Data Cleaning in Activity and Travel S urveys: A Methodology Applied to Walk Trips

摘要

Activity-based household travel surveys are becoming much more common as states andmetropolitan regions contemplate advancing their travel behavior forecasting abilities. Suchactivity and travel surveys are valuable for the estimation and calibration of activity-based traveldemand forecasting models. Before travel survey records can be used, they must be edited usingdata cleaning processes that identify, reject, and/or correct internal inconsistencies, miscodedinformation, and other errors. Literature on travel survey data cleaning is sparse, and few travelsurvey data cleaning standards exist beyond ad-hoc rules of thumb.This paper presents a possible methodology for improving on data cleaning rules ofthumb by borrowing statistical methods, especially from the field of robust statistics. Themethodology was applied to the walk trip records of a household activity and travel surveyconducted during 2011 in the Portland, Oregon, region. First, indicator variables wereconstructed to flag suspect walk trips. Next, visual inspection of the highest-ranking 5% ofsuspect walk trips was performed. The methodology identified 29 walk trips with an incorrectmode, 19 location errors, 39 trips with travel time errors, and 6 walk trips with inaccurate trippurposes. After correcting the mode errors and removing the trips with location errors, key walkcalibration statistics were more reasonable, demonstrating the usefulness of a statistically18derived data cleaning methodology. Finally, the paper concludes with recommended foci for datacleaning efforts of activity-based household travel surveys.