On Tue, 10 Oct 2006, Landa, Sandra wrote:
Dear Gary,
When I found Amelia-II a great happiness came into my life... since I am
fighting with my missing data a long, long time ago...
Unfortunately I am not an expert in statistics (Im really trying
hard...) and I have a lot of problems trying to discover why the results
of an inputation some times are not what I expect...
let me explain you:
I have about 60 variables and over 600 observations. My variables are
ALL ordinal (as they came from a Questionary ) and the answers to the
questions look like this:
the model has p*(p+3)/2 parmaeters with p variables, which means you'll
have 1890 variables if they are all continuous...
O : no problem
1: mild problem
2: moderate problem
3: severe problem
4: complet problem
9: not covered (I won't set it to missing !!)
(.) = missing...
but they're fully categorical, which means you'd need to code these as a
set of binary variables. with only 600 observations, you're asking Amelia
II to do something logically impossible. You could treat this variable as
continuous, but you'd have to do something about category 9.
I exported my dataset from SAS into csv, I have to
admit that I
couldn't read any other kind of source with Amelia, neather from sas
xport nor from spss...
After running Amelia (what I really enjoy due to its simplicity !!!) , I
should obtain a new value like a 0, a 1, a 2, a 3, a 4 or a 9 instead a
( . ) ... is that right ?.. But... I often receive values like
0,0222044604925031 or -0,0111022302462516..
its treating the variable as continuous and so making imputations on a
continuous scale. you could round to the nearest integer if necessary
for your analysis model, or just use it as is if possible.
My question is, Dear Gary : what am I doing wrong ???
Is there any kind of information which I should read ?
I will be very happy to hear from you very soon... and I am already
thankfull for your cooperation...
Warm regards and the very best for you !!!
thanks for the kind words; best of luck with your research.
Gary King
Sandra Landa
-
Amelia mailing list served by Harvard-MIT Data Center
[Un]Subscribe/View Archive:
http://lists.gking.harvard.edu/?info=amelia