Identifying Opportunities to Alleviate Bus Delay

Project Manager

Jonathan Belcher

Project Principal

Annette Demchur

Data Analyst

Nicholas Hart

Graphics

Nicholas Hart

Cover Design

Kim DeLauri

The preparation of this document was supported

by the Federal Transit Administration through

MPO §5303 Contracts #84080 and #91027.

Central Transportation Planning Staff

Directed by the Boston Region Metropolitan

Planning Organization. The MPO is composed of

state and regional agencies and authorities, and

local governments.

December 2016

To request additional copies of this document or copies in an accessible format, contact:

Central Transportation Planning Staff

State Transportation Building

Ten Park Plaza, Suite 2150

Boston, Massachusetts 02116

(857) 702-3700

(617) 570-9192 (fax)

TTY - (617) 570-9193

ctps@ctps.org

ctps.org

Abstract

Reliable bus transportation offers mobility, access, and livability benefits to residents throughout the region. However, delays in bus service diminish these benefits. This study seeks to identify opportunities to alleviate existing delays in bus service. To this end, MPO staff 1) examined factors that lead to increased dwell time at bus stops. This included estimating the amount of time contributed by passengers’ use of each fare payment type, as well as the time added by boardings of baby carriages, wheeled mobility devices, and portable shopping carts—along with any delay caused by the on-board presence of such items. In addition, staff 2) assessed delay caused by operational and scheduling practices—such as, the scheduling of interlines, scheduling of deadhead movements, on-time performance of garage pull-outs, scheduling of supplemental bus trips, scheduling of bus operator swing-ons, and non-utilization of early pull-up opportunity at trip origins.

Abstract

Chapter 1—Executive Summary

Chapter 2—Investigating Factors that Lead to Increased Dwell Time

2.1........ Data Collection—MBTA Routes 116 and 117

2.2........ Estimating a Dwell Time Regression Model

Chapter 3—Assessing Operational and Scheduling Improvements to Alleviate
        Bus Delay

3.1........ Interlines

3.2........ Deadhead Movements

3.3........ Garage Pull-Outs

3.4........ Supplemental MBTA Bus Trips

3.5........ Bus Operator Swing-Ons

3.6........ Early Pull-Up Opportunity

Table 1 Summary of Bus Routes 116 and 117 Boarding Observations

Table 2 Summary of Bus Routes 116 and 117 Load-at-Stop Observations

Table 3 Variables Tested for the Estimation of a Dwell Time Regression Model

Table 4 Coefficients of the Dwell Time Regression Model

Table 5 Estimation of Total Dwell Time and Run Time Contribution

Table 6 Run-Time Variability by Route—Charlestown District, Fall 2015

Table 7 Run-Time Variability by Trip End Point—Charlestown District, Fall 2015

Table 8 Run-Time Variability by Trip End Time—Charlestown District, Fall 2015

Table 9 Trips Leading into Non-Deadhead Interlines that Do Not Pass the Run-Time
 Variability Threshold—Charlestown District, Fall 2015

Table 10 Trips Leading into Deadhead Interlines that Do Not Pass the Run-Time
   Variability Threshold—Charlestown District, Fall 2015

Table 11 Deadhead Movements with Median Run Times Greater than  Scheduled
   Run Times, Spring 2016

Table 12 Summary of Departure Lateness from Garage, Spring 2016

Table 13 Summary of First Trip Arrival Lateness of Pull-Outs that Depart more than
   Three Minutes Late from the Garage, Spring 2016

Table 14 Summary of Early Pull-Up Opportunity by Route

Table 15 Summary of Early Pull-Up Opportunity—All Observations

Figure 1 Observed and Fitted Dwell Time

Figure 2 Estimated Delay from Pay Cash and Add Value Transactions by Trip

Figure 3 Estimated Delay from Baby Carriage Boardings and Alightings by Trip

Figure 4 Estimated Delay from Pay Cash and Add Value Transactions and  Baby
  Carriage Boardings and Alightings by Trip

Figure 5 Departure Lateness from Garage—Systemwide Observations,
  Spring 2016

Figure 6 Procedure for Assessing Swing-On Delay

Figure 7 Procedure for Assessing Early Pull-Up Opportunity

Chapter 1—Executive Summary

Reliable bus transportation offers mobility, access, and livability benefits to residents throughout the region. However, delays in bus service diminish these benefits. Upgrades to infrastructure, such as more efficient fare-collection equipment, as well as operational improvements, such as increasing the effectiveness of scheduling practices, could decrease bus delay, which in turn would improve on-time performance, reduce travel times for transit passengers, and make transit a more attractive travel mode. Using a variety of data sources, the Central Transportation Planning Staff (CTPS) to the Boston Region Metropolitan Planning Organization (MPO) analyzed these factors, and developed solutions to address the specific causes of delay.

Chapter 2 of this report investigates factors that lead to increased dwell time at bus stops by estimating the amount of time contributed by using each fare payment type, and by the boardings and on-board presence of baby carriages, wheeled mobility devices, and portable shopping carts, using observations from a sample of Route 116 and 117 trips. The results from the regression analysis estimate that the baseline boarding time (for a passenger paying with a CharlieCard or CharlieTicket) is about 3.3 seconds. An additional 9.4 seconds is added (12.7 seconds of total boarding time) if the passenger pays with cash, and an additional 6.3 seconds is added (9.6 seconds of total boarding time) if the passenger adds value to their CharlieCard. These results suggest that overall dwell time for the observed trips would be reduced by an estimated 11.3 percent if cash transactions were not permitted and by an approximate 2.6 percent if baby carriages had not boarded. Dwell time accounted for 14.4 percent of total run time. Therefore, staff estimate that total run time for the observed trips would be reduced by only 1.6 percent if cash transactions were not permitted and by 0.4 percent if baby carriages had not boarded.

Chapter 3 of this report assesses delay caused by a series of operational and scheduling practices; the scheduling of interlines, scheduling of deadhead movements, on-time performance of garage pull-outs, scheduling of supplemental trips, scheduling of bus operator swing-ons, and non-utilization of early pull-up opportunity at trip origins.

In analyzing interlines and deadhead movements, CTPS found that 193 of the 364 scheduled weekday interlines from the Charlestown bus district might be considered for de-interlining based on their run time variability observed in fall 2015. In addition, 41 of the 437 scheduled weekday deadhead movements systemwide had median run times greater than their scheduled run times in spring 2016.

In analyzing on-time performance of garage pull-outs, CTPS found that in spring 2016, 31 percent of pull-out trips departed more than three minutes late from the garage; and of these trips, 70 percent of them arrived at the start point of their first vehicle-revenue trip after the scheduled departure time, with an average arrival lateness of 3.9 minutes per trip.

For the analysis of supplemental trips, CTPS provided summaries of origin departure lateness, end-point arrival lateness, and run times, and identified how each trip might be scheduled more effectively. It is important that these trips run as scheduled, as many of them start a string of trips that provide service throughout the PM peak period.

In analyzing bus operator swing-ons, CTPS estimated that swing-ons added an extra 59 seconds to each trip where a swing-on occurred. Staff also found that 24 percent of scheduled swing-ons occurred on trips where the bus arrived at the swing-on location already behind schedule.

In analyzing early pull-up opportunity at trip origins, CTPS found that earlier pull-ups at origins would not contribute to a significant reduction in departure lateness on the routes that staff studied. Of the 25,169 minutes of observed departure lateness, staff estimated that only 1,441 minutes (5.7 percent) would have been saved if buses, that had the opportunity to do so, had pulled up earlier.

Chapter 2—Investigating Factors that Lead to Increased Dwell Time

2.1      Data Collection—MBTA Routes 116 and 117

MPO field staff (CTPS) conducted manual observations of baby carriage boardings, wheeled mobility device boardings, and portable shopping cart boardings on sample trips of MBTA bus Routes 116 and 117 from January 12, 2016 to January 21, 2016. These observations were paired with dwell times, boarding and alighting counts from on-board Automatic Passenger Counters (APCs), and counts by fare transaction type from the MBTA Automatic Fare Collection (AFC) database. A total of 97 trips of usable data was collected from 23 Route 116 inbound trips, 24 Route 116 outbound trips, 23 Route 117 inbound trips, and 27 Route 117 outbound trips. Appendix A provides a summary of boarding observations for each trip; Table 1 displays a summary of all boarding observations; and Table 2 displays a summary of all load-at-stop observations.

Table 1
Summary of Bus Routes 116 and 117 Boarding Observations

(Sampled Trips from January 12, 2016 to January 21, 2016)

AFC = Automatic fare collection.

Note: Counts do not include activity at first and last stop of a trip. Add-value transactions are also counted as CharlieCard transactions.

Table 2
Summary of Bus Routes 116 and 117 Load-at-Stop Observations

(Sampled Trips from January 12, 2016 to January 21, 2016)

Note: Counts do not include activity at first and last stop of a trip.

2.2      Estimating a Dwell Time Regression Model

Using the APC data, AFC data, and data collected from manual observations on sample trips of MBTA Routes 116 and 117, CTPS conducted a regression analysis to estimate the amount of dwell time contributed by a set of variables. Table 3 displays a list of variables that were tested.

Table 3
Variables Tested for the Estimation of a Dwell Time Regression Model

AFC = Automatic fare collection. APC = Automatic passenger counters. WMD = Wheeled mobility devices.

CTPS estimated an ordinary least squares linear regression model1 . The first and last stop of each trip and stops that included a wheeled mobility device boarding or alighting were removed from the regression analysis, as bus doors are often left open for a longer period of time than is required to board and alight passengers at the first and last stop, and the procedure to board or alight a wheeled mobility device may alter the typical manner in which additional customers at the stop board and alight. The functional form of the model is shown below, and its coefficients are shown in Table 4.

Table 4
Coefficients of the Dwell Time Regression Model

AFC = Automatic fare collection. APC = Automatic passenger counters.

Note: Multiple R-squared = 0.641, degrees of freedom = 1,785.

The results from the regression analysis estimate that the baseline boarding time (for a passenger paying with a CharlieCard or CharlieTicket) is about 3.3 seconds. An additional 9.4 seconds is added (12.7 seconds of total boarding time) if the passenger pays with cash, and an additional 6.3 seconds is added (9.6 seconds of total boarding time) if the passenger adds value to their CharlieCard. When the variable representing the number of CharlieTicket transactions was included in the model, it was well outside of a five percent confidence level, suggesting that CharlieCard and CharlieTicket transactions take a similar amount of time to complete. Furthermore, the model estimates that dwell times increase as passenger loads increase, once loads are greater than 30 passengers2 , at a rate of about 0.3 seconds per additional passenger, and that baby carriages contribute 9.6 seconds when boarding and 7.2 seconds when alighting.

Figure 1 displays the relationship between observed dwell time and fitted dwell time. From Figure 1 it appears that the model tends to underestimate when observed dwell times are long. This is because observations with long dwell times are more likely to include additional delay factors that were not specified in the model.

Figure 1
Observed and Fitted Dwell Time

(Sampled Trips from Bus Routes 116 and 117,

January 12, 2016 to January 21, 2016)

The coefficients estimated by the dwell time regression model were applied to the dataset to estimate the total contribution to dwell time and run time3 for each variable, shown in Table 54 . The results estimate that overall dwell time for the observed trips would be reduced by 11.4 percent if cash transactions (pay cash and add-value) were not permitted and by 1.5 percent if baby carriages had not boarded. Dwell time accounted for 15.6 percent of total run time. Therefore, it is estimated that total run time for the observed trips would be reduced by 1.8 percent if cash transactions were not permitted and by 0.2 percent if baby carriages had not boarded.

Table 5
Estimation of Total Dwell Time and Run Time Contribution

(Sampled Trips from Bus Routes 116 and 117,
January 12, 2016 to January 21, 2016)

AFC = Automatic fare collection. APC = Automatic passenger counters.

Note: Three boardings and three alightings of wheeled mobility devices were observed. The combined dwell time for these observations was 502 seconds, accounting for 1.6 percent of total observed dwell time (0.2 percent of total run time).

The relatively small impact of cash transactions on total run time suggests that fare system upgrades that eliminate on-board cash transactions may improve reliability for specific trips or during certain time periods that experience relatively high cash usage. However, these upgrades likely would not reduce run times on the observed trips to a degree that allows service to be scheduled more frequently with the same amount of resources (unless significant time savings are achieved through all-door boarding). Figure 2 shows that the estimated amount of time added by pay cash and add value transactions was less than two minutes for all observed trips, with the largest estimated amount of time added for a single trip being 2.0 minutes (4.7 percent of that trip’s total run time). Figure 2 also shows that trips with relatively larger estimated amounts of time added because of pay cash and add value transactions tended to occur during the midday period of service, with less of an effect during the AM peak and PM peak periods of service.

Figure 3 shows that the estimated amount of time added by baby carriage boardings and alightings was less than one minute for all observed trips, with the largest estimated amount of time added for a single trip being 0.8 minutes (1.2 percent of that trip’s total run time). Figure 3 also shows that, as with trips that have relatively larger estimated amounts of time added because of pay cash and add value transactions, trips with relatively larger amounts of time added because of baby carriage boardings and alightings tended to occur during the midday period of service, with less of an effect during the AM peak and PM peak periods of service. Figure 4 shows that the estimated amount of time added by pay cash and add value transactions and baby carriage boardings and alightings combined was less than 2.5 minutes for all observed trips, with the largest estimated amount of time added for a single trip being 2.3 minutes (5.4 percent of that trip’s total run time).

The variables representing shopping cart boardings and alightings, baby carriages on-board, shopping carts on-board, and wheeled mobility devices on-board were not significant at a five percent confidence level when included in the model; however, this does not necessarily mean that these variables are not significant contributors to dwell time in the real world. It is possible that the dataset did not contain enough observations of these variables to estimate their effects5. Because the on-board presence of baby carriages, shopping carts, and wheeled mobility devices were not significant contributors to the model, no conclusions on their effect on dwell time can be drawn, and it remains unclear whether a seating arrangement designed to provide more room for these devices might produce an overall decrease in dwell time.

The results of this analysis are contingent on the observations of only two bus routes, and a short observation period of seven weekdays6. Therefore, these results are not meant to reflect the MBTA system as a whole, as the effect and rate at which the variables in the model occur likely would vary geographically and temporally.

Figure 2
Estimated Delay from Pay Cash and Add Value Transactions by Trip

(Sampled Trips from Bus Routes 116 and 117,

January 12, 2016 to January 21, 2016)

Figure 3
Estimated Delay from Baby Carriage Boardings and Alightings by Trip
(Sampled Trips from Bus Routes 116 and 117,
January 12, 2016 to January 21, 2016)

Figure 4
Estimated Delay from Pay Cash and Add Value Transactions and
Baby Carriage Boardings and Alightings by Trip
(Sampled Trips from Bus Routes 116 and 117,
January 12, 2016 to January 21, 2016)

Chapter 3—Assessing Operational and Scheduling Improvements to Alleviate Bus Delay

CTPS assessed areas where improvements to operational and scheduling practices can be made to alleviate bus delay, including (3.1) interlines, (3.2) deadhead movements, (3.3) garage pull-outs, (3.4) supplemental trips, (3.5) bus operator swing-ons, and (3.6) early pull-up opportunity.

3.1      Interlines

An interline is when a vehicle is scheduled to complete a trip on one route and conduct its next trip on another route, either between two routes that share a common start/end point or through a deadhead movement. The benefit of interlining is that it allows resources to be utilized more efficiently, with less vehicle movements to and from the garage. A negative effect of interlining is the possibility that a route/corridor with normally reliable service can take on the delay from a route/corridor with poor on-time performance. From a customer-experience perspective, it is generally preferred that customers on one route/corridor are not negatively impacted by recurring delay from another route/corridor, and that routes with poor reliability are self-contained.

The MBTA uses the Horaires et Assignments pour Systems de Transport Urban et Semi-Urban (HASTUS) scheduling software to automatically block (form the string of trips that is assigned to an individual bus) and assign runs (the string of trips that is assigned to an individual operator) to build bus schedules for each bus district. The automatic blocking capabilities of HASTUS provide an opportunity to automatically schedule interlines into each garage’s schedule. CTPS was tasked with assessing run-time variability of trips to provide a list of suggested trips that should not be coded for interlining (assuming that proper run times, recovery times, and other factors are programmed into HASTUS for each trip). CTPS assessed the run-time variability of trips from the Charlestown bus district using fall 2015 Automatic Vehicle Location (AVL) data provided by the MBTA.

Staff utilized two thresholds to assess run-time variability; the 80 percentile run time compared to 120 percent of the scheduled run time if a trip is to be interlined at its end point, and the 90 percentile run time compared to 120 percent of the scheduled run time if a trip is to be interlined with a deadhead movement, to account for additional variability in the deadhead run time. Trips that do not pass these thresholds are not meant to form a definitive list of trips that should not be interlined, as there are scheduling techniques that may reduce the impact of run-time variability, such as increasing the scheduled recovery time. At most, the summary of trips based on these thresholds should be used to identify trips, routes, locations, and times of day where extra attention might be warranted.

Table 6 provides a summary of run-time variability by route; Table 7 provides a summary of run-time variability by trip end point; and Table 8 provides a summary of run-time variability by trip end time for trips from the Charlestown bus district.

Table 6
Run-Time Variability by Route—Charlestown District, Fall 2015

Table 7
Run-Time Variability by Trip End Point—Charlestown District, Fall 2015

Table 8
Run-Time Variability by Trip End Time—Charlestown District, Fall 2015

Table 9 provides a list of fall 2015 trips that were scheduled as a lead in to a non-deadhead interline that did not pass the 80 percentile run-time comparison to the 120 percent of the scheduled run-time threshold (147 out of 290 such trips).

Table 9
Trips Leading into Non-Deadhead Interlines that Do Not Pass the Run-Time Variability Threshold—Charlestown District, Fall 2015