Impact of COVID-19 on Public Transit Accessibility and Ridership
Michael Wilbur, Afiya Ayman, Anna Ouyang, Vincent Poon, Riyan Kabir, Abhiram Vadali, Philip Pugliese, Daniel Freudberg, Aron Laszka, Abhishek Dubey
IImpact of COVID-19 on Public TransitAccessibility and Ridership
Michael Wilbur ∗ , Afiya Ayman † , Anna Ouyang ∗ , Vincent Poon † , Riyan Kabir ∗ , Abhiram Vadali ∗ ,Philip Pugliese ‡ , Daniel Freudberg § , Aron Laszka † , Abhishek Dubey ∗∗ Vanderbilt University , † University of Houston , ‡ Chattanooga Area Regional Transportation Authority , ‡ Nashville Metropolitan Transit Authority
AbstractPublic transit is central to cultivating equitable commu-nities. Meanwhile, the novel coronavirus disease COVID-19and associated social restrictions has radically transformedridership behavior in urban areas. Perhaps the most concern-ing aspect of the COVID-19 pandemic is that low-incomeand historically marginalized groups are not only the mostsusceptible to economic shifts but are also most reliant onpublic transportation. As revenue decreases, transit agenciesare tasked with providing adequate public transportation ser-vices in an increasingly hostile economic environment. Transitagencies therefore have two primary concerns. First, how hasCOVID-19 impacted ridership and what is the new post-COVID normal? Second, how has ridership varied spatio-temporally and between socio-economic groups? In this workwe provide a data-driven analysis of COVID-19’s affect onpublic transit operations and identify temporal variation inridership change. We then combine spatial distributions ofridership decline with local economic data to identify variationbetween socio-economic groups. We find that in Nashville andChattanooga, TN, fixed-line bus ridership dropped by 66.9%and 65.1% from 2019 baselines before stabilizing at 48.4%and 42.8% declines respectively. The largest declines wereduring morning and evening commute time. Additionally, therewas a significant difference in ridership decline between thehighest-income areas and lowest-income areas (77% vs 58%)in Nashville.
Keywords : COVID-19, ridership, socio-economics, spatio-temporal I. IntroductionThe novel coronavirus COVID-19 and the associated socialrestrictions have radically transformed travel behavior in urbanareas throughout the world. While COVID-19 has transformednormal operations in almost all industries, the social distancingmeasures and precautions associated with this virus have hadparticularly devastating effects on public transit. For instance,since the World Health Organization (WHO) declared COVID-19 a pandemic on March 11, 2020 [1] subway ridership in NewYork City has dropped by upwards of 91% [2]. This disruptionhas created pressing operational challenges for public transitagencies. Foremost, agencies must determine how to continue pro-viding adequate service while navigating a rapidly chang-ing environment with reduced resources. This involves firstquantifying the affects of the pandemic to date. However,the decentralized nature of government policies and recom-mendations in the United States makes it challenging toidentify global solutions for local transit agencies. Thereforelocal agencies must identify solutions tailored to their uniquecircumstances. Additionally, the local outlook is dynamic asregulations are lifted or restricted over time. Therefore carefuldata-driven modeling and analysis is required to stay up-to-date on local ridership behavior as well as the financial effectsgoing forward.Perhaps the most concerning aspect of the COVID-19 pan-demic is that low-income and historically marginalized groupsare not only the most likely to be affected financially by thecrisis but are also the least likely to own their own cars [3].Therefore most rely on the public transit system to get to work,school or access child services. Additionally, as most peopleworking in grocery stores, logistics and cleaning have beenlabeled “essential services,” many people from low-incomegroups do not have the luxury of working remotely from home.As resources are limited by drastic drops in ridership, agenciesmust take care in identifying trips to be cut so as to not hurtthose most reliant on local transit services.This work is primarily concerned with two questions. First,to what degree has the COVID-19 pandemic and associatedstate restrictions affected ridership and operations of fixed-line public transit? We focus on Nashville and Chattanooga,TN. We present total lost riders over time compared to 2019baselines and provide a spatio-temporal analysis of ridershipdecline throughout both cities. Secondly, are there disparitiesin ridership changes across socio-economic groups? For thiswe provide a spatial analysis of ridership patterns throughoutthe COVID-19 pandemic and correlate our findings with pub-licly available economic data to draw conclusions regardingchanges in behavior between socio-economic groups.II. Contributions and Key FindingsThe primary contributions of this work are as follows:1) We provide a summary of ridership changes due toCOVID-19 in Nashville and Chattanooga, TN. We findthat ridership dropped by up to 66.9% and 65.1% in a r X i v : . [ phy s i c s . s o c - ph ] A ug ashville and Chattanooga respectively by late Aprilbefore starting a moderate recovery.2) We performed a temporal investigation of ridership preand post-COVID-19 and find that an out-sized propor-tion of changes in ridership occur during weekdaysduring morning and evening rush, indicating that Stay atHome orders and remote work options are a significantfactor in ridership declines.3) Our spatial analysis indicates that change in ridershipvaries greatly between census tracts and neighborhoods.By incorporating economic data at the census tract levelwe found that ridership declined up to 19% more inhigh-income neighborhoods than in the lowest incomeparts of Nashville.The remainder of this article is as follows. First, we sum-marize recent literature regarding the impact of COVID-19 onpublic transit systems and socio-economic transportation stud-ies in Section III. Then we describe the data and processingmethods in Section IV. In Section V, we outline our analysismethods and results. We address possible limitations of thiswork in Section VI and finally we summarize our findings anddiscuss future work in Section VII.III. Related WorkIn this section we cover literature related to COVID-19 inthe context of transportation systems and the interaction ofsocio-economics on transit usage. A. COVID-19 and transportation
Fixed-line bus and rail public transit inherently involvesmoving passengers in an enclosed space. One of the majorreasons there has been such significant declines in publictransit ridership is the fear of COVID-19. In public healthfields, the study of infectious disease transmission throughpublic transit and air travel is well studied [4], [5], [6], [7].While there is a growing number of publications regardingthe spread of COVID-19 by air travel [8], there is a lackof information on how this applies to public transit [9].Regardless of transmission rates on public transit ridership hasdeclined significantly as we show in this work.Given how fast COVID-19 has transformed life in urbanareas the limited amount of work related to virus in thecontext of public transit ridership mostly consists of pre-printpublications and government public releases. The ConnectedCities With Smart Transportation (C2SMART) group at NewYork University has released monthly whitepapers related tothe impacts of COVID-19 on New York City and Seattle. Theyfind that in New York City, average subway and commuter railridership is down 80% while bus ridership is down 50% in thefirst week of July, 2020 with a peak subway ridership declineof 94% in late March [2, 10, 11, 12]. There are similar findingsin European cities [13].There has been some recent work investigating mode shiftaway from public transit. While modeling lasting effects of thepandemic is in its early stages, in some high transit cities evenmoderate shifts from public transit to personal vehicles can increase travel times by 5 to 10 minutes on average for oneway trips [14]. On the other hand, in New York City the bikesharing program CitiBike has been more resilient to loss inridership than the subway system and there is some evidenceof transit users shifting to the shared bike programs [15].
B. Socio-economics in transportation
Previous research indicates different transit behaviors amongsocioeconomic classes. When it comes to public transit, low-income and historically marginalized groups are particularlyreliant on public transportation. Minorities and low-incomehouseholds account for 63% of transit riders in the UnitedStates [16]. Additionally, low-income groups are more likelyto ride buses while high income individuals are more likelyto utilize rail systems [17]. According to a 2017 publicationfrom the American Public Transportation Association, 30%of bus riders have a household income of less than $15,000,while 12% of bus riders have a household income of $100,000or more. Among rail riders, only 13% have household in-comes below $15,000, while 29% have household incomesof $100,000 or more [18]. In terms of public versus privatetransit, a study conducted in Hawaii reported key differencesbetween bus riders and solo drivers. The mean householdincome of a bus rider was 16% lower than that of a solodriver [19]. Bus riders also, on average, owned fewer cars perhousehold (1.7 cars) compared to solo drivers (2.3 cars) [19].A major reason low-income groups are heavily reliant onpublic transportation is their likelihood of owning a personalvehicle. According to an analysis of 2012 California House-hold Travel Survey data, 78% of households without a car donot have a car as a result of economic or physical barriers [3].Together, these studies suggest that individuals of a lower so-cioeconomic background may be disproportionately impactedby changes in public transit availability. It is important to notethat these trends are not unique to the United States; a casestudy conducted in France found that low income individualscomprised a larger portion of public transit ridership than highincome individuals [20].IV. Data Collection and ProcessingIn this section we outline the datasets used in this workwhich consist of ridership boarding information, economicdata per census tract and COVID-19 cases per day as wellas data processing and filtering.
A. Ridership data
Boarding count data was provided by the MetropolitanGovernment of Nashville and Davidson Count for the fixed-line bus systems of Nashville from January 1, 2019 to July1, 2020. Boarding information was also acquired from theChattanooga Area Regional Transportation Agency (CARTA)between January 1, 2020 to July 1, 2020. The ridership datawas derived from farebox units on all passenger vehiclesservicing trips within these time ranges. Farebox included arecord of reach passenger boarding event. It also includeddriver information, shift changes and when vehicles switch
ABLE I : Boarding counts before and after processingand number of census tracts for Nashville and Chattanoogadatasets. Metric Nashville ChattanoogaRaw Boardings(2020 YTD) 2,800,000 464,570Processed Boardings(2020 YTD) 2,800,000 445,987
B. Economic data and COVID-19 new case counts
Economic data was retrieved from the United States CensusBureau [22] and ProximityOne [23]. For each 2010 censustract these sources provided a breakdown of racial demograph-ics, income levels and housing information. New COVID-19cases per day for Nashville and Chattanooga were retrievedfrom The New York Times COVID-19 Dashboard [24] be-tween January 1, 2020 and July 1, 2020.
C. Mapping boarding events to census tracts
To incorporate the census tract level economic data, eachboarding event was mapped to the corresponding census tract where that boarding occurred. As each census tract includeda geometric polygon representing the tract this was a simplespatial join. One limitation of working with aggregated 2019data for Chattanooga was that we could not get baselineridership information at the census tract level. For Nashvillebaseline 2019 ridership at the census level was available.V. Analysis and ResultsIn this section we outline the main analysis and resultsfor this work. We start by giving a high level overviewof COVID-19 restrictions and the corresponding operationalchanges implemented by the transit agencies in Nashvilleand Chattanooga before moving into our analysis of ridershipdeclines in both cities.
A. COVID-19 restrictions and operational changes
Nashville and Chattanooga both receive guidance regardingCOVID-19 related restrictions directly from the State of Ten-nessee and also are available to impose their own regulationsin excess of the state’s recommendations. On March 5th thefirst COVID-19 case was identified in Tennessee and on March8th the first COVDI-19 case was found in Nashville. TheState of Tennessee ordered a State of Emergency regarding thepandemic on March 12, 2020 and a Safer at Home order onMarch 30, 2020 which mandated residents of the state stay intheir homes other than for "essential activities". The TennesseeSafer at Home order ended on April 30 [25].Nashville regulations were more swift. Nashville imposedtheir own Stay at Home order on March 22 which was notlifted until Phase 1 reopening began on May 11 which includedallowing gatherings of up to 10 people while most businesseswere allowed to open at 50% capacity. On May 25 Nashvillemoved to Phase 2 which allowed gatherings of up to 25people and most businesses could operate at 75% capacity[26]. Nashville moved to a Phase 3 opening on June 20, 2020which included provisions a limited opening for small venues(up to 250 people) however reverted back to a Phase 2 openingon July 3, 2020.Both Nashville and Chattanooga reduced the total numberof trips in reaction to COVID-19. Unique trip identifiers werenot available in either dataset. Therefore to tally the number oftrips serviced per week we grouped the data by date, uniquedriver ID, unique vehicle ID, route and direction. Chattanoogamoved to a reduced bus schedule in the middle of Aprilwhile Nashville switched to a reduced schedule on March 29,2020. Prior to the schedule change, Chattanooga serviced 781weekly fixed-line bus trips. During the week of April 19thChattanooga switched all weekdays to their Saturday schedule.Between April 19th and July 1st, Chattanooga serviced 373trips per week. Nashville switched to a reduced scheduleduring the week of April 5. Prior to switching Nashvilleserviced 1954 weekly trips which was reduced to 1035 weeklytrips. As we see in Section V-B, the most significant dropsin ridership occurred well before either city moved to theirrespective reduced schedules. a n 1 s t F e b 1 s t M a r s t A p r s t M a y 1 s t J un 1 s t J u l s t , · . · · R i d e r s h i p , , N e w C a s e s (a) Nashville J a n 1 s t F e b 1 s t M a r s t A p r s t M a y 1 s t J un 1 s t J u l s t , , , R i d e r s h i p N e w C a s e s (b) Chattanooga J a n 1 s t F e b 1 s t M a r s t A p r s t M a y 1 s t J un 1 s t J u l s t − − − − C h a ng e i n R i d e r s h i p NashvilleChattanooga (c) Change in ridership compared to 2019
FIGURE 1 : Weekly ridership compared to new COVID-19 cases per week for (a) Nashville and (b) Chattanooga. NewCOVID-19 cases are at the county level. Nashville is in Davidson County, Chattanooga is in Hamilton County. Phases 1, 2and 3 in (a) are per Nashville’s reopening plan provided by Nashville Metro. (c): Change in ridership compared to last yearfor Chattanooga and Nashville, TN from January through June 2020. Change in ridership was calculated by comparing weeklyridership to the baseline ridership from the same month in 2019.
B. Impact of COVID-19 on city-wide ridership
The fundamental question in this section is to what degreehas COVID-19 changed ridership and what effects do thesechanges have on transit operations. Figure 1a and Figure 1bshow weekly total ridership and weekly new COVID-19 casesin Nashville and Chattanooga respectively. Figure 1c showsdrop in ridership for Nashville and Chattanooga compared toa baseline. The baseline was calculated by taking the averageweekly ridership for the corresponding month in 2019 for bothcities.As shown in Figure 1a, Nashville public transit ridershipstarted to decline on the week of March 1st which correspondswith the first known COVID-19 case in Tennessee on March5th and the Tennessee State of Emergency Order on March 12. Perhaps more importantly there was a major tornado inNashville on March 3rd [27] which helps explain the initialdecline in ridership at this time. Ridership remained constantfor a week before a significant decline started during the weekof March 22nd when the Nashville Safer at Home Order wasput into effect on March 22, 2020. Nashville ultimately reacheda low of 60,620 riders on the week of April 19, a 66.9%reduction from the average ridership in April of 2019 as shownin Figure 1c. From late April to July 1st ridership stabilized.By the week of June 28th ridership in Nashville has recovered22.7% from the low in April. Chattanooga’s steep declinestarted the week of March 5th before hitting a low also on theweek of April 19 with a low of 8,077 weekly riders as shownin Figure 1b. Compared to the 2019 baseline, Chattanooga anuary February March April May June · R i d e r s h i p Route 22 Route 50 Route 52Route 55 Route 56 (a)
January February March April May June , , , R i d e r s h i p Route 1 Route 4 Route 9Route 10 Route 14 (b)
FIGURE 2 : Average weekly ridership per month for the 5most popular routes in (a) Nashville and (b) Chattanooga in2020. Ridership trends follow similar patterns to the demandpatterns across all routes in Figure 1.had a 65.1% in ridership on April 19 Figure 1c. UltimatelyChattanooga ridership recovered to 11,725 riders the week ofJun 28th, an increase of 45.2% from the low of April 19-25.Ultimately, both cities saw a rapid decline in ridership fromearly March to late April before ridership stabilized throughthe end of June. To characterize these findings, we refer to Jan-uary through February as pre-COVID operations and startingin late April a new normal post-COVID operations stabilizedat approximately 60% reduction in ridership compared to theprevious year for both cities.
C. Route level investigation
Figure 2a and Figure 2b show the monthly ridershipdistribution on the top 5 routes for the city of Nashvilleand Chattanooga respectively. We see similar trends to theaggregated ridership analysis in the previous section. Forboth cities ridership decreased rapidly before stabilizing inApril. In Nashville however we see a greater rebound betweenApril to June than in Chattanooga. The rebound in Nashvillecorresponds loosely with Phase 2 reopening.An important note is that for Chattanooga route 14 is oneof the most used routes however it is unique in that it is a freeshuttle service to the University of Tennessee, Chattanooga.When Universities went online in March route 14 initiallycontinued operating on its regular Saturday schedule. Due tothe drastic demand reduction during this time Chattanoogaultimately stopped the service entirely in April.From this section we see that the most populated routesfollow a similar trajectory and magnitude of ridership drop as M ond a y T u e s d a y W e dn e s d a y T hu r s d a y F r i d a y S a t u r d a y S und a y · R i d e r s h i p Jan-FebMay-June (a) M ond a y T u e s d a y W e dn e s d a y T hu r s d a y F r i d a y S a t u r d a y S und a y , , , , R i d e r s h i p Jan-FebMay-June (b)
FIGURE 3 : Daily average ridership for January–Februaryand May–June 2020 for (a) Nashville and (b) Chattanooga.January–February represents baseline pre-COVID ridershiplevels in 2020 while May–June represents ridership after itstabilized post-COVID. , , , R i d e r s h i p Jan-FebMay-Jun (a)
Time of Day (Hour) R i d e r s h i p Jan-FebMay-June (b)
FIGURE 4 : Average ridership per hour of day for January–February and May–June 2020 for (a) Nashville and (b) Chat-tanooga. January–February represents baseline pre-COVIDridership levels in 2020 while May–June represents ridershipafter it stabilized post-COVID.the fixed-line transit system overall. Therefore a more detailedtemporal and spatial analysis is outlined in the followingsections of this paper.
D. Temporal analysis of transit usage and rider behavior
Here we investigate temporal changes in ridership betweenpre-COVID and post-COVID operations. As discussed in Sec-ion V-B, for both cities normal operations spanned from Jan-uary 1st to the end of February and after a rapid drop ridershipstabilized in mid-to-late April. Therefore in this Section weuse January-February to represent pre-COVID operations andMay-June to represent post-COVID operations. In Figure 3aand Figure 3b, we see the ridership distribution of Nashvilleand Chattanooga for each day of the week before COVID-19and after COVID-19. In both cities the drop in ridership on theweekends is less than weekdays with Chattanooga only seeinga 20% decrease in ridership on Saturdays and a 32% decreaseon Sundays compared to an average of 56% on weekdays.Nashville saw a 41% decrease in ridership on Saturdays anda 47% decrease on Sundays compared to an average of 57%decrease for weekdays.Figure 4a and Figure 4b show ridership in January-Februarycompared to May-June per hour of the day. We can see thatthe biggest drops in ridership occur during morning rush andevening rush. This is highlighted in Nashville where morningrush (5:00AM-9:00AM) saw a 64% change in ridership andevening rush (3:00PM-6:00PM) saw a 62% decrease comparedto 42% change between 9:00AM and 3:00PM. This discrep-ancy was not as pronounced with Chattanooga where there wasa 62% and 56% decrease in ridership for morning and eveningrush respectively compared to a 53% between 9:00AM and3:00PM.As we can see in this section, the biggest declines inridership were on weekdays during morning and evening com-muting times. The declines continued after the Nashville andTennessee Stay at Home orders expired showing a persistentshift towards alternative work options throughout the COVID-19 pandemic. This phenomena was however more apparent inNashville than Chattanooga.
E. Spatial analysis of transit usage and rider behavior
In this section we look at spatial variation in ridership. Eachboarding was mapped to a corresponding 2010 census tractin which that boarding occurred. Figure 5 shows the percentdecrease in ridership between pre-COVID (January-February)and post-COVID (May-June) operations per census tract. Asshown, change in ridership was not uniformly spaced through-out either city. Both cities see significant decreases downtown,most likely due to workers working remotely. This was mostvisible in Chattanooga where ridership decreased by up to81%. Chattanooga also saw a significant decrease in ridershipin the census tract that contains the University of Tennessee,Chattanooga reflecting the University’s decision to suspend in-person operations and CARTA’s subsequent cancellation of thefree shuttle servicing this region. While the same patterns arepresent in Nashville, change in ridership was more uniformlikely due to the density of Nashville’s downtown region.Nashville saw significant decrease in ridership to areas heavilydependent on retail and shopping which includes a 87% dropto Opry Mills and a 86% drop to Green Hills.
TABLE II : Pearson Correlation values for change in ridershipafter COVID-19 in Nashville Tennessee. A positive correlationrefers to as the metric increases, the more severe the drop inridership post-COVID.Metric PearsonCorrelationMedian Income 0.21Median Housing Value 0.35Median Rent 0.15% White 0.01% African American -0.02% Hispanic -0.19
F. Socio-economic analysis
As we have seen, ridership varies both temporally andspatially in both cities. In this section we investigate thecorrelation between decreases in ridership and socio-economicfactors. As a proxy for wealth we use median householdincome per census tract. Figure 6 shows change in weeklyridership for 2020 compared to baseline ridership in 2019for the 10% highest income and 10% lowest income censustracts in Nashville. We see a significantly greater decreasein ridership for the high income group which saw a 77%decrease in ridership in the week of April 27th. Meanwhilethe low-income group saw only a 58% decrease in the weekof April 27th, also a low. The trend lines follow a similartrajectory for both groups, no significant time shift was found.Additionally during post-COVID operations both groups sawsimilar upward trends in ridership.Table II provides Pearson correlation statistics for vari-ous income metrics and racial demographics for ridershipdrop post-COVID. The moderate positive correlation betweenmedian income and housing value reiterates the effects ofsocio-economic status on ridership post-COVID. In Table IIa positive correlation refers to the case in which as themetric increases the more severe the drop in ridership. Thehighest positive correlation with drop in ridership was withmedian housing value, i.e. census tracts with high medianhousing costs had the greatest reduction in ridership from2019 baseline. There was also a moderate negative correlationbetween the percent of the population that was Hispanic andchange in ridership. There was no correlation between thepercentage of the census tract that was African American orWhite.There are two likely reasons that low-income areas see lessof a decrease in public transit usage. First, low-income familiesare less likely to own a car [28]. Second, workers in grocerystores, sanitation and cleaning, and logistics are often labeled“essential” workers and still required travelling to their placeof work. Additionally, as local resources and jobs are oftenlimited in low-income areas [29] travel by public transit is anecessary component of life for these groups, regardless ofCOVID-19.
FIGURE 5 : Change in ridership between pre-COVID (January–February) and post-COVID (May–June) 2020 per census tractfor (left) Nashville and (right) Chattanooga. J a n F e b M a r A p r M a y J un J u l − − − − − Low-incomeHigh-income
FIGURE 6
Transport Policy , 60:152–159, 2017.[4] Annie Browne, Sacha St-Onge Ahmad, Charles R Beck,and Jonathan S Nguyen-Van-Tam. The roles of trans-portation and transportation hubs in the propagation ofinfluenza and coronaviruses: a systematic review.
Journalof travel medicine , 23(1):tav002, 2016.[5] Jason R Andrews, Carl Morrow, and Robin Wood. Mod-eling the role of public transportation in sustaining tuber-culosis transmission in south africa.
American journal ofepidemiology , 177(6):556–561, 2013.[6] Fang Zhao, Thomas Gustafson, et al. Transportationneeds of disadvantaged populations: where, when, andhow? Technical report, United States. Federal TransitAdministration, 2013.[7] Andras Bota, L Gardner, and Alireza Khani. Modelingthe spread of infection in public transit networks: Adecision-support tool for outbreak planning and control.In
Transportation research board 96th annual meeting ,2017.[8] Matteo Chinazzi, Jessica T Davis, Marco Ajelli, CorradoGioannini, Maria Litvinova, Stefano Merler, Ana Pastorey Piontti, Kunpeng Mu, Luca Rossi, Kaiyuan Sun, et al.The effect of travel restrictions on the spread of the2019 novel coronavirus (covid-19) outbreak.
Science
Sustainability , 12(9):3870, 2020.[14] Yue Hu, William Barbour, Samitha Samaranayake, andDan Work. Impacts of COVID-19 mode shift on roadtraffic. arXiv preprint arXiv:2005.01610 , 2020.[15] João Filipe Teixeira and Miguel Lopes. The link be-tween bike sharing and subway use during the covid-19 pandemic: The case-study of new york’s citi bike.
Transportation Research Interdisciplinary Perspectives
Bulletin of thepsychonomic society , 27(1):64–66, 1989.[20] K Mohamed, Etienne Côme, Johanna Baro, and LatifaOukhellou. Understanding passenger patterns in publictransit through smart card and socioeconomic data.
Ur-bComp,(Seattle, WA, USA)
Transportation , 44(3):495–510,2017.[29] Evelyn Blumenberg and Michael Manville. Beyond thespatial mismatch: Welfare recipients and transportationpolicy.