The S-U algorithm for missing data problems (Q5943413): Difference between revisions

A new Monte-Carlo method for finding the solution of an estimating equation for missing data problems is proposed. Let \(Z_i\), \(i=1, 2, \ldots, N\), be random vectors. Suppose that \(Z_i\) is partitioned into components \(X_i\) and \(Y_i\), observations of \(Y_i\) are available, but \(X_i\) is unobserved or missing. Let \(S_i(Z_i |\theta) = S_i(X_i, Y_i |\theta)\) denote the \(i\)-th summand of an estimating function and let \(F_{\theta}(\{ X_i\} |\{ Y_i\})\) denote the cumulative distribution function of the missing data conditional on the observed data. The estimator \(\hat\theta\) is obtained from solving \(S_T(\{ Y_i\} |\hat\theta) =0\), where \[ (1)\qquad S_T(\{ Y_i\} |\hat\theta) = \sum_{i=1}^N S_i(Y_i |\theta) =\sum_{i=1}^N \int S_i(X_i, Y_i |\theta) dF_{\theta} (\{ X_i\} |\{ Y_i\}). \] The authors propose an S-U algorithm for finding the solution of the equation (1). The algorithm alternates between two steps: an S-step in which the missing data are simulated either from the conditional distribution described above or from a more convenient importance sampling distribution, and a U-step in which parameters are updated using a closed-form expression that does not require a numerical maximization. The authors prove that this algorithm converges and compare the S-U algorithm with Monte-Carlo maximum likelihood. They apply this algorithm to semi-Markov models with interval censored data.