Estimating correlation with missing data

There's a great article here on estimating correlations and means when you have missing data in your datasets. It uses the Expectation Maximisation algorithm to calculate the missing values, and what is interesting is how much the implied correlation changes after applying EM.