We discuss graphical models and parameter estimation for the problem of missing data presented by Frangakis and Rubin (2007).We use a graph model to describe the non-ignorable missing mechanism and the EM algorithm to find the maximum likelihood estimators (MLE).Then we relax their assumptions and extend the model.We show that under certain condition, the parameters of interest are also identifiable, and we can find the MLE using the EM algorithm.With weak assumptions, we can obtain the bounds of causal effect we are interested.Finally we show simulation results.