|
Abstract:
|
Statistical inference of longitudinal data based on non-smooth estimating equations can be accomplished by constructing large sample confidence intervals based on the asymptotic covariance matrix. However, this matrix often involves an unknown density function, which can make direct estimation difficult and unreliable. We introduce Markov chain marginal bootstrap (MCMB) (He and Hu 2002) as an alternative to construct confidence intervals. When there are p parameters, instead of solving a system of p-dimensional nonlinear equations at each step for classical bootstrap method, MCMB sequentially solves p one-dimensional equations, thus greatly reducing the computational difficulty, especially when p is large. Our simulation study shows favorable performance of the MCMB method in accuracy and reliability compared to the direct method of estimating the asymptotic covariance matrix.
|
