Home

Behind the scenes with MBTA data.

In the last post, we took a broad look at ridership on the MBTA in 2020, and dove into the details on which types of passengers continued to ride the system. In this post, we’ll examine where passengers rode the system and how that changed from the patterns we typically see.

The above map (click to enlarge) shows the MBTA’s bus and rapid transit routes and lines and is colored by the amount of ridership change they saw, comparing Fall 2019 with Fall 2020 data. The ridership is normalized per revenue vehicle hour to account for changes in service levels between the two time periods. While all routes lost some ridership, the amount varied greatly – the least affected routes (colored in the yellow end of the color scale) retained about 60% of their ridership, while the most affected routes (deep red) lost nearly all of their ridership (while some had no service at all in this time period). You can explore this data in this file.

From this map, a few broad types of routes seem to have retained ridership particularly well:

  • Routes in Roxbury / Dorchester / Mattapan, Chelsea / East Boston, and Lynn / Salem
  • Routes that travel a long distance, such as the 70
  • Routes that provide the only service to a particular area, such as the 34E

Each of these categories fits with the ongoing research that transit is currently serving “essential” trips, and these are the types of routes that the MBTA prioritized in its Forging Ahead service changes.

Total Ridership Change by Station

We also took a look at how ridership changed by rapid transit station (excluding the Surface Green Line as detailed data is unavailable). This file shows the % change in average weekday validations from January / February (combined) to each month in 2020. The top and bottom 10 are shown below for October (excluding Lechmere and Science Park which were closed for Green Line Extension construction). October was chosen as the point post-pandemic when ridership was the highest, which would best illustrate the differences between stations.

  Jan./Feb. Avg. Weekday Validations Jan./Feb. Rank Oct. Avg. Weekday Validations Oct. Rank % Change
Revere Beach

2918

54 1680 34 -42%
Suffolk Downs 755 62 376 60 -50%
Beachmont 3266 51 1505 39 -54%
Maverick 11206 13 5046 2 -55%
Andrew 5232 39 2321 21 -56%
Orient Heights 4438 42 1900 29 -57%
Fields Corner 4770 40 2027 25 -57%
Airport 7011 26 2977 17 -58%
Bowdoin 2327 55 967 49 -58%
Charles/MGH 10387 16 4162 3 -60%

Most of the stations that had high levels of retained ridership were on the Blue Line. This is also reflected in the line-level data that we covered in our last post. Other stations on the top ten list likely have a high number of passengers without vehicle access (Andrew and Fields Corner), or are near major medical facilities (Charles / MGH) which of course continued operation throughout the pandemic.

  Jan./Feb. Avg. Weekday Validations Jan./Feb. Rank Oct. Avg. Weekday Validations Oct. Rank % Change
Kendall / MIT 16870 4 2382 20 -86%
South Station 24385 1 3666 7 -85%
Alewife 11295 12 1857 30 -84%
Courthouse 3600 48 647 57 -82%
Arlington 6595 29 1198 44 -82%
Porter 8284 20 1518 37 -82%
Oak Grove 6236 30 1185 45 -81%
Davis 11397 11 2181 24 -81%
Park Street 15544 7 3000 16 -81%
Harvard 16546 5 3313 10 -80%

It seems safe to conclude that the stations that lost the most ridership tended to be a combination of:

  • those with high numbers of usual riders who can work from home, 
  • high concentrations of college students, or 
  • those with high numbers of passengers who drive and park.

It should be noted that these are also some of the busiest stations on the system. This is important for two reasons: First, even with many usual passengers not riding, these stations still served significant numbers of riders. The 3,000+ taps seen daily at Park St, Harvard and South Station (not even counting commuter rail passengers) is comparable to the ridership at stations like Wollaston or Stony Brook during normal times. Even though they were among the stations with the highest portion of ridership lost, Park Street, Harvard, South Station and even Kendall were still in the top 20 busiest stations in October. Second, it confirms that the subway system is largely designed around bringing passengers to work in the usually busy areas in the center of the city, while the bus system tends to serve more radial and outlying trips. While this is not a new observation, rarely is it revealed in such a stark manner. Even starker is the drop in ridership on Commuter Rail, which is even more focused on serving the center of Boston and oriented around peak travel than subway and bus.

In our next post, we’ll look into how ridership patterns changed by time of day and throughout the week.

Ridership on the MBTA and public transit in general has dropped dramatically as a result of the COVID-19 pandemic. For this series of posts, we wanted to take a longer look at the year to review how ridership changed in three dimensions: by mode, over time, and by location.

Reports on ridership during the early months of the pandemic can be found in recent posts on the blog. We also continue to keep the public folder updated with downloadable datasets from which these charts and descriptions are based.

Notes on the Data

As a reminder, data on ridership is collected differently on different vehicles and therefore has different quirks and levels of reliability. Unless otherwise noted: 

  • Rapid Transit ridership (Blue, Red (including Mattapan), Green, and Orange Lines) is based on validations on MBTA fare equipment. While looking at the following charts, it is important to note that fares were not collected on buses, surface Green Line, and the Mattapan Line from mid-March through mid-July. For this time period, front-door boarding was suspended out of concern for safety of riders and MBTA employees. As a result, AFC (Automated Fare Collection) data for the early pandemic period does not account for riders on the mentioned modes.
  • Bus ridership is based on automated passenger counters (APCs) on-board buses. 
  • Commuter Rail and Ferry ridership are based on manual counts recorded by conductors and other staff. For Commuter Rail, not all trips are counted, so a placeholder is used based on the most recent real observations if there is no observation for a particular trip on a particular day.
  • The RIDE ridership is based on trips booked in the RIDE’s software.

Chart showing weekday ridership on all MBTA modes indexed to February 2020

The above chart (click to enlarge) shows ridership change over the year (weekdays only) compared to the baseline week of February 24-28, just before emergency orders were issued. Indexing every mode to this week hides the difference in volume of ridership between modes, but allows us to compare how much each mode and rapid transit line were affected. Showing just weekdays makes the trends easier to follow while still showing some day-to-day variation.

Some observations from this chart:

  • Bus and the RIDE ridership were affected the least initially, but still dropped to roughly 20% of the baseline ridership. They then recovered to about 40-45% of the baseline by September and have been largely steady since.
  • The Blue Line has retained higher ridership than the other rapid transit lines. The Blue Line initially dropped to about 12-13% of the baseline, but by August was comparable with bus and the RIDE, before dropping a bit below them in October.
  • The other Rapid Transit lines showed similar trends, with the Orange Line retaining the most ridership, then the Red Line, then the Green Line.
  • Finally, the Ferry and Commuter Rail retained the least amount of their original ridership. Ferry data is seasonally adjusted here and each day is compared to the weekday average from the same month in 2019. This was done to provide a better trendline since ferry ridership is very seasonal and the February baseline week would be quite low. 

Passenger Trends

Chart showing Gated station taps in 2020 compared with 2019

The above chart shows total validations at all gated stations, totaled by week (including weekends), compared to the same week in 2019. Weekly gated station validations dropped to their minimum for the year during the week of April 13th. During this week, there were 226,273 validations at gated stations or approximately 8% of the number of validations of the corresponding week in 2019. Gated station validations remained at or below 10% of 2019 levels until the end of May. Unlike 2019 ridership, which sees minor dips and spikes corresponding with significant events (e.g., a dip of about 500,000 validations during the week of July 4th), pandemic period 2020 ridership has grown and declined more steadily.

In absolute numbers, pandemic ridership reached a peak at the end of September, with weekly gated station validations reaching nearly 800,000 during the week of September 21st. Weekly gated station validations reached a maximum of 29% of 2019 levels during the first week of September.

In order to understand the changes that were happening on the system we were interested in finding out whether we gained back rides (individual trips) or riders (distinct people) faster. The below plot shows three different values relative to their 2019 levels. They depict the change in total validations, distinct cards seen, and what we call “frequent riders”: distinct cards that have validations on at least 3 days in a week.

Chart showing ridership patterns among different types of passengers compared to 2019

Chart showing ridership among different types of cards compared with 2019, focusing on August through October 2020

While the patterns over the summer are difficult to interpret because of the lack of bus fare collection, we see an interesting but perhaps not surprising pattern in the last few weeks of data. Relative to their 2019 levels, we see a greater decrease in the “frequent” cards than we do in the distinct cards seen. This indicates that frequent users of the T left the system at a higher rate than more occasional users. This could indicate that changes in the structure of work, with far fewer people going into an office 5 days per week, is impacting ridership patterns. That said, it could also be influenced by changes in the amount of “churn” (people losing their cards, or otherwise switching cards) or changes in who has signed up for Perq (usually frequent riders).

We are also curious about who is using the system during the pandemic months. Like ridership patterns, understanding this is important to structuring service and pass products. For example, throughout the pandemic, reduced fare products were used at a much higher rate than in previous years, likely indicating that the people who are still riding the T are disproportionately low-income.

Chart showing reduced fare validations as a proportion of all validations, comparing 2019 and 2020

This shows that riders who rely on reduced fare products to ride the T have made up a significantly larger proportion of weekly taps than during 2019. While the difference between 2019 and 2020 peaked early in the pandemic, this difference has persisted to the end of October. This is particularly interesting given that many reduced fare riders are students, and many schools were held remotely during this time. 

Other research has noted that these essential trips are the ones remaining on transit, so it should be unsurprising that reduced fare users, passengers on the RIDE and bus riders have continued to ride the system while many have chosen other modes or not made trips at all. In the next post, we will take a look at where and when travelers took the MBTA during the pandemic.

Fare collection was temporarily suspended when the MBTA implemented mandatory rear-door boarding on buses and trolleys at street-level stops due to safety reasons during the pandemic. On July 20, after safety protocols were put in place, fare collection on buses and trolleys at street-level stops resumed along with the resumption of front-door boarding.

While ridership remains far from normal, the return to fare collection provided a natural experiment for learning more about how the system is being used and how passengers respond to fares. Being curious, we dug into the data a bit to examine what happened after fare collection resumed. Usually, when fares change, we see a corresponding change in ridership that can be explained to some extent by elasticity: as cost to passengers increases, ridership drops, and vice versa. In this case, there are some confounding circumstances which are likely affecting ridership:

  1. Most importantly, because of the pandemic, many usual riders are not traveling, and non-essential trips are discouraged. The concern over safety and general guidance to stay home is likely to have had a far greater impact on travel than changing fares. We would therefore expect a lower elasticity effect on the trips still being taken, since we believe more trips than usual are by passengers with few other choices.
  2. The subway system continued normal fare collection throughout the entire time period, and the MBTA provides free transfers between bus and rail. The T also has a number of bus routes that “feed” into the subway system. This means that for many trips, there was no difference in cost when bus fares were not collected, as passengers would need to pay the subway fare on the other leg of their journey.
  3. Fare collection was only paused for four months, and the pause was for safety reasons, so we would not expect to see passengers change their behavior in the way they might if it were a permanent change (for example, selling their car). Economists usually describe both “short-run” and “long-run” elasticity; in this case we only saw the short-run.

The rest of this post will discuss the changes and provide some hypotheses about what might be causing the changes and what they might mean.

Limitations

Unfortunately, we by definition do not have any Automated Fare Collection system (AFC) data from the time when fares were not collected. That means we are unable to run our Origin-Destination-Transfers model (ODX) for that time period, nor can we follow particular cards or tickets over that time to see how their behavior may have changed when fares were re-instated. We at OPMI are used to having multiple data products at our disposal; during the pandemic, we’ve been unable to use many of them and have been limited in our ability to do field work or in-person surveys. So, we have to work with what we can. In this analysis, we largely used the Automated Passenger Counter (APC) data, which does not depend on fare collection and previous runs of ODX.

Chart showing ridership by day on buses and at gated stations for June and July 2020.

The above chart shows the bus ridership (from APCs) and the total gated station validations (from AFC) for the month before and after fare collection was resumed, along with a 7-day moving average. As you can see, while the trend on both bus and subway was slightly positive, when fare collection resumed on July 20, we saw a drop in bus ridership. This drop was about 9% from the week immediately previous to the fare collection change.

At the route level, we see a wide range in the ridership changes. Comparing the average ridership by route for the two weeks before fare collection with the average ridership for the two weeks after (and excluding routes with fewer than 500 passengers daily), the routes ranged from the SL3, which increased its ridership by 4%, to the 41, which dropped by 22%. Among Key Bus Routes, the biggest drop was the 1 with 16% fewer riders, and on the other end was the 28, which gained 2%. We have mapped the routes and shaded them by the amount of the change in the following map. You can also download the source data here.

Map of the change in ridership by route before and after resumption of fare collection.

There does not seem to be strong geographical variation in which routes were most affected. On the whole, routes which run further outside of the city center and have less overlap with the subway system seem to be less affected, but there are counter-examples as well. It is possible that riders on these routes have fewer other options, so they continued to take trips via bus at the same rate. For trips where passengers had a subway option, perhaps some passengers chose the bus leg when it was free but then either switched to subway or just walked further when bus fare collection resumed.

To test whether routes that are more often part of a multi-leg journey were less likely to lose ridership (discussed in bullet point 2 above), we first queried ODX to find how often a trip on each route was part of a multi-leg journey. We used data from July 2019 for this purpose – obviously data from 2020 would be preferable, but was not robust enough at this time. Below is a scatterplot of the transfer rate plotted against the change in ridership from before and after fare collection.

Scatterplot comparing the change in ridership with the transfer rate on each route.

From the scatterplot, the two measures don’t seem to correlate: the relationship is weak, and it is in the opposite direction from what we hypothesized. As transfer rate of the route increased, ridership actually tended to drop more. While the weak correlation does not disprove this hypothesis, it is likely that other factors are in play.

While we were unable to draw many conclusions from the difference in routes, we will keep thinking about this and as we get better data from 2020, will revisit the analysis. If you have any hypotheses about what factors may have affected these changes, email us at This email address is being protected from spambots. You need JavaScript enabled to view it.. We’ll update on the Blog if we discover anything!