The primary objective of rider surveys is to produce a dataset that is a reflection of the real world. Accurate surveys are vital in transit planning and forecasting. Long-standing tradition states that a 10% sample of boardings provides statistically accurate results. While statistical theory implies that this overall sampling rate is valid, ample empirical evidence leads one to reject this tradition. The Federal Transit Administration (FTA) recently led a research project that analyzed the sampling rates of two recent rail rider surveys. The results of this analysis may provide insight to how practitioners should approach sampling and determine sampling rates.
The FTA and its team applied the sample size formula with the finite population correction factor to a recent Washington, DC, rail rider survey. The formula was applied to a situation where the objective is to produce accurate estimate of flows by income group. It was applied to one high-volume and one low-volume station given three different assumptions:
1. Accurate information on flows is available, but no prior income distribution Information is available,
2. Accurate information on flows is available, but prior income distribution Information is available, and
3. Accurate information on flows is available, prior income distribution Information is available, and pre-survey screening is employed to avoid over-sampling populations.
Each assumption represented an individual “case study”. For each case study, FTA and its team examined the actual statistical accuracy achieved by the survey for income groups by each station-to-station group movements. The team then computed an optimal sample size given confidence interval and ridership constraints. By comparing the actual versus optimal sample sizes, the team was able to draw valuable insights into current survey practices and how they affect statistical accuracies of the results.