Transcript Document
NTD Reporting Estimating boardings and passenger miles with automatic passenger counters and Ridecheck Plus APC University October 17, 2014 Houston, TX Nancy Edmonson Consultant and Qualified Statistician Houston, Texas NTD Reporting by APC Data Mining FTA Requirements • Considered an “alternative sampling technique” • Even with 100% APC-equipped vehicles, more than 2% of data is missing or discarded • Meet NTD standard of 95% confidence and +/10% accuracy • For APC-based data, parallel estimation for boardings and passenger miles for one year NTD Reporting by APC Data Mining Statistician’s view of APC data mining • Applies up to 68 business rules to eliminate bad or suspicious data • Business rules can be customized to ensure relevance at each agency • Amount of discard data varies by agency, but typically 20% to 30% • Resulting sample is huge – typically more than 200,000 trips • Takes advantage of huge dataset with automated “rigor” to ensure validity NTD Reporting by APC Data Mining Key is Data Cleaning • Covered earlier by Rich • Key strength of the methodology • Business rules set to be appropriate for each agency • Block level and trip level checks • Eliminates bad data as well as adjusts good data (e.g., carryover rule) NTD Reporting by APC Data Mining Basic Process – Work up from one survey 1 NTD Reporting by APC Data Mining Summary of same trip 2 NTD Reporting by APC Data Mining Average trip for entire month 3 NTD Reporting by APC Data Mining Factor average trip into monthly total 4 NTD Reporting by APC Data Mining Process boardings for all trips and routes 5 NTD Reporting by APC Data Mining Process PM for all trips and routes 6 NTD Reporting by APC Data Mining Survey sample huge -count by route and month NTD Reporting by APC Data Mining Process can include adjustments such as accuracy factor NTD Reporting by APC Data Mining Check for bias, and check again • Is system 100% APC-equipped? • Are all routes, contractors and vehicle types represented in sample? • On lightly patronized trips, is operator activity substantially affecting ridership • In areas of high ridership, how well do the APCs deal with crowds? • Are proper adjustments make for service interruptions, missed service, or special schedules? • Does agency have proper procedures for checking and calibrating APCs? • If accuracy factor is used, is checker data representative of system? • Were the checkers themselves reliable and accurate? NTD Reporting by APC Data Mining Legacy method versus APC data mining over one fiscal year Legacy Method APC Data Mining Difference 3,149,013 3,263,994 +3.7% 10,179,840 10,592,219 +4.1% Boardings 11,320,995 11,443,495 +1.1% Passenger Miles 26,539,392 27,101,448 +2.1% Agency A Boardings Passenger Miles Agency B I, Nancy Edmonson, a qualified statistician, certify the APC data mining methodology described herein yields estimated boarding and passenger miles at the 95% confidence level within 10% precision as required by NTD.