A similar alternative might be to impute logged values and then exponentiate.
Mark Lunt
Ren� Wevers wrote:
> I am working with an extensive dataset (10.000+ observations) and
> logically some values are missing. I decided to use the -ice- package to
> impute certain missing values, but the result simply makes no sense to me.
>
> For instance I am estimating missing variables for the full time
> equivalent (FTE) of employees of a company mainly based on the absolute
> number of employees. (Logically) the absolute number of employees is never
> below zero. Also when I run a regression between the FTE number and
> absolute number I get a highly significant relation with positive
> coefficient and a positive constant estimates. Nevertheless when I run
> -ice-, the imputed values are extremely often (far) below zero (!!!). Also
> worrying is that the imputed values are practically all completely
> different (over factor 100) from the absolute number of employees, where a
> closer relation is (logically) expected.
>
> Is there some explanation for this or are we making any mistakes.
>
>
ICE assumes that continuous variables are normally distributed: if that
is not the case, impossible values can appear. In particular, if you
have lots of companies with a few employees and a few companies with
lots of employees, ICE will impute negative numbers of employees. One
possible solution is to use the "match" option of ICE. Alternatively, I
have written some ado-files which convert variables to normal-scores and
back: you can convert to normal scores (which are normally distributed),
perform the imputation on these variables, then convert back to your
original distribution. If you are interested in using these ado-files, type
net from http://personalpages.manchester.ac.uk/staff/mark.lunt
into stata, then click on the blue "nscores"
Hope that's of some use
*
* For searches and help try:
* http://www.stata.com/support/faqs/res/findit.html
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/
