Accurate forecasts of benefits and costs from planned transportation investments are required for good decision making. Recent research from Flyvbjerg, Bain and others has noted large inaccuracies in benefits and costs from large transportation projects. One cause of these inaccuracies is known in behavioral research as the “planning fallacy”, a systematic underestimation of costs and risks and overestimation of benefits made when decisions are made under uncertainty (Kahneman). To de-bias forecasts, researchers (Kahneman, Flyvbjerg) and analyses (the “Pickrell report”) have recommended providing a reference class forecast alongside the forecast(s) developed using standard practice. Reference class forecasts, or “outside views”, provide the historical accuracy results given various project and corridor characteristics.
Toward this end, the author developed a database recording the characteristics of 62 transit forecasts, including components of the projects and forecasts. The database includes an assessment of the ridership forecast inputs by assessing the accuracy of “upstream forecasts” that travel forecasters are entirely dependent upon. Assessments were made for supporting and competing transit network, roadway congestion, project service level, and fare assumptions. Another unique feature of this database is it records the year in which the forecast was made, the opening year, and multiple observations of actual ridership. This database is being used by the author to provide reference class forecasts, insight on “upstream forecasts” that contribute to ridership forecast inaccuracies, and supplemental research.
The presentation will include the following items:
• Descriptions of the database’s transit projects, components, and accuracy assessment criteria,
• Review of the challenges the author faced when gathering the information for the database and recommendations for improved forecast archival,
• Tabulations of the database that provide preliminary insight into key forecasting accuracy questions:
o Has forecasting accuracy increased over time?
o Is forecasting accuracy better for extensions, new lines or new systems?
o Is forecasting accuracy better for certain transit modes?
o Does accuracy improve throughout project development?
o What are the main contributors of forecast uncertainty?
o Are forecasts made closer to project opening more accurate than those made further from opening?
o Are “upstream forecasts” always accurate when the ridership forecast is accurate?