Accurate representations of transit travel flows are vital for informed transit planning and forecasting. However, empirical evidence suggests that traditional surveying approaches produce poor representations of actual flows. The Federal Transit Administration (FTA) recently conducted a research project to confirm this hypothesis, investigate reasons for the poor representation of flows and investigate approaches that generate reliable estimates of travel flows. FTA and its research team utilized faregate data and recent rider surveys from the Washington DC and Atlanta rail systems for this effort.

The team’s results appear to confirm that traditional surveying approaches produce poor representations of travel flows. Response rates vary by station-to-station movements, and that traditional approaches do not satisfactorily address these variances. Auxiliary flow count data is therefore required to detect and correct for these biases.

The team also investigated methods to utilize flow counts and rider survey data to produce reasonable estimates of flows. The team conducted several survey expansion experiments using traditional expansion approaches and iterative proportional fitting (IPF) methods with varying intensities of entry-to-exit flow sampling rates. Some IPF experiments also examined the impacts of maintaining zero-cell values in the sample flows versus replacing them with constant or modeled values. The results of the experiments were compared to the faregate data across time periods, volume categories, and suburban/Central Business District movements. The team also analyzed how varying sampling rates and expansion methods impact high-, medium-, low- and zero-value cells.

Lower sampling rates were found to fail in capturing small-valued cells thereby increasing overall error. The results indicate that while a 100% sample is optimal, for these two systems, a 20% sampling rate of flow counts produces reasonable flow estimates.

The paper will present the methodology, metrics and results of these experiments. Insights into surveying practice and potential implications to other systems will also be discussed.