18 Aug
2004
18 Aug
'04
5:26 p.m.
Dear Jong Hee,
Do you have any covariates, other than votes and lagged votes? I note
that in your reshape command you have a "7" meaning that you are telling
GAUSS there are 7 variables. However you only list six. Reshape may be
creating a 7th variable that is completely missing (However, that is not
the way reshape generally works--if it runs out of observations to put in
the new matrix it just starts reading from the beginning again--see
guesses below).
The error message you mention from Amelia states that Amelia is concerned
that a variable in the imputation dataset is either completely missing
(and you wouldn't ever want to impute a variable that is completely
missing) or constant (thus not a "variable" and not something to be added
to a multivariate Normal model).
So, take a look at the data you are getting after the reshape command.
Insert a line like: print vote[1:10,.] and see if it looks right. Is
there a 7th variable that doesn't belong?
My bold leap of intuition, is that you've done something similar already,
and had to change Reshape to have seven columns to make the matrix "look
right". This might be because your "load" command was reading the endline
character in your original ".txt" file as an observation of a variable.
If this pure guess work is near the mark, then the easiest fix is just to
change your fifth line to: vote=vote[2:225,1:6] so as to trim that
column also.
let me know if this works, or trouble continues,
James Honaker
(tercer(a)ucla.edu)
On Wed, 18 Aug 2004, Jong Hee Park wrote:
Dear Amelia users,
I use Amelia for Gauss to input missing obs in multiparty votes.
I have problem in putting data in Gauss. Since I'm not a Gauss user and use
Gauss this time first, I may make dumb mistakes.
My data(vote) look like these;
-----------------------------------------------------------------------------
party1 party2 party3 party1(t-1) party2a(t-1) party2b(t-1) party3(t-1)
0.52 0.43 0.04 0.41 0.33 0.17 0.06
0.46 0.51 0.00(missing)0.53 0.43 0.03 0.00
----------------------------------------------------------------------------
Party1 is the reference party and party 2 and 3 have missing values.
So I use following command;
new;
library amelia;
load X[ ] = c:\data\newelecg.txt; # call data
vote = reshape(X, 225, 7); # make data as matrix
vote = vote[2:225,.]; # delete vnames
_AMempri=3; # use ridge prior
_AMkknp=3; # number of parties
_AMkkpfo= 1; # position of the reference party
_AMemt = 1; # use t-distributed ECME algorithm
dbuf = amelia(vote);
And I got the next error message.
----------------------------------------------
Invalid input, program ended due to input error:
Variable in dataset is either a constant or completely unobserved.
Delete this variable and rerun Amelia.
Math coprocessor exceptions:
Invalid Operation
Missing values may have been encountered
----------------------------------------------
Could somebody tell me what's wrong with my command and how can I fix it?
Thank you in advance.
Jong Hee Park
Department of Political Science
Washington University in St.Louis
Campus Box 1063
One Brookings Drive
St. Louis, MO 63130
Office Phone: (314) 935-6764