Hello folks,

I have got a conceptual question concerning multiple imputation with  
Amelia. I have got a TSCS data set with quite a few missing variables.

I have got a country X for which I observe the variables A, B and C  
during a range of t-10 to t. For my analysis, I need the following  
values for X: A and B at time t and C at t-1 (C being the lagged  
variable, e.g. GDP).

The question is now if I should run the imputation on the TSCS  
dataset and then extract the observations I need or if I should  
compile the dataset exactly the way I will need it in the end prior  
to the imputation?

Thank you for your advice,


Raffael



--------------------------
Raffael Himmelsbach, PhD candidate
Institut d'études internationales et politiques (IEPI)
Université de Lausanne