Model validation is an important step in travel demand modeling. It serves to ensure that the calibrated model produces outputs that are consistent with the observed data. In the Florida Standard Urban Transportation Model Structure (FSUTMS), various consistency checks for the validation of the distribution and assignment models are suggested for highway networks.
These checks involve the comparison of predicted link volumes with ground counts at screenlines, cordon lines, and/or cutlines. Statistics such as system-wide vehicle-mile-traveled (VMT), vehicle-hour-traveled (VHT), link volume by facility type, area type, facility size (in term of number of lanes), and volume group Root Mean Square Error (RMSE) are major criteria used for validation. Errors between model outputs and what are observed from the actual networks may come from such sources as measurement, sampling, model misspecification, prediction, and aggregation process. In this study, sampling and model misspecification errors were considered in deriving accuracy standards for highway networks. The required number of count stations to be sampled for each type of facility to achieve a specific accuracy standard at a certain level of confidence was also derived and a general rule of thumb was recommended for validation purposes. For transit networks, the accuracy standards were derived to account for statistical variability, aggregation errors, and prediction errors. Errors in modal share subjected to aggregation and prediction variability were derived for both multinomial logit and nested logit models. However, no results are presented due to a lack of data support.