Consider parametric models that are too complicated to allow calculation of a likelihood but from which observations can be simulated. We examine parameter estimators that are linear functions of a possibly large set of candidate features. A combination of simulations based on a fractional design and sets of discriminant analyses is then used to find an optimal estimator of the vector parameter and its covariance matrix. The procedure is an alternative to the approximate Bayesian computation scheme. © 2012 Biometrika Trust.

Original publication

DOI

10.1093/biomet/ass030

Type

Journal article

Journal

Biometrika

Publication Date

01/09/2012

Volume

99

Pages

741 - 747