <http://gking.harvard.edu/> - King(a)Harvard.edu -
@kinggary<http://twitter.com/kinggary>- 617-500-7570 - Asst 495-9271 -
Fax 812-8581
2011/4/8 Elias Naumann <elias.naumann(a)uni-mannheim.de>
Dear all,
I'm currently using amelia to impute income in cross-sectional data (about
30% are missing). The distribution of the available income data is right
skewed, so I transformed the income using the natural logarithm.
Unfortunately the income data is also top-coded or right-censored - that is
all respondents with an income above a certain limit are summarized into one
category. (see example in the following paper, dealing exactly with this
problem: http://www.fcsm.gov/07papers/Buettner.X-C.pdf)
[image: Comparison of observed and imputed data]
(I've also added the comparison of imputed and observed data in a jpeg-file
where the imputation worked and I hope that the graph is visible)
So I have two problems:
1.) In some of the data sets, when there are a lot of people in the top
category the imputation is not succesfull at all.
2.) But even if the imputation is working, the imputed values in the upper
categories are smaller than the observed values - and only very few imputed
values are above the maximum of the observed values.
Is there any way to deal with these problems in Amelia? Or do I have to
transform the income data to get a good imputation?
Thanks for your help,
Elias
--
Elias Naumann
SFB 884
Universität Mannheim
