QUASI-LIKELIHOOD FOR MULTIPLICATIVE RANDOM EFFECTS

A quasi-likelihood approach similar to that described by Morton (1987) is found to be particularly simple in the case of measurements subject to two or more sources of multiplicative error. Still further simplification occurs when the effects of interest also are multiplicative. The multiplicative analogue of the balanced 'split plot' model is considered in detail: an analysis of gamma deviance replaces the familiar analysis of variance. The efficiency of the quasi-likelihood method, relative to a more conventional analysis of logarithms of the data, is explored briefly.