Overcoming the (apparent) problem of inconsistency in origin-destination matrix estimators

Recently a number of models for the estimation of origin-destination trip matrices from traffic counts have been presented. One class of models assumes knowledge of the proportional usage of each link by the interzonal traffic. These models result in an underspecified, and often inconsistent, system of linear equations. Several authors have addressed the problem of inconsistency by, in different ways, changing the traffic count data so as to achieve consistency. The authors argue that, modeling-wise, this is not a sound approach because inconsistent data is a natural part of any origin-destination matrix estimation problem. By using a stochastic programming approach the inconsistent input data becomes a natural part of the estimation process. This is done by viewing the traffic counts as realizations of some unknown underlying distribution of traffic flows.