How-to: Quantify Uncertainty in Travel Forecasts

Federal Highway Administration2018PDFTech ReportChen, JasonBernardin, VinceTuttle, SteveTrevino, StevenUnited States. Federal Highway AdministrationRSG, Inc.Sun, SarahUnited States. Federal Highway AdministrationUnited StatesTraffic forecastingTraffic modelsUncertaintyMonte Carlo methodRisk analysisForecastModelUncertaintySensitivity analysisMultivariate response surfaceMonte Carlo simulationFHWA-HEP-20-021DTFH61-12-D-00013, T-12002Sun, SarahTravel 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 to derive those forecasts. This uncertainty translates into risks that decisions based on the forecasts will be misguided - either failing to meet objectives or simply not performing as well as some other alternatives that could have been selected. From an agency perspective these risks can lead to lawsuits, loss of credibility, unwarranted expenditures, or suboptimal allocation of funding. Formal risk analysis includes several dimensions including identifying sources of risk, quantifying their likelihoods of occurrence and estimating the consequences of those occurrences. In the case of travel forecasting, quantifying uncertainty in travel forecasts and its effects on key performance measures is a non-trivial task given the complexity of the factors which can impact travel. This report provides details on how uncertainty in travel forecasts and related performance measures can be quantified. Formal methods for quantifying risk or uncertainty profiles in key performance measures have been developed and are increasingly common in the context of investment grade traffic and toll revenue studies. This How-to guide will illustrate how these methods can be applied to other performance measures such as system VMT, delay, transit ridership, walk and bicycle mode shares, and greenhouse gases and other emissions. Three different risk analysis approaches are explained and illustrated using an activity-based model for Chattanooga, TN, and a four-step model for Toledo, OH: 1) traditional sensitivity analyses with simple risk profiling (similar to FTA guidance); 2) risk profiling based on univariate sensitivity analysis with Monte Carlo simulation and 3) more robust risk profiling using multivariate response surface methods and Monte Carlo simulation.119