摘要

The combination of road accident frequencies before and after a similar change at a giver. number of sites are considered. Each target site includes different accident types and is linked to a specific control area. At any one target site it is assumed that the total number of accidents recorded is multinomially distributed between the before period and the after period and also between several mutually exclusive types. The parameter of the distribution depends on the different accident risks in the control area linked to each site as well as on the average effect of the change. A method of estimating simultaneously the average effect and the accident risks in control areas is suggested. Some simulated accidents data allow us to study the existence and consistence of the linear constrained estimator of the unknown vector parameter.