Dear all,

I have simulated non-normal data to check what Amelia ist doing when the assumption of multivariate normal distribution is not considerd. I generated a variable with two peaks as a mixture of two normal distributions.

In the density plot of Amelia 1.6.1 one can see that the overimputations are normal distributed anyway. That's what I expected because of the algorithm. In Amelia 1.7 the density of the overimputations are looking nearly exactly like the original variable with two peaks.

Can you explain to me what was changed in the algorithm?

To a better understanding I attached the density plots. The black lines are the simulated values, the red lines the values after overimputation.

Thanks a lot!

Sara