Quantifying Uncertainty in Travel Forecasts and Performance Measures
Corresponding Author: Jason Chen, RSG
Presented By: Jason Chen, RSG
Abstract
Travel forecasts inherently embody some level of uncertainty as a result of uncertainties around many of the key inputs, simplifications, and statistical methods that are used. Good planning is best supported by an understanding of the uncertainty in travel forecasts and the performance measures derived from them. However, quantifying uncertainty in travel forecasts and its effects on key performance measures can be a non-trivial task given the complexity of the factors which can impact travel.
To help support planners and modelers, TMIP recently developed a How-To manual illustrating how uncertainty in travel forecasts and related performance measures can be quantified. The proposed methods were applied using the Toledo trip-based model and Chattanooga activity-based model to evaluate the level of uncertainty in several planning performance measures such as VMT, VHT, transit ridership, etc., derived from their travel demand forecasts. The uncertainty factors were assumed to come from both land use and non-land use attributes that would significantly affect travel choices.
Traditional univariate sensitivity analyses were performed and compared with a more robust approach based on Monte Carlo simulation using a reduced form model estimated from a limited number of travel model runs selected based on experimental design criteria. The approach illustrates an experimental design with twenty model runs designed to establish the relationship between performance measures and uncertainty factors. A response surface analysis is used to establish the relationship between performance measures and influencing attributes in the form of a reduced form model. Monte Carlo methods are then used to numerically simulate the full probability distribution of performance measures.
While these methods may be new / unfamiliar to many practitioners, the goal of TMIP’s project is illustrate how they can be implemented in a practical way to quantify uncertainty in performance measures. The results can be important and helpful for planning, illustrating, for instance, how some measures like VMT show a high degree of uncertainty while other measures like travel time saving can show relatively limited uncertainty.