Demographic and Built Environment Predictors of Public Transportation Retention and Work-from-Home Changes in Small- to Medium-Sized Massachusetts Cities, 2011–2021

Abstract

Transportation uses substantial energy and is a significant household expense in the United States; public transportation and working from home present opportunities to reduce energy use and increase household affordability. However, during COVID-19, transportation systems reduced service, and nationwide, public transportation use has been declining. Focusing on six small-to-medium-sized “Gateway Cities” in Massachusetts—more affordable cities with lower-than-state-average median income and lower-than-state-average education—that have regional transit systems and are within Boston’s commuter rail area, we analyzed the changes in public transit ridership and work from home. We estimated linear and hierarchical linear regression models to understand the association between demographics and built environment and lower emission modes to work between 2011 and 2021. We used GIS to visualize the distribution of public transit ridership and work from home over time and space. We found that the block groups in our sample retained public transit users over the study period and saw increases in working from home. Across all cities, transit ridership was more likely to increase in block groups with higher accessibility to jobs and more frequent transportation to those jobs; work-from-home was more likely to increase in block groups with a lower percentage of Hispanic residents and lower rent burden. We found that most block groups either saw an increase in ridership or working from home, suggesting that work from home and public transit users are spatially segmented groups.

1. Introduction

In planning for a sustainable city, it is important that planners find ways to reduce energy use per capita. Nearly 30% of the energy use in the United States of America (U.S.) is for transportation, which also is the second most expensive category for households after housing [1]. One way to reduce energy use in transportation and household transportation costs is by increasing the use of public transit. Having more workers work from home can also contribute to reduced energy use since this also reduces commuting by personal vehicles. While transit use in the U.S. is in crisis with falling ridership in many cities [2], in the post-COVID city, working at home may be a new option for some types of workers. Those who can work from home, however, may be very different from those who take transit to work [3,4].

In this context, we examine the change in transit ridership and work-from-home percentages by workers in six Massachusetts “Gateway Cities” near Boston. Massachusetts defines “Gateway Cities” as small-to-midsize cities that have below-average income and below-average education, while also having affordable housing stock and infrastructure [5]. We focus specifically on Brockton, Fall River, New Bedford, Worcester, Lawrence, and Lowell, former manufacturing centers defined as “Gateway Cities” since 2007 [5]. These cities do not have easy access to the main transit provider for the city of Boston and the towns that surround it—the Massachusetts Bay Area Transit Agency (MBTA). However, they do have local transit providers and are relatively close to the commuter rail stops that provide heavy rail connections to the city of Boston and its surrounding towns, which are the largest employment hub in the region.

Massachusetts policies direct resources to redevelop and economically develop these Gateway Cities; however, these improvements may have less-than-sustainable outcomes, like increased commutes by personal vehicles and increased housing costs. To investigate the current landscape of transportation use and housing affordability in these towns, we ask: What local characteristics are associated with changes in sustainable journey-to-work practices—working from home and traveling to work on public transit—between 2011 and 2021 in these cities?

To answer this question, we first compare the percentage of workers who used transit and work from home in 2011 and 2021 to identify if there were patterns of change over space and time in these cities. Then, we estimate models that predict transit use and change as well as work-from-home use and change over this time period. Other studies have noted the importance of socioeconomic and housing characteristics as well as the built environment in predicting transit use; we also account for these characteristics [6,7]. While we are generally interested in public transit use retention and work from home changes, we are also interested in understanding how changing demographics as well as housing and the built environment may be impacting these dynamics. In the context of these two objectives, we review the literature related to transit and work-from-home characteristics in the next section.

1.1. Transit Service and Ridership Changes before and after COVID-19 Lockdowns

We situate this question about changes in transit ridership and work from home in small-to-midsize U.S. cities within the context of U.S. public transportation usage decreases, the relationship between transportation and the built environment, the intersection between transportation access and affordability, and current research on the relationship between public transit usage and sustainability. While larger U.S. metro areas are experiencing a decline in public transit ridership and a reduction in convenient service to low-income people, recent immigrants [8], and ethnic and racial minorities, it is unclear the degree to which these trends are present in smaller urban areas with public transit.

Before the COVID-19-related transportation declines in 2020, U.S. transit ridership had already been declining, particularly bus ridership [9]. Prior research points to three major explanations: a decline in effectiveness of public transportation service, the introduction and growth of rideshare services, and the increasing affordability of automobile transit. Rideshare, or TNCs (Transportation Network Companies), usage is associated with some increases in public transportation use, making up for anticipated declines based on reduced effectiveness, particularly for rail ridership, as riders use rideshares to travel between rail stations and their trip origins and destinations [10,11]. However, TNC negatively impacts bus ridership, and is not a complementary service for bus riders as it is for rail users [10,12]. Finally, increased car ownership is associated with declines in public transit use between 2000 and 2019, explained by relatively lower costs of vehicles and gasoline in the U.S. [13,14,15]. More convenient individual modes of transit were overtaking public transportation use even before COVID-19 declines.

During the COVID-19 lockdowns in U.S. cities, public transportation use and service declined further. Transit-dependent riders became the majority of users during the lowest ridership times during pandemic-related shutdowns [16,17]. Analyses of changes in transit usage during the pandemic found that areas with lower income, higher poverty, and a higher percentage of Hispanic and Black residents retained ridership, and areas with higher percentages of work from home and automobile access lost riders [18]. Further, during COVID-19, transit system service reductions mostly impacted places with already vulnerable populations; cities with more sprawl were more likely to have service cuts than compact cities [19]. Even as transit systems served their vulnerable ridership least, mobile tracking showed that low-income riders had fewer reductions in trips during the pandemic, suggesting less flexibility in how much travel they required [16]. COVID-19 service reductions, rider concerns about COVID-19 spread in confined buses or trains, and COVID-19 and post-COVID-19 work-from-home opportunities dramatically changed public transit usage [20].

Overall, transit usage since the COVID-19 lockdowns has been lower than before, though ridership began to increase in the U.S. in 2021 [21]. Ridership changes include lower peak transit use and more infrequent bus riders but continued off-peak use and consistent ridership by pre-pandemic bus riders in large North American cities [3,22]. Noting that most research on the post-COVID-19 lockdown transit use decline focuses on large cities, a case study in a smaller U.S. county found that changes in ridership varied based on sociodemographic characteristics, though affordable or free bus ridership was more resilient than less-widely-available bikeshare, even with reduced bus service [23]. Ngo and Martin [23] found that bus ridership declined for some sociodemographic groups but increased for others, suggesting that even in relatively homogeneous small counties, changes in access to and use of transportation may not be equitable.

1.2. The Relationship between Transit, Built Environment, and Affordability

Previous research on the relationship between housing and public transit use focuses on who is using transit and where, with the intent to better match public transportation services to people who need them most—lower-income workers without cars, typically minorities and recent immigrants. In the U.S., most cities are designed for automobile transportation, and public transportation is a service that allows non-automobile users to travel through automobile-centric spaces [15]. In this landscape, car ownership is prevalent, and those unable to or unwilling to own cars rely on other transportation means to get to and from work and other destinations [24].

The combination of transportation and housing costs has been used by policymakers to assess overall location affordability, since transportation costs are typically a U.S. household’s second-largest expense after housing. For lower-income households, the combination of transportation and housing costs can be unaffordable, particularly when lower-priced housing is located far from urban, employment-rich areas, requiring a vehicle or long and/or expensive travel times to reach city center jobs [25,26]. Analyzing location efficiency to identify the optimal locations for minimizing housing and transportation costs is one way of assessing and planning for this combined expense; however, analyses typically focus on where rapid transit contributes to lower transportation costs (versus car usage), not bus usage [27,28]. While empirical evidence about whether reduced transit costs actually offset increased housing costs, research on location efficiency for low-income and poorer households finds that the cost of living in transit-rich areas may not be affordable regardless of the offset cost of transportation [29].

Finally, just because public transit is available does not mean that the people who need it most will be able to afford to live there or will be able to stay. Research on transit gentrification has mixed findings; however, some studies point to increases in housing prices near new light rail stations in the time around station opening [30], and in Eastern Massachusetts, areas near commuter rail service may have higher house values [31]. Introducing bus rapid transit can increase transit accessibility without increased housing costs, potentially because the amenity value of transit does not outweigh the nuisance effects of nearby bus transit [32].

1.3. How Travel-to-Work and Work-from-Home Contribute to Sustainable Transportation Systems

Prior research identifies that the people likely to use public transit are minorities, women and low-income households employed in essential services [33,34]. Remote work increased dramatically with the onset of the pandemic and appears likely to remain elevated for many years to come [35,36]. Speroni et al. [36] note that while remote work or telecommuting was seen in the past as a panacea for traffic congestion and vehicle emissions, though researchers have found that it is more likely to increase overall driving by remote workers. On the other hand, many researchers have noted that public transit systems, in contrast to street and highway systems, have lost ridership due to the rise of remote workers since they depended on commuters to downtowns and job centers [9,34]. The demographics of remote workers are also quite different from those who depend on transit, since they tend to be both wealthier and have higher educational attainment [37,38].

Prior research on the sustainability benefits of public transit focuses on reducing vehicle miles traveled and improving accessibility for all residents in a city [39,40]. A recent review on the sustainability implications of TNC found both complementarity and replacement in TNC’s impact on public transportation ridership [41]. Further, they pointed to the longer-term question of whether it would ultimately impact individual vehicle ownership [41]. Indeed, Soria et al. [34] note that seamlessly incorporating TNC within the public transit system with shared payment systems would entice riders back to public transit. Fare-free public transit leads to increases in ridership [42].

While working from home reduces the emissions associated with transportation to and from work, Moos and Skaburskis [43] found that people who worked from home were more likely to have single family homes, regardless of their demographics, and that the type of housing stock supporting at-home workers may not offset their emissions reduction in travel. While working from home may reduce commute-to-work travel Lachapelle et al. find that this has sustainable transportation outcomes only when a person actually works at home, rather than working from a non-work, non-home location [44]. Further, the opportunity to work at home is not evenly distributed across types of work or workers. Lower-wage workers—disproportionately immigrant, younger, ethnic minorities, and lower-educated—were less likely to be able to work from home during COVID-19 [45].

To make transportation in cities sustainable it is key that planners find ways to make public transit more attractive to both commuters and those who work-from-home and need to make non-work trips. To do this, it may be important to understand that both the built environment and other modes like TNC, walking, and biking, as well as sociodemographic and transit capacity, are all intricately linked.

2. Materials and Methods

2.1. Selection of Cities

Six Massachusetts “Gateway Cities” in Eastern Massachusetts were selected for analysis. The cities were selected because they all had local bus transit systems and were not part of the MBTA Rapid Transit system and are within driving distance of the MBTA commuter rail that connects to the city of Boston. Figure 1 shows the six cities and their relationship to Boston and the commuter rail. Two of the cities are also in the commuter region for Providence, Rhode Island. All cities qualify as Massachusetts municipalities, defined by the State of Massachusetts for economic development incentives as cities between 35,000 and 250,000 people, and below statewide median income and college degree attainment [46], and all six were part of the original 11 gateway cities defined in the MassINC and Brookings Institution report on Massachusetts cities not transitioning effectively from mill towns to the knowledge economy [5]. These six cities are particularly interesting to our query as they are near the higher-cost Boston metro area but are affordable cities for lower-income people and recent immigrants to move to.

2.2. Population, Transportation, and Built Environment Characteristics

The data used for this paper were from the five-year estimates of the American Community Survey (ACS) of the U.S. Census for the years 2007–2011 and 2017–2021, queried via the database Social Explorer. In total, there were 607 census block groups in 2011 and 535 in 2021 in the cities that were included in this study. The Census estimates were aggregated to the census block group level for both years. General Transit Feed Specification (GTFS) data from local transit agencies were used to calculate transit frequencies for 2012 and 2022 for both local bus and commuter rail. The local agencies from which we collected GTFS data include Worcester Regional Transit Authority (WRTA) for Worcester, Southeastern Regional Transit Authority (SRTA) for Fall River and New Bedford, Lowell Regional Transit Authority for Lowell, MVRTA—Transit for the Merrimack Valley for Lawrence, Brockton Area Transit Authority for Brockton, and MBTA for Commuter Rail. The GTFS data were at the stop level, which was then aggregated to averages by census block group. The U.S. Environmental Protection Administration’s (EPA) smart location database [47] was used to quantify the built environment in terms of its walkability and transit-friendliness. They provide data aggregated to the census block group level. The comparison time periods—2007–2011 and 2017–2021—were the most recent non-overlapping time periods for which demographic (ACS) and GTFS data were available.

Dependent variables included the percentage of the block group that took transit to work amongst workers over the age of 16. The census asks respondents who worked in the last week about their primary means of transportation to work. Respondents select from a list the method they used for the longest distance in their typical trip to work, and we use “Public Transportation (includes Taxicab)” and “Worked at home” percentage use among the total number of workers 16 years and over in the census block group as the dependent variables. We also calculated the ratio of the percentage using transit in the two years to find locations where transit use remained stable. The change in the percentage of workers working from home was also one of the dependent variables in this paper. Since some block groups were split or combined, we spatially joined overlapping block groups across the two years.

In summary, we calculated the following dependent variables at the block group level:

$$\mathrm{T}\mathrm{r}\mathrm{a}\mathrm{n}\mathrm{s}\mathrm{i}\mathrm{t} \mathrm{P}\mathrm{e}\mathrm{r}\mathrm{c}\mathrm{e}\mathrm{n}\mathrm{t}\mathrm{a}\mathrm{g}\mathrm{e} \mathrm{C}\mathrm{h}\mathrm{a}\mathrm{n}\mathrm{g}\mathrm{e}=\mathrm{T}\mathrm{r}\mathrm{a}\mathrm{n}\mathrm{s}\mathrm{i}\mathrm{t} \mathrm{p}\mathrm{e}\mathrm{r}\mathrm{c}\mathrm{e}\mathrm{n}\mathrm{t}\mathrm{a}\mathrm{g}\mathrm{e} 2021-\mathrm{T}\mathrm{r}\mathrm{a}\mathrm{n}\mathrm{s}\mathrm{i}\mathrm{t} \mathrm{p}\mathrm{e}\mathrm{r}\mathrm{c}\mathrm{e}\mathrm{n}\mathrm{t}\mathrm{a}\mathrm{g}\mathrm{e} 2011$$

(1)

$$\mathrm{T}\mathrm{r}\mathrm{a}\mathrm{n}\mathrm{s}\mathrm{i}\mathrm{t} \mathrm{R}\mathrm{e}\mathrm{t}\mathrm{e}\mathrm{n}\mathrm{t}\mathrm{i}\mathrm{o}\mathrm{n}={\displaystyle \frac{\mathrm{T}\mathrm{r}\mathrm{a}\mathrm{n}\mathrm{s}\mathrm{i}\mathrm{t} \mathrm{p}\mathrm{e}\mathrm{r}\mathrm{c}\mathrm{e}\mathrm{n}\mathrm{t}\mathrm{a}\mathrm{g}\mathrm{e} 2021}{\mathrm{T}\mathrm{r}\mathrm{a}\mathrm{n}\mathrm{s}\mathrm{i}\mathrm{t} \mathrm{p}\mathrm{e}\mathrm{r}\mathrm{c}\mathrm{e}\mathrm{n}\mathrm{t}\mathrm{a}\mathrm{g}\mathrm{e} 2011}}$$

(2)

$$\mathrm{C}\mathrm{a}\mathrm{r} \mathrm{P}\mathrm{e}\mathrm{r}\mathrm{c}\mathrm{e}\mathrm{n}\mathrm{t}\mathrm{a}\mathrm{g}\mathrm{e} \mathrm{C}\mathrm{h}\mathrm{a}\mathrm{n}\mathrm{g}\mathrm{e}=\mathrm{C}\mathrm{a}\mathrm{r} \mathrm{p}\mathrm{e}\mathrm{r}\mathrm{c}\mathrm{e}\mathrm{n}\mathrm{t}\mathrm{a}\mathrm{g}\mathrm{e} 2021-\mathrm{C}\mathrm{a}\mathrm{r} \mathrm{p}\mathrm{e}\mathrm{r}\mathrm{c}\mathrm{e}\mathrm{n}\mathrm{t}\mathrm{a}\mathrm{g}\mathrm{e} 2011$$

(3)

$$\mathrm{C}\mathrm{a}\mathrm{r} \mathrm{R}\mathrm{e}\mathrm{t}\mathrm{e}\mathrm{n}\mathrm{t}\mathrm{i}\mathrm{o}\mathrm{n}={\displaystyle \frac{\mathrm{C}\mathrm{a}\mathrm{r} \mathrm{p}\mathrm{e}\mathrm{r}\mathrm{c}\mathrm{e}\mathrm{n}\mathrm{t}\mathrm{a}\mathrm{g}\mathrm{e} 2021}{\mathrm{C}\mathrm{a}\mathrm{r} \mathrm{p}\mathrm{e}\mathrm{r}\mathrm{c}\mathrm{e}\mathrm{n}\mathrm{t}\mathrm{a}\mathrm{g}\mathrm{e} 2011}}$$

(4)

$$\mathrm{W}\mathrm{o}\mathrm{r}\mathrm{k}\mathrm{f}\mathrm{r}\mathrm{o}\mathrm{m}\mathrm{h}\mathrm{o}\mathrm{m}\mathrm{e}\mathrm{P}\mathrm{e}\mathrm{r}\mathrm{c}\mathrm{e}\mathrm{n}\mathrm{t}\mathrm{a}\mathrm{g}\mathrm{e}\mathrm{C}\mathrm{h}\mathrm{a}\mathrm{n}\mathrm{g}\mathrm{e}=\mathrm{W}\mathrm{o}\mathrm{r}\mathrm{k}\mathrm{f}\mathrm{r}\mathrm{o}\mathrm{m}\mathrm{h}\mathrm{o}\mathrm{m}\mathrm{e}\mathrm{p}\mathrm{e}\mathrm{r}\mathrm{c}\mathrm{e}\mathrm{n}\mathrm{t}\mathrm{a}\mathrm{g}\mathrm{e}2021-\mathrm{W}\mathrm{o}\mathrm{r}\mathrm{k}\mathrm{f}\mathrm{r}\mathrm{o}\mathrm{m}\mathrm{h}\mathrm{o}\mathrm{m}\mathrm{e}\mathrm{p}\mathrm{e}\mathrm{r}\mathrm{c}\mathrm{e}\mathrm{n}\mathrm{t}\mathrm{a}\mathrm{g}\mathrm{e}2011$$

(5)

Socioeconomic variables included median income, the percentage of the census block group that were Black, the percentage of the census block group that were Hispanic, the percentage of adults that had less than high school education, the percentage of households living in poverty, and the percentage of the occupied housing units where there were no vehicles available. Housing variables included the percentage of the block group that was renter occupied and the percentage of renter-occupied households paying gross rent exceeding 50% of their income.

The built environment variables included a transit accessibility index defined by the EPA as an index of the relative accessibility of a block group compared to other block groups in the same metropolitan region, as measured by travel time to the working-age population via transit [47]. When the index has values closer to 1, the block group is more accessible. The walkability index was calculated by a weighted formula using results of indicator rank scores developed by the EPA. The index uses selected variables on density, diversity of land uses, and proximity to transit from the Smart Location Database. Index values that are closer to 20 are the most walkable, and values close to 1 are the least walkable.

2.3. Evaluating Change in Sustainable Transportation Usage

To assess changes over time, ratio variables were calculated to assess changes in sustainable transit behavior per census block group. Transit Retention summarizes the ratio of the percentage of transit use for work from 2011 to 2021. It aims to summarize the amount of 2011 transit usage that remains present in 2021; values greater than 100 indicate ridership decline, while values lower than 100 indicate ridership growth. This serves as a proxy for whether this location retained transit riders or lost them during this period. The same value was calculated for Car Retention to describe the ways that car ownership changed.

Further, to assess changes over space, we summarize the percentages of work from home and public transit use in census block groups in 2021 separately, using bivariate symbology to visualize which locations have >5% of either or both. Five percent was used as the critical value since it is near the average usage, thus representing above- and below-average use of these sustainable transportation methods. Change in transit and work from home summarizes the distribution areas where the absolute change in percentage using these transportation modes from 2011 to 2021 was 5% or greater.

2.4. Linear Regression and Hierarchical Models

General linear regressions and hierarchical models with fixed and random effects were tested to see if there were differences across cities in predicting transit and work-from-home percentages as well as transit retention and work-from-home percentage change over time. We also tested the models for spatial autocorrelation of the residuals and found that spatial regressions were not necessary. Since linear regressions at the city level could be biased by the small sample sizes, we used hierarchical models to test for city effects as suggested in the literature [48]. The small sample size could have biased our results, but there was no available spatial data at the individual level for these cities.

The data from the different sources (GTFS from transit agencies, US Census, and US EPA) were collated in an ESRI ArcGIS Pro 3.1 environment using ArcPy 3.1, a Python package developed by ESRI. The statistical data analysis was conducted using R. Packages used in R included lme4, dplyr, and psych.

3. Results

Table 1. City-wide descriptive statistics, 2021, with percent change from 2011 in parentheses. Data from Social Explorer, 5 year ACS (2017–2021, 2007–2011).

	All Massachusetts	Brockton	Fall River	Lawrence	Lowell	New Bedford	Worcester
Population	6,991,852 (7%)	104,216 (11%)	93,339 (5%)	87,798 (15%)	114,804 (8%)	100,309 (6%)	203,867 (13%)
Black Percentage	7.3 (18%)	41 (27%)	7 (155%)	5 (68%)	9 (61%)	6 (−24%)	13 (36%)
White Percentage	75 (−2%)	34 (−26%)	75 (−12%)	38 (36%)	58 (5%)	60 (−19%)	65 (−5%)
Hispanic Percentage	12 (42%)	12 (16%)	12 (81%)	82 (31%)	18 (18%)	23 (62%)	24 (40%)
Median Income (inflation-adjusted)	$89,026 (35%)	$68,067 (37%)	$49,613 (43%)	$39,583 (32%)	$64,489 (25%)	$50,581 (35%)	$56,746 (16%)
Renter Percentage	38 (11%)	43 (8%)	64 (9%)	71 (18%)	57 (19%)	61 (16%)	58 (21%)
Renter-Occupied Gross Rent ≥50%Percentage	23 (6%)	23 (−17%)	22 (−2%)	28 (−5%)	23 (21%)	24 (−2%)	24 (14%)
Transit-to-Work Percentage	9 (5%)	7 (16%)	2 (47%)	6 (21%)	3 (18%)	3 (41%)	4 (42%)
Car-to-Work Percentage	73 (2%)	85 (14%)	89 (6%)	85 (27%)	84 (8%)	88% (10%)	80% (9%)
Drove-Alone-to-Work Percentage	66 (2%)	74 (12%)	75 (2%)	69 (43%)	74 (14%)	74 (10%)	68 (6%)
Work from Home Percentage	12 (217%)	5 (219%)	5 (272%)	3 (409%)	7 (275%)	4 (134%)	8 (172%)
Transit System	NA	BAT	SRTA	MVRTA	LRTA	SRTA	WRTA
Average commute to work (minutes)	30 (7%)	32 (10%)	25 (14%)	22 (−4%)	27 (8%)	26 (8%)	25 (9%)

BAT—Brockton Area Transit Authority; SRTA—Southeastern Regional Transit Authority; MVRTA—Merrimack Valley Regional Transit Authority; LRTA—Lowell Regional Transit Authority; WRTA—Worcester Regional Transit Authority.

shows the overall statistics by city compared to the state of Massachusetts. These cities’ median incomes range from about $40 K to $68 K, which is lower than the state-wide median of $89 K. All have had an increased median income from 2011 to 2021, and Worcester had the lowest median income increase of 16%, while Fall River’s median income increased the most, by 43%. Statewide, the median income increased by 35%. Four of the cities—Fall River, Lowell, New Bedford, and Worcester—are majority White, as is the state of Massachusetts as a whole. Brockton has the highest percentage of Black residents (41%), while the other five cities range from 5% to 13%, compared to the statewide 7%. All cities (and the state of Massachusetts) had increases in the percentage of Black residents from 2011–2021 except for New Bedford. Lawrence has the highest proportion of Hispanic residents (82%), while the five other cities range from 12–24%, and the state has 12%. All six cities, as well as the whole state, saw an increase in the percentage of Hispanic residents from 2011–2021. Five of the cities—Fall River, Lawrence, Lowell, New Bedford, and Worcester—are renter-majority cities, ranging from 58–71% renters. Massachusetts has 38% renters, and the only owner-majority city, Brockton, has 43% renters. Amongst these renters, the proportion of renters with severe rent burden is similar across the six cities and the state, ranging from 22–28% rent burdened. All six gateway cities had a higher percentage of people who drove to work—80–89%—than the statewide percentage of 73%. All six gateway cities also had a lower percentage of people working from home than statewide—3–8% compared to the 12% statewide average—but all increased two-fold to five-fold between 2011 and 2021. Two cities, Fall River and New Bedford, participate in the same regional transit system; the other cities each have different regional transit systems. Figure 1 shows the locations of these gateway cities. Two of the cities are farther north of Boston, near New Hampshire (Lawrence and Lowell), and two are farther south of Boston, on the coast and closer to Rhode Island (Fall River and New Bedford). Two are termini for heavy rail to Boston, and two are en route; the two nearest to the Providence commuter rail are not connected to Boston by rail.

Table 2. (a) Descriptive statistics for dependent and explanatory variables. (b) Descriptive statistics for change and retention variables.

(a)
Variable	Mean 2011	Mean 2021	St. Dev. 2011	St. Dev. 2021	Median 2011	Median 2021
Socioeconomic
Black percentage	4.2	4.2	10	10.5	0	0
White percentage	95.8	95.8	10	10.5	100	100
Hispanic percentage	19	25.5	23.7	25.8	9.4	17.3
Poverty percentage	16.5	14.3	16.5	14.7	12.2	10.2
Median Income	48,910	55,978	23,343	32,977	46,447	54,181
Housing
Renter percentage	50.4	57.4	28.2	28.1	53	62.1
Renter-Occupied Gross Rent ≥ 50% percentage	26.1	22.7	20.6	17.7	23.7	20.4
Built environment
EPA Walk Index	13.7	14.5	3.1	2.7	14.3	14.7
EPA Transit Index	0.01	0.01	0.02	0.02	0	0
Bus Frequency per hour	2.4	4.3	1.8	4.4	2	2
Rail Frequency per hour	0.6	0.2	1.2	0.3	0	0
Number of Bus Lines	21.2	40.7	32.2	30.2	4	34
Number of Rail Lines	0.6	0.6	2.2	1.7	0	0
Transportation
Zero-Veh. household percentage	16	17.3	14.7	15.4	12.2	13.4
Transit-to-work Percentage	4	4.3	6.2	5.7	1.7	2.3
Car-to-work Percentage	88.3	83.5	11	13.1	91.2	85.9
Drove-alone-to-work Percentage	75.4	71.1	15	15	78.6	73.5
Work-from-home percentage	2.1	5.8	3.4	7.1	0	3.3
Commute over 60 min	4.6	9.9	5.5	10.7	3.1	7
N	607	531	607	531	607	531
(b)
	Mean Change		St. Dev Change		Median Change
Transit percentage change (2021–2011)	0.1		6.9		0
Transit Retention percentage (2011/2021)	85.4		219.4		0
Car percentage change (2021–2011)	−4.3		13		−3.8
Car Retention percentage (2011/2021)	143.2		125.9		115.1
Work-from-home change (2021–2011)	3.7		7.2		1.9
N	607

a shows the summary statistics for the socioeconomic and built environment variables for all the cities. Table 2b includes summary statistics for transit use and work-from-home changes and retention for all six cities by census block group. From Table 2a, it is evident that the sample percentage of Black residents remained stable over the study time period as did the White percentage, though the Hispanic percentage grew from a median percentage of about 9% to 17%. There was also a higher variance in the Hispanic percentage across the sample than that of the White or Black percentage of residents within a block group. The median income rose over the period, and the percentage of households living in poverty fell in the census block groups in these cities. The standard deviation of the median income also increased, suggesting a greater range of incomes in the block groups since 2011, unlike the standard deviation for poverty, which fell over this time period in these block groups on average. The percentage of renters increased from about 50% (53% median) in the census block group to 57% (62% median). The average percentage of rent-burdened households fell slightly, while the percentage of zero-vehicle households appeared to stay the same in the census block groups over the 2011–2021 comparison. The average bus-use frequency increased, though the median frequency remained the same in both years at about two buses per hour. The standard deviation of the bus frequency increased, suggesting more variance across the block groups. The number of bus lines increased significantly from 2011 to 2021 based on both the mean and median values, though the number of rail lines remained the same. Much of this increase was driven by bus route increases in New Bedford and Fall River. Lowell and Brockton saw small decreases in average bus frequency, though they are still higher than Lawrence, which has the lowest bus frequency in both years.

Table 2b suggests that both the work-from-home percentage and the transit-to-work percentage increased, while the drive-to-work percentage fell. The transit retention percentage was positive on average for the census block groups in our sample, suggesting perhaps that transit ridership, in general, was maintained or increased within census blocks rather than transit usage decreasing in some and increasing in others. The median value for the percentage of those taking transit to work was quite low but increased from 1.7% to 2.3%. In contrast, the median value for public transit supply, including rail and bus frequency, was unchanged across the years. The median percentage of workers who worked at home was 0% in 2011, and this rose to 3% in 2021. All the cities show a widening in the box plots in 2021 when compared to 2011 (Figure 2), suggesting that the range of work-from-home percentages increased across the census block groups. Brockton and Lawrence had the highest transit use in 2011 (6.7% and 6%), while Lawrence had almost caught up by 2021 (6.5% for both cities). Fall River had the lowest transit use at 1.6% in 2011, which rose to 1.9% in 2021. All the other cities increased, though Worcester and Lawrence showed the most increase over this time period. All the cities saw work-from-home percentages almost double or even triple, with Brockton (5%), Lowell (7%), and Worcester (8%) having the highest average percentage of worker at home in 2021.

To understand whether the census block groups across the cities were very different from each other, random intercepts and slope models were estimated for all outcomes using the same dependent variables as the general linear regressions. However, the multilevel models were not a significant improvement on the estimates from the general linear regression, except for the 2011 transit percentage. The variance partition estimates from the multilevel models showed that only 5–10% of the variance in transit ridership percentage was across the cities, and 90–95% of the variance was within the city block groups. This suggests that most of the variation in transit use happened within the cities but not across the cities themselves. Block groups with similar socioeconomic and built environment characteristics had similar transit use, regardless of which city they were in. Spatial regressions were also not required based on Lagrange multiplier tests.

The results of the random intercepts (for 2011 transit) and general linear regression models (for all the other models) are shown in

Table 3. (a). Regression models predicting transit and work-from-home percentages at the census block group level, comparing 2011 and 2021, with and without built environment and transportation variables. (b) Regression models predicting transit retention and work-from-home changes in a census block group, with and without built environment and transportation variables.

(a)
	Dependent Variable:				Dependent Variable:
	Transit Percentage				Work-from-Home Percentage
	2011		2021		2011		2021
	Sociodemographic only	With built environment and transportation variables	Sociodemographic only	With built environment and transportation variables	Sociodemographic only	With built environment and transportation variables	Sociodemographic only	With built environment and transportation variables
Black Percentage	0.003	0.001	−0.04	−0.01	0.01	0.01	−0.03	0.01
Hispanic Percentage	0.04 ***	0.04 ***	0.04 ***	0.03 ***	−0.01	−0.01	−0.03 **	−0.01
Renter Percentage	0.04 ***	0.04 ***	0.03 ***	0.03 ***	−0.01 **	−0.01 **	−0.03 **	−0.01 **
≥50% rent burden	0.02 **	0.02 **	0.005	0.003	0.003	0.004	−0.005	0.004
Walk Index		−0.1		0.20 **		0.02		0.02
Transit Index		−37.91		−26.42		9.91		9.91
Bus frequency average		0.39 ***		−0.13		−0.08		−0.08
Rail frequency average		0.02		3.56 ***		−0.11		−0.11
Constant	0.29	0.22	1.54 ***	−1.26	2.87 ***	2.80 ***	8.42 ***	2.80 ***
Observations	607	607	531	531	607	607	531	607
Log Likelihood	−1934.4	−1928.7	−1652.4	−1635.8	−1603.4	−1601.8	−1787.6	−1601.8
Akaike Inf. Crit.	3882.8	3879.5	3314.9	3289.6	3216.7	3221.6	3585.2	3221.6
(b)
Variables	Dependent variable: Transit Retention				Dependent variable: Work-from-Home change
	Sociodemographic only		With built environment & transportation variables		Sociodemographic only		With built environment and transportation variables
Black Percentage	−0.68		−0.38		−0.05 *		−0.04
Hispanic Percentage	0.17		0.26		−0.02 *		−0.03 **
Renter Percentage	0.29		0.4		−0.02		0.003
≥50% Rent Burden Percentage	−0.06		0.12		−0.03 *		−0.03 *
Walk Index			−2.95				−0.21 *
Transit Index			1056.16 ***				4.57
Bus frequency average			−1.21				−0.18 **
Rail frequency average			22.05 ***				−0.37
Constant	51.88 ***		66.82 **		5.45 ***		8.30 ***
Observations	607		607		607		607
Log Likelihood	−3920		−3905		−2051		−2045
Akaike Inf. Crit.	7853		7830		4115		4110

Note: * p < 0.1; ** p < 0.05; *** p < 0.01.

a,b. The model results suggest that Hispanic, renter, and housing burden percentages were significant in predicting transit use in census block groups in 2011 as well as 2021. A unit increase in the Hispanic percentage in a tract led to a 0.04% increase in transit use on average. When built environment and transit variables were included, these variables were still significant in 2011, though rent burden was no longer significant in 2021. Bus and rail frequency were both significant in 2011, though in 2021, the bus frequency was no longer significant. Once built environment and transit variables were accounted for, renter percentage was the only significant variable in predicting the work-from-home percentage in a block group. The coefficient suggested that the higher the percentage of renters, the lower the work-from-home percentage in a census block group.

Table 3b suggests that the only significant predictors of transit retention in a census block group are the transit index and rail frequency. Both were positive, suggesting that better accessibility to jobs and higher rail frequency lead to higher transit retention in a block group. None of the socioeconomic variables were significant. On the other hand, in predicting work-from-home change, the Hispanic percentage and rent burden percentage were significant and negative, suggesting that block groups with higher Hispanic and rent burdened percentages were less likely to work from home. The average bus frequency was also negative, suggesting that locations with better bus frequency had lower work-from-home percentages.

Figure 3 and Figure 4 show how these relationships manifest in space. Figure 3 illustrates the percentage of each block group that used public transit (a) and worked from home (b). While the preceding analysis found some predictors based on characteristics of the block group and rail frequency, we do not observe any significant pattern in the distribution of locations of public transit use or working from home. Indeed, in Figure 4, where we illustrate the locations of jointly high work from home and transit use, we see that most block groups are either higher in work from home (5% or more) or higher in transit use (5% or more). In 2021, these follow a pattern, with higher public transportation use being more central and working from home being more peripheral. We also investigated areas of change and found that few areas had increases in both work from home and public transportation use, and the changes, in general, were aligned with the locations of higher work from home and public transit use from 2021, suggesting that the illustrated change is the magnitude of use change (which we saw across all cities, see Table 1 and Figure 2), not a change in the locations of public transit users or from-home workers.

4. Discussion

We asked what local characteristics were associated with transit ridership change and work-from-home change between 2011 and 2021 in six Gateway Cities in Massachusetts. Considering demographic, built environment, and affordability factors, we found that ridership was more likely to increase in block groups with higher accessibility to jobs and more frequent transportation to those jobs. Work from home was more likely to increase in block groups with lower Hispanic percentage, lower rent burden percentage, and lower bus frequency. Not surprisingly, public transit to work and work from home also seem somewhat spatially segmented—block groups with higher public transit use were usually not the block groups with higher work-from-home percentages.

Our analysis also focused on cities as well as block groups, and the differences in findings between cities as a whole and block groups within these six Gateway cities, in general, point to the importance of the local characteristics in predicting public transit use rather than the city itself. We note that the random intercept analysis did not generate any significant predictors of working from home or public transit use, meaning that within each city, built environment and demographic factors were not significant. However, across all of these six cities, a clearer pattern emerges. Future research should investigate the robustness of these patterns by including a larger set of cities. Additionally, future research should consider the roles of other factors in informing sustainable transportation use, including traffic congestion and commuting patterns.

We also note that our focus on block groups across cities and within cities raised the variance of types of block groups within each city. As demonstrated in Figure 2, all six cities had wide ranges of transit use, transit change, and work-from-home change. Summary statistics of commuting type change at the city level versus at the block group level also illustrate that increases in transit use are spatially segmented—while there are increases in both work from home and public transit use, these trends at the city-wide level mask the limited areas where these changes occurred. Further, Figure 3 and Figure 4 demonstrate this spatial heterogeneity within each city. As expected, based on previous literature, these increases in sustainable commuting approaches are not widespread and are likely to be related to specific socioeconomic and occupational classes. Increasing overall public transit usage may require converting non-public transit users into public transit users in areas and amongst demographics where transit usage is not already prevalent.

While overall driving to work is high across the study area, the few places that had more than 5% of workers working from home and more than 5% of workers taking public transit to work were scattered across the cities. The same is true of locations that had above-average increases of 5% or more in both public transit use and work from home. This suggests that neighborhoods may be changing in different ways. Yu et al. [49] suggest that the implications for retaining bus ridership are dependent on the city itself and that rather than targeting driving, it would be more fruitful for local policymakers to target the built environment. Studies suggest that making transit stops safer to access via traffic-calming strategies could be a sustainable strategy [50,51]. Our study also suggests the importance of some built environment factors. Further research should investigate the demographic, labor, and built environment changes in public transit retention block groups to understand how this sustained use could be maintained; research into the areas with both work-from-home and public transit increases also warrants further understanding of how to sustain or grow these lower-emission commuting practices.

We were surprised to find that ridership had not declined in these cities, and between 2011 and 2021, the percentage of people taking public transportation to work increased in each city and within block groups. While we have previously discussed the complexity of what actual travel actions are included in the category “public transit to work”, and indeed, people may be taking rideshares to work or taking the heavy rail, we were curious about what each city reported as their local bus ridership.

Two of the regional transit authorities in these gateway cities recently implemented fare-free ridership. The Merrimack Valley Regional Transit Authority (Lawrence) had ridership in 2021 at 66% of pre-pandemic levels (995 K riders) and had declined from highs of 2.3 M riders in 2016 to 1.5 M riders in 2019. While they were functionally fare-free due to low-contact COVID-19 operating procedures, in 2022, outside of our study timeframe, most services went fare-free [52]. News reports indicate that since introducing fare-free service, fixed-route and paratransit use now exceeds pre-pandemic levels [53] The Worcester Regional Transit Authority (Worcester) also instituted fare-free service in 2020, and by 2022, had exceeded pre-pandemic ridership levels and is anticipating riders exceeding its previous ridership peak in 2016 [54]. As noted in our literature review, fare-free transit is a potential way to increase ridership in smaller cities.

Other regional transit authorities’ reports indicate declining ridership, some seeking further understanding of the factors impacting decline. In a 2019 study of rideshare (TNC) use and Brockton’s public transit usage decline, researchers from the Metropolitan Area Planning Council did not conclude whether TNC use was impacting transit ridership, citing many other factors that could also indicate a decline in transit use. This report found a 3% decline in Brockton’s daily fixed ridership in 2015–2018 and pointed to similar declines in other regional bus services in Massachusetts. Further, this demonstrates pre-COVID-19 concerns about replacement by TNC [55]. This aligns with the broader literature on the factors leading to ridership decline before COVID-19. These locally measured reports of system ridership show a different trend than the travel-to-work measures in the American Community Survey data we used. While these might indicate that non-regional transit use may be increasing or that there is a conceptual disconnect between the measure of daily ridership and people who say that they use public transit to get to work, this suggests future research to better understand the actual sustainable transit practices of individuals.

We did find that the work-from-home percentage increased and the percentage of workers using cars decreased, though most census block groups retained the percentage of workers on average. This suggests that the workers in these cities are not changing their travel behavior drastically, though the time period does include COVID-19. The growth in the work-from-home percentage suggests that some of the workers in these cities are able to substitute their car or transit commutes. The model results show that those block groups that have seen changes in work-from-home percentages tend to have lower Hispanic and rent-burdened percentages. On the other hand, locations that had higher Hispanic and Renters tended to have significantly higher transit use both in 2011 and 2021. This suggests perhaps that workers using transit and those working at home may be quite different in terms of their socioeconomic characteristics.

The role of the built environment in predicting transit use was evident in both 2011 and 2021, suggesting that transit use works in conjunction with pedestrian friendliness and regularity of transit. It was also a significant predictor of transit retention. The role of transit in connecting workers to jobs was the most prominent factor in predicting retention. It is worth noting that it was not bus frequency but rail frequency that was significant. This suggests that even in these six cities, access to jobs via rail was important for workers. Conversely, it is worth noting that none of the built environment variables were significant in predicting work-from-home percentages in both years. However, bus frequency was negatively correlated with work-from-home change, suggesting that census block groups with higher bus frequency locations were less likely to work from home. It appears that these two types of workers (those who take transit and those who work from home) may perhaps live in very different census block groups within these cities. Future research should further investigate these apparently different cohorts of transit users and workers from home in small- to mid-sized cities to understand their different transportation practices and needs.

Finally, we did not find a significant relationship between our measure of housing affordability public transit use, and thus, we do not have an answer to whether or how public transportation users in these cities have location-efficient housing. We did find that places with increased work from home were also places with lower housing cost burden, suggesting that people able to work from home are also more likely to be less financially overextended in these cities, and these locations were also less frequently served by buses. Further research should investigate public transit users specifically to more clearly understand whether they are living in affordable housing and whether their public transit use offsets some of their potential transportation costs.