Better Bus Project: Estimating Ridership Impacts of Service Change Proposals – Part 1

Read the below for an explanation of the methodology the MBTA’s Service Planning Department used to estimate the impacts on riders of the 47 Better Bus Project proposals affecting 63 bus routes.

This post has three parts; this is part 1.


Part 1

Methodology Overview

  1. Estimating Passenger Travel Time Impacts
  2. Estimating Change in Frequency and Passenger Wait Times

Part 2

  1. Estimating Passenger Walk Time Impacts
  2. Estimating Passenger Transfer Time Impacts
  3. Estimating Stranded Passengers
  4. Estimating Impacts Across the Day

Part 3

  1. Estimating Ridership Impacts
  2. Combining All Time and Ridership Impacts to Calculate Net Impacts


For more on the Better Bus Project and to learn how you can give feedback, please see the Project Page.

NOTE: A few of the summary documents created for each of the 47 proposals could not fit every potential service impact. In some cases, we had to omit some of the benefits or some of the trade-offs of a proposal.

For a full list of all service impacts—including weekend impacts—of all 47 proposals, see the document linked here:

Read the Document

Part 1

With the Better Bus Project, we endeavor to address the gaps and weaknesses in existing bus service. Too many of our bus routes fail to live up to our own standards—we want to change that.

Before we add more resources—more buses, more bus drivers, more bus garages—we have some long-overdue work to do. We need to focus our existing buses and operators where the demand is greatest. We need to eliminate unproductive route extensions or variations that don’t make sense anymore. We need to reduce the complexity of our network so that our services can be easy to use and competitive. That way, when we do add to our system, our customers will get better service for every dollar invested. To use a metaphor, we need to lay a better foundation before building a bigger house.

We have 47 proposals to update and modernize 63 existing routes. These are good and long-overdue changes. Taken together, they’re laying the foundation for a bus network with more frequent, reliable service that provides better connectivity in Greater Boston. A summary of each individual proposal can be found at These proposals include a description of each proposed change by route number, a map of the change, and estimates of the impacts in terms of time and ridership.

This post goes into the details of how those estimates were calculated. We’ll use the proposal to combine Route 1 and Route CT1 into a single route to demonstrate the application of various concepts and tools, along with references to other proposals when the 1/CT1 example doesn’t apply. Each proposal incorporates trade-offs — some riders will see changes to how they ride or walk to transit with the intent of providing better service. The goal of this methodology is to quantify these changes and to estimate whether each proposal has a net positive impact.

The methodology used to estimate proposal impacts incorporates the following calculations as appropriate.

1. Estimating Passenger Travel Time Impacts

To estimate the impact on rider travel times that a routing change will have, we use our automated passenger counter (APC) system if available, which tells us the number of passengers who board or exit the bus at each stop on a trip along with bus arrival times at each stop. 

We query the APC database to get the travel times between two stops by time of day.  When using the query of the APC database, we define the origin stop and the destination stop, along with the direction, day type (weekday, Saturday, or Sunday), and the date range. We set the date range large enough to get a good sample size. 

Where APC stop-to-stop data does not exist, either because there is a new route or the route doesn’t operate during certain times of the day, we use Google vehicle travel times between stops using the high end of the ranges provided.

As an example, consider the Route 1 route change proposal to omit the loop around Harvard Square and instead turn left onto Dunster Street from Massachusetts Avenue to Mt. Auburn Street. The APC travel times around Harvard Square from Massachusetts Avenue at Holyoke Street to Mt. Auburn Street at DeWolfe Street range between 7 and 9 minutes throughout the day on Saturdays. The corresponding Google travel times for trips between these two stops with the new route range from 3 to 6 minutes throughout the day. So the estimated travel time savings for any riders who travel through this segment ranges from 2 to 6 minutes across the day.

Map showing the change under new routing around Harvard Sq.

2. Estimating Change in Frequency and Passenger Wait Times

For any potential schedule change, the first step is to ensure that we provide sufficient scheduled travel time to allow for reliable operation.  To do that, we use run time data from the MBTA’s automatic vehicle location (AVL) systems. Each bus is equipped with a GPS device that records when the bus starts and ends the route.  With thousands of samples collected every day, we can review the statistical distribution of these travel times.  We set the scheduled run time between the route’s startpoint and endpoint to the median observed travel time. We set “layover” time, which is the scheduled time at the end of the bus trip to allow recovery from traffic and other variability, equal to the difference between the 90th and the 50th percentile run time.  So, the “run plus layover” time in each direction is set to the 90th percentile run time. The new “cycle” time is the round-trip scheduled run time plus layover and equals the sum of the inbound and outbound 90th-percentile run times.

Chart showing how to calculate cycle time

Once we have the recommended cycle time, we compare it to the current cycle time. The difference between these two times, summed across an entire service day for all trips, equals what we call the “scheduled deficiency,” or the gap between the time we currently provide in our schedules and what we should be providing. As the chart below shows, this gap can vary across the day, both in size and in direction. Generally, however, we find that current cycle times fall short of the recommended cycle times.

Chart showing current vs. 90th-percentile Cycle Times

One more bit of math and then we’ll get to how all this affects the analysis. Cycle time is related to headway (the time between scheduled trip departures) and vehicle count as follows:

Cycle Time = Headway × Vehicle Count

When we update scheduled cycle times to the 90th percentile run time (which typically is an increase), and vehicle count is constant, then the typical result is an increase in headway which means more time between trips or reduced frequency.

Using our example of the 60-minute cycle time from above, assume that we have 5 vehicles that we can assign to this theoretical route. Under the formula, this results in a headway of 12 minutes (60/5). Now assume that we updated the cycle time to match the 90th percentile, and it resulted in a cycle time of 70 minutes. Keeping the vehicle count of 5 constant, the result is an increase in the headway from 12 minutes to 14 minutes.

Once the cycle times are updated to reflect most recent travel time data, the next step for the estimated passenger wait time impact is to calculate the change in headway. This is the difference between the new headway resulting from the proposal (which assumes both the new cycle time and updates to the 90th-percentile run time) compared to what the headway would be if cycle times were only updated to reflect the 90th-percentile run time.

As an example, consider the Route 1 proposal above on Saturdays. The proposal is estimated to save 2 minutes at the median and up to 6 minutes in the morning. The cycle time savings is estimated to result in a 1-to-3-minute reduction in headway, from 15 minutes to 12 minutes from 7:00 AM to 9:00 AM. Note that the baseline headway assumes that cycle times are updated to reflect the 90th percentile (what they should be—in this case 15 minutes), not what scheduled headways currently are (12 minutes). 

The other type of change that would affect headway is shifting resources between routes or variations. While the travel time change adjusts the cycle time, a resource shift adjusts the vehicle count. All of these proposals assume that no new buses are available. However, many of the proposals shift resources, often the result of the elimination of a route or variation. In these cases, the vehicles that are currently used for a route or variation that is eliminated would be added to the remaining route or variation.

Going back to our example, consider the proposal to consolidate Routes 1 and CT1 on weekdays into a single Route 1 along with the Harvard Square change described above. We calculated the vehicle count for Route CT1 across the day, then added these counts to the Route 1 vehicle counts.

The existing headway for Route 1 at 7:30 AM is 10 minutes, and the updated run times increase that headway to 11 minutes. When the vehicle count from Route CT1 is added, the headway improves to 9 minutes. When the travel time savings from the Harvard Square route change is also added, the headway improves to 8 minutes.

Assuming random passenger arrivals, a rider’s wait time equals one-half of the headway. If a headway is 10 minutes, for example, some passengers will arrive just after the bus leaves and wait for 10 minutes, while others will arrive just as the bus arrives and wait for 0 minutes; on average, we assume that passengers wait 5 minutes.  If the headway change is estimated to save 2 minutes and go from 10 minutes to 8 minutes, the new expected wait time would drop from 5 minutes to 4 minutes. The wait time change is 1 minute or half the headway change of 2 minutes.

Next: Part 2 of this series, which explains how the MBTA estimated walk times, transfer times, stranded riders and how we estimated all these impacts over the course of the day.