Home

Behind the scenes with MBTA data.

At the onset of the pandemic, the MBTA lost a vast majority of its ridership (and therefore a large portion of its revenue) yet played a critical role in transporting the essential workers that kept the Boston area running. We’ve talked about some general patterns in COVID-era MBTA ridership on the data blog previously (see 2020 ridership in review posts). This post discusses results of a regression analysis of bus and rapid transit ridership and what it can tell us about who was riding the system as well as the benefits and drawbacks of some publicly available data sources. For more detail, a longer report detailing the results of these analyses is available here

In general, the MBTA lost a greater proportion of riders on rapid transit and the commuter rail than on bus services. However, there is significant variation between individual bus routes as well as between rapid transit lines. By analyzing similar variables for both rapid transit and bus, we can gain more insight into what drove ridership retention (or lack thereof) during the pandemic.

Changes in ridership were analyzed for two different periods: February to April 2020 (the time period with the lowest MBTA ridership) and February to October 2020 (which had the highest ridership during the pandemic). To measure bus ridership, Automatic Passenger Counter (APC) boardings on weekdays were averaged for each stop and then summed to the census tract level. Routes that ran run as directed (RAD) buses may have had some missing data during the height of the pandemic. Rapid transit ridership was measured by average weekday validations at gated stations. Half-mile catchment areas around stations were used to aggregate information about jobs and population for each station.

Two different sources of demographic information were used — the 2019 American Community Survey (which collects information on residential demographics) and the 2015-2017 MBTA Systemwide Passenger Survey. In general, the riders of the MBTA tend to vary from the residents in areas surrounding stations. This analysis confirmed that fact — for February to October ridership, better model fit was observed when using the demographics from the MBTA Passenger survey (even though the data are now several years old). Interestingly, when looking at February to April ridership retention, we saw better fit on the models using ACS data. In April, with many destinations closed and, in general, much fewer trips being made, people were more likely to make home-based trips than during non-pandemic times. This fact may have contributed to better model fit in the analysis of April ridership. 

The variable with the strongest relationship to rider retention across time periods, modes, and data sources was an index aimed at capturing the percentage of riders (or residents) in a given area that are low-income, people of color, or have limited access to a vehicle. Similar to the process used for Forging Ahead, this index is an attempt to capture “transit criticality.” Throughout the pandemic, at least on bus and rapid transit, routes and station catchment areas with higher proportions of people of color, low-income residents/riders, and people with limited vehicle access had greater ridership retention. For example, for April 2020, the following model results indicate a very strong positive (i.e., a higher index value indicates higher ridership/less ridership decline) relationship between the transit criticality index and ridership change at gated stations. 

Term Estimate Standard Error t-statistic P-value
(Intercept) -0.99174 0.019248 -51.524 0
Population -1.06E-06 9.42E-07 -1.127 0.265
Number of Jobs -3.16E-07 1.47E-07 -2.158 0.035
Number of Medical Jobs 1.64E-06 7.67E-07 2.142 0.037
Number of Educational Jobs -4.48E-06 1.40E-06 -3.213 0.002
Number of Retail Jobs 3.00E-08 3.04E-06 0.01 0.992
Number of Bus Routes in 1/4 Mile 0.002973 0.001067 2.786 0.007
CR Station within 1/1 mi -0.01761 0.01182 -1.49 0.142
Transit Criticality Index 0.078516 0.015527 5.057 0

Observations: 62, R2 = 0.550, Adj. R2 = 0.482

The influence of what types of jobs are near stations or along bus routes can also be seen. For example, rapid transit ridership retention is correlated with the number of medical jobs in the station catchment area. A greater number of jobs in the medical field correlated with greater ridership retention. This aligns with expectations given the unique circumstances of the pandemic as a ridership disruption, where throughout the pandemic medical workers continued to make trips. Consistent with the system-wide pattern that the model confirms, we see this pattern play out at Charles/MGH and Kendall/MIT. Validations at Kendall dropped to a lower point and stayed lower throughout the year than did validations at MGH, just over the Longfellow Bridge. 

Chart comparing ridership at Kendall/MIT and Charles/MGH stations in 2020

The influence of occupation type is also seen on bus ridership. In addition to the transit criticality index, the number of retail-sector jobs was significantly correlated with greater ridership retention throughout the pandemic. This highlights the variation in trip generation among different types of occupations. Retail jobs generate more trips than other types of work — in addition to workers themselves, retail stores serve as destinations, resulting in more trips on a given route. In contrast to the total number of jobs in a census tract was negatively correlated with ridership retention (consistent with the high proportion of teleworking). 

The analysis builds off some of the themes discussed in previous ridership data exploration, statistically confirming some hypotheses about ridership during COVID. Further, the difference between residents and riders can be seen in the data. The longer report can be found here, and visualizations of ridership and other trip-making can be found on the MassDOT Mobility Dashboard.

In our review of the impacts of the COVID-19 pandemic on ridership, we showed the following charts that display average entries at gated stations over the day. This shows us how ridership during the pandemic has been not only lower in volume, but also less focused on the traditional peak times around 8 AM and 5 PM on weekdays.

Chart showing average validations over the course of the day for Fall 2019

Chart showing average validations over the course of the day for Fall 2020

While the “traditional” pattern is clearly very peaky, the contrast becomes even more stark if you start to dig into the data more granularly. Here is a chart of average entries on weekdays (all lines) in Fall 2019, sorted into 5-minute buckets (compared to the 30-minute buckets above).

Chart showing average validations by 5 minutes over the course of the day for Fall 2019

In the five-minute period from 5:10 to 5:15 PM, the T averaged just over 6,200 taps at gated stations in Fall 2019. But, just a few minutes later from 5:25-5:30, the T averaged 4,998 taps, and a few minutes earlier, from 4:55-5:00, we averaged 4,679 taps per weekday.

Here is a chart showing the same time period, focusing just on the downtown stations on the Red Line: South Station, Downtown Crossing and Park Street:

Chart showing average validations in 5-minute increments over the course of the day at downtown stations for Fall 2019

From 4:55-5:00 PM, these three stations see an average of 840 taps per weekday. But just a few minutes later from 5:05-5:10, we average 1,240 taps. Some of these passengers entering at these stations are headed for the Green, Orange or Silver Lines, but other passengers are also transferring behind the gate from these lines to the Red Line. Fortunately, ODX estimates these boardings, and we have them already aggregated here by 15 minute buckets: 

Chart showing average boardings on Red Line over the course of the day for Fall 2019

Counting transfers and non-interaction, ODX estimates that over 4,000 people board the three core stations on the Red Line in the time period from 5:15-5:30 PM. By 6-6:15 PM, this has dropped to about 2,800 passengers.

If three trains pass through these stations in that time period and we assume passengers are traveling in both directions equally, that’s over 600 passengers (most of an entire train!) who are boarding each train during the busiest time. When you add in additional passengers who enter at busy workplaces like Kendall, Central, and Broadway, not to mention any passengers traveling across downtown, then it’s no wonder that crowding reaches extreme levels at this time of the day. With trains at capacity, there is very little margin for error: When there is even just a slight delay, or a particularly busy day of ridership, passengers are at crush capacity, are left behind, and ultimately have a bad experience. For some passengers, the crowding levels even when things are working perfectly is unacceptable and forces them to choose other options (We examined this with some focus groups and surveys in a blog post here).

Mass transit’s particular advantage over other transportation modes is its ability to move large numbers of people from a wide area into a particular place at the same time without them needing to park a vehicle. But when transit is over capacity, we have a “tragedy of the commons” situation where the service works significantly less well. And adding peak capacity is very expensive, and sometimes impossible. 

One of the potentially positive things to come from the pandemic is the likelihood of increased flexibility in work hours for many people who were previously part of the peak-of-the-peak crush load. Additionally, transit agencies throughout the country have realized (or emphasized) that their essential ridership – those who have continued to ride transit throughout the pandemic – are often riding outside of peak hours. Importantly, these non-work trips are also the types of trips that transit needs to be competitive for in order to reduce car ownership among inner core neighborhoods. 

If traditional 9-5 passengers are more flexible in when they leave work, and transit agencies are able to provide more service off-peak and in neighborhoods to better serve non-work trips, we could eventually reach a scenario where ridership is similar (or even higher) than pre-pandemic, but less concentrated on the peak-of-the-peak. Consider the following chart, which shows hypothetical Red Line boardings during the PM peak, if they were more spread out:

Chart showing average validations at peak times for Fall 2019 and a hypothetical post-pandemic scenario

To create this chart, a “capacity” of 3400 boardings per 15 minutes was applied to the data from Fall 2019 shown above. This amount of demand, given peak-level service, would create full, but not overcrowded trains in each direction. Additional boardings above this capacity were distributed among the other time periods between 3 and 7:30 PM in the same proportion as the existing ridership. To keep things simple, 54 boardings were eliminated entirely. In this purely illustrative scenario, we can see that nearly the exact same number of passengers board the Red Line between 3 and 7:30 pm, but they are simply less concentrated at the busiest time. Add in some additional ridership off-peak and on weekends, and you could see a scenario with even higher ridership than before. 

Time Period Fall 2019 Avg. Hypothetical Future Scenario
3:00 PM 1,397 1,469
3:15 PM 1,605 1,687
3:30 PM 1,700 1,787
3:45 PM 1,674 1,761
4:00 PM 2,085 2,193
4:15 PM 2,530 2,661
4:30 PM 2,744 2,885
4:45 PM 3,025 3,180
5:00 PM 3,685 3,400

5:15 PM

4,460 3,400
5:30 PM 3,738 3,400
5:45 PM 3,285 3,400
6:00 PM 2,867 3,014
6:15 PM 2,615 2,750
6:30 PM 2,325 2,445
6:45 PM 1,819 1,913
7:00 PM 1,604 1,687
7:15 PM 1,364 1,434
TOTAL 44,521 44,467

The MBTA would not, and likely could not enforce a particular capacity on boardings. But, this thought experiment illustrates that if future flexibility in passengers’ schedules allowed more passengers to board outside of the busiest time, the MBTA could carry just as many passengers as pre-pandemic. Just as importantly, this scenario would likely provide more timely (due to reducing the likelihood of being “left behind”), comfortable trips for all passengers, and more reliable service overall.

In the past two posts, we’ve given an overview of how ridership changed during the pandemic, both over the course of the year and spatially throughout the system. In this post, we’ll take a look at how patterns of ridership changed temporally on a weekly and daily level.

Ridership by Time of Day

Chart showing validations at MBTA faregates by time of day for Fall 2019

Validations by time of day, Weekdays 9/1/2019 - 12/31/2019

Chart showing validations at MBTA faregates by time of day for Fall 2020

Validations by time of day, Weekdays 9/1/2020 - 12/31/2020

Early on in the pandemic we showed these charts that show taps over the course of the day. We have updated them here, showing total validations per half-hour both for weekdays in Fall 2019 (9/1/19-12/31/19) and Fall 2020 (9/1/20-12/31/20). The updated charts look pretty similar – the pattern of ridership over the day did not change a lot from the beginning of the pandemic to the end of 2020. This suggests that those “essential” workers who were still riding in March continue to drive the patterns of ridership, while some other riders have returned but not concentrated in particular times.

The most interesting thing about these charts is the broad spread of the peaks. While passengers in normal times are highly concentrated in the peaks (and really, in the peak-of-the-peak) at 8 AM and 5 PM, we see in Fall 2020 that the number of boardings at 3 PM is almost exactly the same on average as the number at 5 PM. Additionally, the midday ridership is up to about 50% of the peak, while in Fall 2019 it was less than a third of the peak level. While we do not expect these patterns to continue in precisely the same way once we reach the “new normal”, even small changes in the concentration of passengers during the peak would have big implications for service provision, as often the extreme crowding at the peak-of-the-peak slows trains, increases dwell times, and reduces overall capacity.

Ridership by Day of the Week

We also saw different distributions of ridership throughout the week than we usually do. The following table shows the total validations at all gated stations on the average day of the week, for the same Fall 2019 and Fall 2020 periods as above. Holidays where the MBTA ran different levels of service than a weekday schedule are excluded.

Year Sunday Monday Tuesday Wednesday Thursday Friday Saturday
Fall 2019 Validations 155,783 454,830 473,372 505,383 503,383 474,962 210,193

(% of the week)

5.6% 16.4% 17.0% 18.2% 18.1% 17.1% 7.6%
Fall 2020 Validations 64,150 110,633 116,405 115,734 110,999 123,988 87,863
(% of the week) 8.8% 15.2% 16.0% 15.9% 15.2% 17.0% 12.0%

We have also made a chart showing validations by time of day over each day of the week. These are similar to the charts at the beginning of the post, but all lines are added together. We’ve highlighted Fridays, Saturdays and Sundays as they show the biggest changes (click to enlarge):

Chart showing ridership by day of the week, comparing Fall 2019 and 2020

Usually, Wednesdays and Thursdays have the highest ridership levels, which makes sense since people tend to take days off either at the beginning or the end of the week. All weekdays in Fall 2019 ranged between 16.4% and 18.2% of the total validations for the week. In Fall 2020, however, weekdays were a smaller proportion of weekly ridership, ranging from 15.2% to 17% of the total, which makes sense given the lack of 9-5 commuters. Interestingly, Fridays were the busiest weekday in 2020 by a significant margin. In the time of day charts, it appears the Friday morning peak is similar to the other weekdays, but ridership increases in the afternoon and evening. It is possible these additional passengers are working from home most of the week, but then ride the system on Fridays as they end their week.

Weekend days provided a higher proportion of ridership than usual in Fall 2020. While Saturdays in Fall 2019 were 44% of the average weekday, in Fall 2020 they had 76% of the average weekday’s ridership. Sundays showed a similar pattern, increasing from 32% of the average weekday (in 2019) to 56%. Again, this suggests that passengers who rely on transit for travel continue to ride each day or have different schedules than the usual peak patterns, while those who tend to mostly ride during the week are no longer traveling or have changed their modes.

While research continues, the characteristics of ridership continue to show that transit provides an essential service for those who don’t or can’t drive cars and who are unable to work from home. The MBTA continues to keep a close watch on ridership and monitor passenger behavior in multiple dimensions: volumes over time, volumes by route and location, and behavior over the course of the day. While we expect 9-5 commuters to return to the system, the pandemic has emphasized that hundreds of thousands of passengers rely on the MBTA who cannot work from home, and those passengers tend to more often travel at times outside the traditional peak. To best serve those essential trips well on into the future, we will need to carefully examine travel behavior using all available tools, and plan carefully, thinking beyond the usual emphasis on peak travel.