Multilevel simultaneous equation model: a novel specification and estimation approach (Q2332699)

For linear regression problems having hierarchical or grouped data, linear statistical models are studied involving groups of linear equations such as: \(y_i = X_i \beta + Z_i \mu_i + \epsilon_i\), \(i=1, \dots, l\). Here, \(y_i\) denotes a response-vector for the \(i\)-th group, \(X_i, Z_i\) are design matrices, \(\beta\) is the unknown vector to be estimated, and \(\mu_i, \epsilon_i\), \(i=1, \dots, l\), are independent zero mean normal-distributed random vectors. More general models of this type are presented. Several covariance structures for the random variables are discussed, and maximum likelihood estimators for the unknown vector \(\beta\) of coefficients are considered. Numerical examples are given.