AGGREGATE PREDICTION WITH DISAGGREGATE MODELS: BEHAVIOR OF THE AGGREGATION BIAS

Disaggregate travel demand models are being increasingly applied to predictions of aggregate demand. This is usually done by applying these models to zonal aggregated data, but is causes a bias (the aggregation bias) in the predictions obtained. The purpose of this paper is to study empirically the characteristics of the bias for a variety of conditions. The empirical analysis focuses on the bias in aggregate predictions of mode choice for work trips in Washington, D.C. Two main factors of bias are identified as the geographic aggregation level and the level of detail by which the distribution of explanatory variables is represented. Both the magnitude and the behavior of the aggregation bias are examined for a wide range of geogrpahic aggregation levels, for several approximate representations of the distribution of explanatory variables, and for two different transportation options. The simplest aggregate prediction method uses average zonal variable values in the disaggregate model. The results of this study indicate that, by applying this method, substantially biased predictions may result. Applying more accurate distribution representation reduces this bias significantly but does not ensure its complete elimination. A residual bias of significant magnitude still remained in many of the situations examined. The implication is that sophisticated methods for bias reduction should be developed in order to make aggregate predictions with disaggregate models a more reliable analysis tool. /Author/