smacmullan/transit-analysis

Planning

Requirements

• Take an array of bus visit timestamps at a stop and a timeframe. Calculate the AWT
• Take a timeframe, bus visit timestamps, and GTFS data. Calculate the EWT
• Combine AWT/EWT from multiple stops
• Combine AWT/EWT from multiple timeframes?
• bus bunching metric
  • Actual AWT vs AWT if buses were evenly spaced?
  • ATA 2018 analysis used percent of buses that arrived after another within a certain timeframe.
• bus route slowing metric - what sections of the route are the slowest? At what time of day?
  • MPH on route section?
  • dwell time on route section?
• actual performance vs theoretical ideal performance
  • How long does the bus take to make its route?
  • If the bus went at a constant speed with no stops, how long would it take?
  • Bus goes with no stops but a labeled speed. How long?
  • Labeled speed, constant stop time. How long?
  • What is the cause of additional timing
• AWT weighted by ridership during the day

External

• Package for grabbing CTA bus data, GTFS data, train data
  • Chicago has bus speed data in historical congestion tracker. Unfortunately it doesn't cover all the arterials.
• Live map of buses
• Use rideshare data to identify latent transit demand. trip data

Formulas

• AWT = SUM(D^2) / (2T)
  • Average value of a sawtooth function
• EWT = SWT - AWT
• SWT = scheduled wait time
• EWT = excess wait time
• AWT = actual weight time
• D = duration between buses

Scratchpad

Multiple stops

AWT_stop1 = SUM(D_1^2)/(2T) AWT_stop2 = SUM(D_2^2)/(2T)

AWT_avg = (AWT_s1 + AWT_s2)/2

• same denominators

ATW_avg = SUM (AWT_stop)/n

• where n = total number of stops

AWT partitioning

T = D AWT = D^2/(2D) = D/2

2nd partition (half before bus arrives) d = t = D/2 AWT = (D/2)^2/(2(D/2) = (D^2/4)/D = D/4

1st partition (half after previous bus) t = D/2 area = D^2/8 + D^2/4 = 3/8 * D^2 AWT = area/t = 3/4*D

• valid; average for people waiting in first partition will be longer than those in second partition
• doesn't intuitively represent what we want to show; normal service seems better or worse depending where you cut the timeframe
• need to find way to apply EWT continuously