Carola
Let me just give you my $0.02 on this.
The imputed values are OK. When you are imputing continuous
variables, you are imputing them with the assumption that the variable
is continuous normal. Bounded estimates are not continuous normal,
though I have seen them treated as such. This is probably what you
are doing too when you use these percents in an analysis.
What you need to remember is that these imputed values are nothing
else but abstractions without a meaning that will get you unbiased
estimates. So even if your values are out of range, you should not
worry about that if your statistical analysis treats them as
continuous variables.
Some software (SAS Proc MI for example) will reimpute if it receives a
value that is out of range. I have seen people who round to the
border value within the range. Amelia has some facilities for priors
where you can set the range of your value (but I never used it because
I do not agree with the practice). It is probably fine to do that in
most cases, but when you engage in such a practice, you decrease the
variance of the imputed values. This could lead to artificially
decreased standard errors and overconfident estimates.
If I were you, I would worry not about what the imputed values are,
but if it is OK to treat these percents as continuous, if your model
and results make sense and if you are meeting the assumptions of MAR
with your imputation model. If you can confidently answer Yes to all
these questions, your results will be unbiased even if some
imputations are out of bounds.
I hope this helps
L
On Jul 5, 2008, at 8:30 AM, Carola Herrera wrote:
Hello
Amelia software is running on my computer but I've noticed that
variables
that on my original dataset range from 0 to 100 (they are
%) in the imputed datasets take negative values (the
reported "min" in the summary stats is a negative
number). I wonder how to fix this.
Thanks (I'm new to Amelia and MI)
Best regards,
Carola
-
Amelia mailing list served by Harvard-MIT Data Center
[Un]Subscribe/View Archive:
http://lists.gking.harvard.edu/?
info=amelia
-
Amelia mailing list served by Harvard-MIT Data Center
[Un]Subscribe/View Archive:
http://lists.gking.harvard.edu/?info=amelia