EVALUATION OF AUTOMATIC PASSENGER COUNTERS: VALIDATION, SAMPLING, AND STATISTICAL INFERENCE

Whereas automatic passenger counters (APCs) offer the potential for cost-effective data recovery, they introduce new complications in the data recovery process. Three issues associated with the use of APCs are addressed on the basis of the experience of the Tri-County Metropolitan Transportation District of Oregon. The first issue is validation, which concerns both recovery and accuracy of APC passenger data. The second concerns the design of a sampling methodology for APCs compatible with UMTA's Section 15 reporting requirements. Third is inferring system-level ridership from sample data in the presence of selective APC failures. APCs provided systematically accurate passenger counts. Given that APCs recover operating data for all bus trips making up a vehicle schedule, a cluster sampling method was developed. Selective data recovery failures can bias estimates of system-level ridership. When data recovery rates vary by bus type, route type, or time of day, inferences may over- or under-represent total system ridership. In these circumstances, post hoc stratification of the sample is recommended. Several alternative corrections based on a priori knowledge of the mix of bus types and schedule characteristics in the system are presented.