Equity Implications of Ride-Hail Travel during COVID-19 in California

Abstract

COVID-19 has shocked every system in the U.S., including transportation. In the first months of the pandemic, driving and transit use fell far below normal levels. Yet people still need to travel for essential purposes like medical appointments, buying groceries, and—for those who cannot work from home—to work. For some, the pandemic may exacerbate extant travel challenges as transit agencies reduce service hours and frequency. As travelers reevaluate modal options, it remains unclear how one mode—ride-hailing—fits into the transportation landscape during COVID-19. In particular, how does the number of ride-hail trips vary across neighborhood characteristics before versus during the pandemic? And how do patterns of essential trips pre-pandemic compare with those during COVID-19? To answer these questions, we analyzed aggregated Uber trip data before and during the first two months of the COVID-19 pandemic across four regions in California. We find that during these first months, ride-hail trips fell at levels commensurate with transit (82%), while trips serving identified essential destinations fell by less (62%). Changes in ride-hail use were unevenly distributed across neighborhoods, with higher-income areas and those with more transit commuters and higher shares of zero-car households showing steeper declines in the number of trips made during the pandemic. Conversely, neighborhoods with more older (aged 45+) residents, and a greater proportion of Black, Hispanic/Latinx, and Asian residents still appear to rely more on ride-hail during the pandemic compared with other neighborhoods. These findings further underscore the need for cities to invest in robust and redundant transportation systems to create a resilient mobility network.

Keywords

data and data science urban transportation data and information systems ridehailing data planning and analysis effects of information and communication technologies (ICT) on travel choices shared mobility public transportation innovative public transportation services and technologies transportation network companies (TNC)

COVID-19 has had profound effects on travel and transportation in the United States. In the first months of the pandemic, driving dropped to 60% below what would be expected without a pandemic, and transit ridership plummeted between 70% and 90% nationwide ( 1 – 3 ). Yet people still need to travel for essential purposes like medical appointments, buying groceries, caring for family, and—for those who cannot work from home—to work. Households without cars have long faced more limited access to essentials such as jobs and healthy foods compared to households with cars ( 4 , 5 ). The pandemic may exacerbate preexisting travel challenges as transit agencies across the country modified or cancelled services ( 6 ). With more limited available options alongside heightened public health concerns, travelers without cars may seek other modes to reach essential destinations. One possible mode is ride-hail (also known as transportation network companies). How travelers use ride-hail during COVID-19, however, and the potential equity implications of that use, remain unknown.

This research explores the potential equity implications of ride-hail trip-making during COVID-19 using a combination of aggregated Uber and census data. We compare Uber trip volumes eight weeks before and eight weeks during the early months of the COVID-19 pandemic in neighborhoods across four regions in California: Bakersfield, Fresno, Los Angeles, and San Francisco. Following this introduction we organize this paper as follows: first, we review the existing literature on ride-hail travel and traveler responses to COVID-19. We then present the data and methods employed in this study. Next, we present results for how both all and essential trips changed—both in absolute numbers and by neighborhood characteristics—during the first months of the COVID-19 pandemic. We conclude with a discussion of results and implications for transportation policy and planning.

Literature Review

Ride-Hail Travel

Neighborhood analyses typically examine ride-hailing in relation to the local built environment and rider characteristics. Survey research consistently finds that ride-hail users are younger and own fewer cars compared to the general population ( 7 – 9 ). Research is more mixed about ride-hailing by income, finding that a greater share of ride-hail users earn higher incomes ( 7 – 10 ), but that ride-hail travelers earning lower incomes and living in low-income neighborhoods use ride-hail as much or more often compared to higher-income travelers and neighborhoods ( 8 , 9 , 11 , 12 ). People use ride-hail for a variety of trip purposes. Studies based on survey data show that people ride-hail disproportionately for entertainment, social, and recreational purposes compared with other modes of transportation ( 10 , 13 ), but that a not insubstantial share also use ride-hail to make essential and household-serving trips. Henao ( 14 ), for example, finds that one-quarter of users surveyed in the Denver area use ride-hail for shopping and errands. Others report lower (4% to 9%) shares of ride-hail travel for grocery, shopping, and errand purposes ( 10 , 13 ). Household-serving travel remains challenging to quantify, however, in part due to the breadth of trip types that may fall under this umbrella; for example, Young and Farber ( 10 ) examine trips to “Facilitate passengers” and “Other” purposes, which could include trips to meet essential household needs, such as transporting a young or elderly relative to a doctor’s appointment.

As with individual- and neighborhood-level socioeconomic characteristics, built environment factors are associated with ride-hail travel, albeit more weakly ( 11 ). Research finds that ride-hail trips are associated with transit density and quality (11, 15–18). Ride-hail trips are also positively associated with population and job density ( 11 , 12 , 15 ) as well as a greater land use diversity ( 15 ). Research also finds that a high number of ride-hail trips occur between highly walkable areas ( 19 ), as well as in high-density and congested neighborhoods ( 20 ).

Travel Responses to COVID-19

Travel plummeted during the first months of COVID-19 as businesses shuttered and firms moved to remote work. As cities and states urged residents to stay at home, driving fell by 60% compared to what would have been expected without a pandemic, and transit ridership dropped between 70% and 90% nationally ( 1 – 3 ). Drops in transit ridership varied by mode, with the largest drops in commuter rail services and the least in intracity bus ridership ( 3 ). Puentes ( 3 ) argues that varied decreases in transit ridership reflect income differences across riders: bus riders have a lower income compared to rail commuters, suggesting both that they have fewer alternative travel options and are often essential workers who must be physically present at their workplaces. Even as transit remained critical to the transport of essential—and often low-income—workers, transit agencies across the county cut service in response to COVID-19 ( 21 – 24 ). In Los Angeles, for example, LA Metro cut bus service by 29% and rail service by 14%, reducing weekday service frequency and hours in line with prepandemic Sunday schedules ( 22 ). In San Francisco, the San Francisco Municipal Transport Agency anticipates that pandemic-spurred service cuts—that have the agency running at about 70% of pre-pandemic service hours--will endure for months ( 24 ). Some cities have responded by offering free or discounted trips on shared mobility services like bikeshare and e-scooters (see e.g., DDOT [25]). Other travelers have responded by purchasing a “pandemic car.” Early pandemic car sales show upticks among young urban adults even as overall car sales were down; while these urban young adults may have previously viewed a car as “more expense than it’s worth,” they may now see cars as the “ultimate personal protective equipment” for safe travel during the pandemic ( 26 ). But what choices exist for travelers who still need to travel but cannot afford to purchase a car or face restricted mobility options?

This paper considers one possible alternative—ride-hailing—and asks two related questions about ride-hail travel during COVID-19: first, how do neighborhood-level patterns of Uber trips compare pre-pandemic and during the first months of the COVID-19 pandemic? And second, how does essential trip-making both in aggregate and across neighborhoods compare before and during the pandemic? We hypothesize that ride-hail trips fell less in neighborhoods where fewer people can transition to remote work and where personal car access is limited. Specifically, we hypothesize that census tracts with lower car ownership, lower household median incomes, and higher transit ridership will experience smaller drops in ride-hail travel. We also hypothesize that a greater share of trips serve essential destinations during compared to before the pandemic, both because people limited personal travel and because many non-essential destinations (e.g., bars and restaurants) closed during the first months of the pandemic.

Data and Methods

Data

To examine how ride-hail travel changed during the COVID-19 pandemic, we analyzed Uber data aggregated at the census tract level before and during the first two months of the pandemic, as well as the change in trip volumes between the two periods. We define the periods based on California’s statewide stay-at-home order, issued on March 19, 2020 ( 27 ). In sum, we include 16 total weeks of data: eight weeks before the stay-at-home order, from January 4 to February 28, 2020, and eight weeks after, from April 4 to May 29, 2020. These periods were chosen to avoid transitional effects (people beginning to avoid travel immediately before the stay-at-home order, and adjusting their travel and work patterns immediately afterwards), while minimizing the effect of long-term trends.

We analyze these 16 weeks of data across four geographic regions in California: Bakersfield, Fresno, Los Angeles, and San Francisco. Table A1 in the Appendix provides the background socioeconomic and transportation characteristics of each region. We define each region based on its Uber market boundary, and include all census tracts with non-zero populations that had at least 10 Uber trip origins and/or destinations between January and February (see Figure 1). Our final sample includes 4,445 tracts across the four regions; the tracts included are home to 98.1% of the population in the selected regions ( 28 ). Compared to tracts included in the analysis, excluded tracts (n = 91/4,546, 2.0%) have lower population and job densities, a lower median household incomes, smaller shares of transit commuters, and lower shares of Asian or Black residents.

Figure 1.

Study area.

To create a measure of ride-hail travel across space, we aggregated Uber trips based on trip origin census tract. The data set includes completed trips across all Uber ride-hail products (e.g., UberX, UberXL, Uber Black) with the exception of trips to and from airports. We exclude airport-serving trips because (a) they are numerous (see for example Lavieri and Bhat [13]) and could bias results in census tracts near airports, (b) trips to/from airports primarily reflect passengers rather than employee commutes based on reported mode splits of airport employees ( 2 ), (c) airport census tracts contain zero population and, therefore, lack resident socioeconomic characteristics, and (d) we aim to measure the effects of COVID-19 on day-to-day household travel rather than its effect on air travel, which is almost certainly the primary purpose for airport-serving ride-hail trips. We note that UberPool, the service that matches riders traveling along similar routes, was suspended on March 17, 2020; therefore, UberPool trips appear in the January–February data for the San Francisco and Los Angeles markets but not in the April–May data. Table 1 shows the variables obtained from Uber for each census tract, aggregated over eight weeks before and eight weeks during the pandemic. In addition to total trip origins per census tract, we counted how many trips were matched to a specific location (rather than just an address) using Foursquare, a location data platform that connects addresses with specific destinations, and assigns categories based on the destination type ( 29 ); for more information about the user interface between Uber and Foursquare, see Crook ( 30 ). In total, 44.4% of trips across the two time periods were associated with a Foursquare location at the origin or destination. We then coded Foursquare-matched trips as serving essential destinations if they fell into one of the following six categories: groceries (including supermarkets, convenience, warehouse, and discount stores, but excluding liquor stores); hardware stores; healthcare and medical (including hospitals, emergency rooms, doctor and dentist offices, and medical labs); pharmacies; government buildings and services (including municipal buildings, post offices and shippers, courthouses, police and fire stations, and embassies); and financial services (including banks, credit unions, and check cashing services).

Table 1.

Analysis Variables According to Census Tract

Variable	Year/time period	Source
Number of trip origins	January–February 2020	Uber
Number of trip origins	April–May 2020	Uber
Number of trip origins with an origin or destination matched to any Foursquare location¹	January–February 2020	Uber Foursquare
	April–May 2020	Uber Foursquare
Number of trip origins with an origin or destination matched to an essential destination on Foursquare²	January–February 2020	Uber Foursquare
	April–May 2020	Uber Foursquare
Number of trip origins aggregated by day of week and hour of day³	January–February 2020	Uber
	April–May 2020	Uber
Low- to mid-wage (<$1,250/month) job density (jobs/square mile)⁴	2017	LEHD, 2017
High-wage (>$3,334) job density (jobs/square mile)	2017
Population density (people/square mile)	2014–2018	Five-year American Community Survey
Number of adults aged 20–44
Number of adults, Asian
Number of adults, Black
Number of adults, Latinx/Hispanic
Number of commuters, ride transit
Number of households owning zero cars

Note: LEHD = Longitudinal Employer–Household Dynamics Survey.

Foursquare is a location data platform that connects addresses with specific destinations. Foursquare destinations are integrated into the Uber app and some travelers identify their origin/destination by place name rather than solely address.

Essential destinations matched using Foursquare include trips to/from the following destination types: groceries, healthcare, government, financial services, pharmacies, and hardware stores (see text for more details).

For example, the number of trips during January and February 2020 taken on Mondays from 4:00 to 4:59 pm.

We combined the low- and mid-wage categories because a worker earning minimum wage in California ($12/hour) and working full-time (40 h/week) would earn $1,920/month assuming four work weeks in a month. The low- and mid-wage categories, therefore, include all minimum wage and part-time positions.

We contextualized Uber travel using a suite of built environment and sociodemographic data from the 2014 to 2018 five-year American Community Survey (ACS) and the 2017 Longitudinal Employer–Household Dynamics (LEHD) survey (see Table 1). We selected these variables based on factors found to influence travel behavior (and ride-hail travel specifically) from the literature.

Methods

This project focuses on changes in ride-hail travel during the first months of the COVID-19 pandemic. We first calculated the changes in number of Uber trips to explore the changes in ride-hail trips overall, across regions, and by census tracts between January–February and April–May 2020. We distinguished between “all” trips and “identified essential” trips, and calculated the share of identified essential trips in two ways: first, as a proportion of all trips; and second, as a proportion of trips with an identified Foursquare origin or destination. The former likely underestimates the share of essential trips within our sample because trips can serve an essential destination without being matched to a Foursquare location; the latter method likely overestimates the number of trips serving essential destinations. We discuss both possibilities further in the “Limitations” section.

To determine which neighborhood characteristics are most associated with changes in trip-making, we specified three models and selected the best-fit models using the Akaike Information Criterion/Bayesian Information Criterion. First, we specified a linear regression model to examine associations between changes in Uber travel—measured as the ratio of trips made during April and May to the number of trips made pre-pandemic in January and February—and independent neighborhood characteristics from the LEHD and ACS, listed in Table 1. Second, using the same suite of independent variables, we modeled the share of trips before COVID-19 (January and February) that served identified essential destinations, to determine which neighborhood characteristics were associated with travel to these destinations pre-pandemic. As a robustness check, we tried an alternative specification with identified essential trips as a proportion of Foursquare-matched trips; model coefficients and statistical significance are robust to whether the denominator of the dependent variable is all trips or Foursquare-matched trips. Finally, we modeled the ratio of identified essential trips during April and May to those identified essential pre-pandemic in January and February to assess which neighborhood characteristics were most strongly associated with changes in essential trip-making. We tested city fixed effects in each model but excluded them from final model specifications because they did not contribute meaningfully to model goodness of fit, nor did they change model coefficients or statistical significance.

Preregistration

After completing the research design, but before data analysis, we preregistered the research hypotheses, lists of variables, and analysis methods with the Open Science Framework (osf.io/qbc96).

Limitations

The data used in this study are limited in three primary ways. First, rider information is not included, so we cannot determine the degree to which the same cohort of riders was using Uber before the pandemic and during it. In addition, the people who use Uber in a neighborhood may not reflect the demographic characteristics of the neighborhood as a whole, and some are likely not residents of that neighborhood. Second, we code trips as “identified essential” if its origin or destination matched to a Foursquare location that falls into the above-enumerated “essential” categories. This represents a best estimate of the share of trips serving essential destinations, but is almost certainly an undercount. Essential trips are not identified if travelers entered only an address rather than a suggested location within the Uber app. For example, if two riders were both traveling in separate Ubers to a Safeway grocery store and one entered the address “1209 Main Street” and one entered “Safeway,” only the latter trip would be matched as an essential trip within our data, even though both traveled to the same essential destination. Also, because users often request trips from their current location rather than entering it into the Uber app, some trips originating at essential places may be matched to a nearby address rather than the rider’s true origin. We therefore calculated a second metric: identified essential destinations as a share of trips with matched Foursquare locations. Although this latter metric likely overestimates the share of trips to the enumerated essential destinations—Foursquare locations are disproportionately commercial and the measure may, therefore, reflect essential trips as a proportion of largely commercial trips—it also helps to account for the changing number of total trips with matched Foursquare locations, which fell from 47% in January and February to 31% in April and May 2020. Second, the list of essential categories is limited to an unambiguous subset of essential goods and services that people regularly access, to provide a relatively “clean” proxy for relative changes in essential travel before and during the pandemic. Other categories of essential travel (trips between non-business locations, for example, to care for family members, or workers commuting to warehouses, factories, construction, and food service jobs, and the like) are therefore not captured by this metric. Finally, each market area comprises one or more counties, and substantial variation likely exists between individual municipalities and neighborhoods within each market.

Findings

Figure 2 shows the decrease in Uber trips between January–February and April–May 2020; Figure 3 shows these same changes across space. Across the four regions, the number of all trips dropped by 82% and the number of identified-essential trips fell by 62%. Pandemic effects varied across regions, neighborhood characteristics, and trip type (all versus identified essential). Fewer trips during the first months of the COVID-19 pandemic do not likely stem from a reduced service quality; service quality remained markedly consistent during January and February and during April and May, with both the share of unfulfilled requests, and wait times remaining similar over time.

Figure 2.

Percentage change in all trips and those identified essential.

Figure 3.

Percentage change in number of Uber rides, January–February to April–May 2021.

Ride-Hail Travel before and during COVID-19

Between January-February and April-May 2020, ride-hail travel dropped precipitously across the four regions; Uber trips fell by 82% from pre-pandemic levels by April–May, similar to the large drops reported in local transit ridership. For example, transit demand was down 85% to 91% in the San Francisco Bay Area between April and May 2020 compared to demand expected without a pandemic ( 31 ). Trip decreases varied according to region and neighborhood characteristics.

Table 2 shows that trip reductions were less pronounced in the less populated, less dense, and lower-income Bakersfield and Fresno regions relative to either Los Angeles or San Francisco. Pre-pandemic, a higher share of trips—and number of trips per capita—began in high-income compared to low- and mid-income neighborhoods. By April-May, the trend reversed, with twice as many trips per capita beginning in low- compared with high-income neighborhoods (0.63 and 0.31 trips per person, respectively). The reversal stems from steeper drops in trip-making to/from high-income neighborhoods (–89%), compared to neighborhoods with a low median household incomes (–73%).

Table 2.

Uber Travel According to Region and Neighborhood Characteristics, before and during the Pandemic

	Number of total trips		Total trips per capita¹
	January–February 2020 (%)	April–May 2020 (%)	January–February 2020	April–May 2020	% Change
Overall	100.0	100.0	2.50	0.45	−82
Bakersfield	0.7	1.2	0.44	0.14	−68
Fresno	0.8	1.2	0.27	0.07	−72
Los Angeles	55.3	65.4	2.86	0.61	−79
San Francisco	43.2	32.2	2.67	0.36	−87
Median neighborhood income^
Low (≤$50,431)	23.2	34.2	2.37	0.63	−73
Middle ($50,431–$103,833)	48.6	48.9	2.39	0.43	−82
High (>$103,833)	28.2	16.9	2.84	0.31	−89
Proportion of households without a car^
Low (<2.8%)	12.7	12.5	1.21	0.21	−83
Middle (2.8%–11.3%)	40.2	44.9	1.97	0.40	−80
High (>11.3%)	47.1	42.7	5.17	0.84	−84
Population density^
Low (<4,785 people/square mile)	14.9	14.3	1.37	0.24	−83
Middle (4,785–15,077 people/square mile)	44.1	44.8	2.23	0.41	−82
High (>15,077 people/square mile)	41.0	40.9	4.35	0.78	−82

^Low defined as the bottom 25%, middle as the middle 50%, and high the top 25% of census tracts across all regions.

per resident. Trips per capita were calculated by summing the total number of people and trips that occurred in each neighborhood group and then dividing total trips by total population.The shaded portion indicates the “Total trips per capita”, and the non-shaded portion indicates “number of total trips”.

Changes were more uniform across neighborhoods based on household car ownership and population density: in neighborhoods with varied levels of car ownership, the number of Uber trips fell between 80% and 84%. Both before and during the pandemic, Uber trip origins were about four times higher per capita in neighborhoods with higher shares of zero-car households (>11.3%) compared with neighborhoods with relatively few zero-car households (<2.8%). Uber trips originated disproportionately in medium- to high-density neighborhoods (about 85% of trips from 75% of census tracts), although drops in trip-making in the first months of the pandemic were relatively uniform across tract population densities (about 82%). Table A2 in the Appendix further breaks down these results by individual region.

Table 3 shows the associations between neighborhood characteristics and the ratio of Uber trips made during compared to before COVID-19 related shutdowns (dependent variable = number of trips April–May / number of trips January–February), determined from a linear regression model. Positive coefficients indicate a higher ratio of trips made during relative to pre-pandemic, meaning that trips in April and May were closer to their pre-pandemic levels. Negative coefficients show the reverse: greater reductions in trips compared with pre-pandemic levels. Model results show that higher shares of Asian, Black, and Hispanic/Latinx residents in a tract were positively associated with trips during the first two months of the pandemic compared with pre-pandemic levels; in other words, the number of trips fell by less in these neighborhoods, controlling for other neighborhood characteristics. Neighborhoods with higher shares of transit commuters, zero-car households, and younger adults (aged 20–44), and greater low- to mid-wage job density showed steeper declines in Uber trips during the pandemic compared with other areas. Higher median household incomes were also associated with fewer trips during the pandemic compared with pre-pandemic levels, all else equal. By contrast, residents in densely populated areas undertook more trips during the pandemic compared with pre-pandemic levels. Tracts with light or heavy rail stations (n = 237 across Los Angeles and San Francisco) also had greater declines in trips during the first months of the pandemic; this squares with many rail stations being concentrated in downtown business districts, where many workplaces shifted to remote work.

Table 3.

Association between Neighborhood Characteristics and the Ratio of Uber Trips Made during the Pandemic to those Made before It

	Coefficient	Sig.
Median household income ($1,000s)	−0.0006 (–11.01)	^***
% transit commuters	−0.0008 (–3.35)	^***
% zero-car households	−0.0008 (–3.25)	^***
% adults aged 20–44	−0.0016 (–9.43)	^***
Race/ethnicity
% Black	0.0037 (25.01)	^***
% Asian	0.0007 (6.70)	^***
% Hispanic/Latinx	0.0029 (35.12)	^***
Population density (1,000/square mile)	0.0010 (5.17)	^***
Job density (1,000/square mile)
Low- to mid-wage jobs¹	−0.0029 (–5.67)	^***
High-wage jobs	0.0002 (0.79)	NS
Rail station (yes)²	−0.0257 (–3.62)	^***
Constant	0.2429 (21.10)	^***
n (tracts)	4,455	na
R²/Adj R²	513/0.511	na

Note: Sig. = significance; NS = not significant; LEHD = Longitudinal Employer–Household Dynamics Survey. t-statistics in parentheses. na indicates not applicable.

p < 0.05, ^**p < 0.01, ^***p < 0.001.

Wage categories defined by LEHD; low-wage are jobs paying <$1,250/month, mid-wage are jobs paying $1251–$3,333/month, and high-wage are jobs paying >$3,334/month.

Indicates that a heavy or light rail station is located in the census tract.

Ride-Hail Travel Serving Essential Destinations

Before the pandemic, identified essential trips accounted for 5.6% of all trips and 12% of trips with a matched Foursquare location (see Table 4). Echoing patterns observed for all trips (see Table 2), more identified essential trips occurred per capita in high-density tracts and neighborhoods with higher shares of zero-car households compared to lower-density neighborhoods and tracts with lower shares of zero-car households. Unlike with all trips, about the same number of identified essential trips per capita (trips per residents) began in low- rather compared to high-income neighborhoods before the pandemic (0.14 and 0.13 trips per resident, respectively); identified essential trips also accounted for a higher proportion of all trips in low- compared to high-income neighborhoods (6.1% and 4.7%, respectively).

Table 4.

Number of Identified Essential Trips According to Neighborhood Characteristics, before and during the Pandemic

	Number of trips serving identified essential destinations per capita¹			Proportion of all trips serving identified essential destinations²		Proportion of matched Foursquare trips serving identified essential destinations³
	January–February 2020	April–May 2020	% Change	January–February 2020 (%)	April–May 2020 (%)	January–February 2020 (%)	April–May 2020 (%)
Overall	0.14	0.05	−62	5.6	12.0	12.1	39.4
Bakersfield	0.02	0.01	−51	5.2	8.2	16.6	43.8
Fresno	0.02	0.001	−55	6.2	9.9	13.7	35.7
Los Angeles	0.17	0.07	−58	5.9	11.7	13.0	40.0
San Francisco	0.14	0.05	−67	5.3	13.0	10.7	38.4
Median neighborhood income^
Low (≤$50,431)	0.14	0.07	−52	6.1	11.0	14.1	40.1
Middle ($50,431–$103,833)	0.14	0.06	−61	5.9	12.9	12.5	40.2
High (>$103,833)	0.13	0.04	−73	4.7	11.8	9.7	36.1
Proportion of households without a car^
Low (<2.8%)	0.07	0.03	−61	5.7	12.6	12.5	40.0
Middle (2.8%–11.3%)	0.13	0.05	−60	6.4	12.9	13.9	41.0
High (>11.3%)	0.26	0.09	−64	4.9	11.1	10.3	37.5
Population density^
Low (<4,785 people/square mile)	0.08	0.03	−63	5.8	12.5	12.1	38.1
Middle (4,785–15,077 people/square mile)	0.14	0.05	−61	6.3	13.5	13.3	41.7
High (>15,077 people/square mile)	0.21	0.08	−62	4.9	10.4	10.6	37.1

^Low defined as the bottom 25%, middle as the middle 50%, and high the top 25% of census tracts across all regions.

per resident.

Identified essential destinations (groceries, healthcare, government, financial services, pharmacies, and hardware stores) as a proportion of total trips.

Identified essential destinations as a proportion of trips matched to Foursquare locations.

As with all trips, the number of identified essential trips, regardless of how we measured them, fell in April and May relative to January and February. Yet the number of identified essential trips decreased by less than trips overall. While all trips decreased by 82% between January–February and April–May, identified essential trips fell by 62% in total. Drops in identified essential trips ranged across regions from between –51% in Bakersfield to –67% in San Francisco. As a result, the share of all trips identified essential more than doubled during the first months of the pandemic (12.1%) compared with before it (5.6%). As a share of matched Foursquare location trips, essential trips rose even more dramatically, from 12% of trips January–February to 39% of trips April–May.

Altered temporal patterns before compared with during the COVID-19 shutdowns may reflect the changed composition of trip purposes during the pandemic. Figure 4 shows the temporal distribution of trips across each study period. Trips in January and February show expected weekday morning and afternoon peaking, along with an increased number of trips during Friday and Saturday evenings, reflecting social and recreational travel. In April and May the weekday rush hour and weekend evening peaks are erased or muted. Instead, a far greater share of trips in April and May occur on weekdays in the early afternoons, when people would typically have been at work, and may reflect travel during times that essential businesses are open.

Figure 4.

Temporal trip patterns before and during the pandemic.

Table 5 shows the results for two identified essential trips models. The first shows associations between tract characteristics and the share of Foursquare-matched trips that were identified as serving an essential destination before the pandemic. Results show that neighborhoods with higher incomes, higher population densities, and a greater share of zero-car households are associated with a lower share of essential trips. By contrast, neighborhoods with a higher share of Black, Asian, and Hispanic/Latinx residents are associated with greater shares of trips identified as serving essential purpose, all else equal.

Table 5.

Association between Neighborhood Characteristics and Identified Essential Uber Trips before the Pandemic and Ratio of Essential Uber Trips Made during the Pandemic to those Made before It

	(1) Number of identified essential trips, prepandemic¹		(2) Ratio of essential trips made during the pandemic to those made before it
	Coefficient	Sig.	Coefficient	Sig.
Median household income ($1,000s)	−0.0003 (–6.46)	^***	−0.0003 (–2.52)	^**
% transit commuters	−0.0003 (–1.58)	NS	−0.0014 (–2.88)	^***
% zero-car households	−0.0007 (–3.08)	^***	0.0000 (0.04)	NS
% adults aged 20–44	−0.0004 (–2.42)	^**	−0.0016 (–4.47)	^**
Race/ethnicity
% Black	0.0014 (9.99)	^***	0.0024 (7.83)	^***
% Asian	0.0008 (9.13)	^***	0.0016 (8.05)	^***
% Hispanic/Latinx	0.0011 (14.15)	^***	0.0027 (15.58)	^***
Population density (1,000/square mile)	−0.0005 (–2.94)	^***	−0.0015 (–3.68)	^***
Job density (1,000/square mile)
Low- mid-wage jobs	0.0001 (0.27)	NS	−0.0007 (–0.70)	NS
High-wage jobs	−0.0002 (–0.95)	NS	−0.0003 (–0.52)	NS
Rail station (yes)²	−0.0247 (–3.80)	^***	−0.0261 (–1.83)	^*
Constant	0.1603 (15.19)	^***	0.4116 (17.26)	^***
n	4455	^***	4455	na
R2/Adj R2	0.147/0.145		0.122/0.120	na

Note: Sig. = significance; NS = not significant. t-statistics in parentheses. na indicates not applicable.

p < 0.05, ^**p < 0.01,^***p < 0.001.

Number of trips with identified Foursquare locations.

Indicates that a heavy or light rail station is located in the census tract.

The second model presents the associations between neighborhood characteristics and the ratio of identified essential trips during the first months of the pandemic compared to those made before it (ratio = number of identified essential trips in April–May/number of identified essential trips in January–February). As with the ratios previously presented in Table 3, positive coefficients reflect that trips in April and May were closer to their pre-pandemic levels. Negative coefficients indicate greater reductions in trips compared to pre-pandemic levels, all else equal. Model results reveal that, controlling for other factors, higher shares of transit commuters and adults aged 20–44 in a tract are associated with fewer essential trips during the pandemic relative to pre-pandemic. Neighborhood income, population density, and high-wage job density were also negatively associated with the ratio of during to pre-pandemic trips. Conversely, neighborhoods with a higher number of Asian, Black, and Hispanic/Latinx populations all show positive associations, suggesting lower drops in essential trip-making during the pandemic relative to other neighborhoods.

Discussion

At the beginning of 2020, patterns of ride-hail use superficially squared with the prevailing narrative of who uses ride-hailing ( 7 – 10 ): more Uber trips were taken per capita in densely populated neighborhoods and in tracts with high shares of zero-car households and higher median household incomes. Model results coupled with destination category data (see Table 5, column 1), however, suggest a more nuanced reality: neighborhoods with a higher median household incomes took lower shares of trips to identified essential destinations, while neighborhoods with higher shares of Asian, Black, and Hispanic/Latinx residents relied on ride-hail relatively more for essential trips, even after controlling for income and other factors. Together, these findings suggest that differences between neighborhoods (and, presumably, the populations who live in them) significantly correlate with differences in ride-hail trip purposes. Findings reinforce research by Lavieri and Bhat ( 13 ), who report that higher-income travelers make a smaller share of ride-hail trips for errands, and non-Hispanic white travelers take a greater share of trips for recreational purposes compared to lower-income and other ethnic groups.

The emergence of COVID-19 wrought dramatic changes in travel and travel behavior globally, and ride-hail was no exception. Uber trips in the four California markets studied here fell by more than 80% overall, on a par with the decline seen in transit ridership in many cities. Intriguingly, however, the prevailing narrative of ride-hail users described above appeared to reverse during the first months of the pandemic: the highest-income neighborhoods showed the lowest number of trips per capita in April and May, while per capita trips were twice as high in low-income areas. Even when controlling for income, neighborhoods with more older (45+) residents, and higher shares of Black, Hispanic/Latinx, and Asian residents still appear to rely more on ride-hail during the pandemic, relatively speaking, than other areas. These shifts are particularly remarkable given that UberPool–which was available in San Francisco and Los Angeles pre-pandemic and is disproportionately used by travelers living in low-income and majority non-white neighborhoods ( 32 )–was not available in April and May, suggesting that many former UberPool users now rely on the relatively more expensive UberX service. On the other hand, tracts with higher shares of transit commuters and higher shares of zero-car households all showed steeper declines in trip-making during the pandemic relative to other neighborhoods.

At first blush, these findings may appear counterintuitive or contradictory; in particular, the last two findings run counter to our initial hypotheses that ride-hailing would see lower trip declines in tracts with less personal car access (ownership) and where people more reliant on transit may be seeking alternatives due to reduced transit service during the pandemic (see, for e.g., WMATA [21]), and/or early perceptions—albeit not widely supported by evidence—that transit posed a potential public health risk ( 33 ). Together with less trip erosion in neighborhoods with higher shares of Asian, Black, and Hispanic/Latinx populations, however, results may reflect existing transportation disparities and suggest distinctly different travel responses to COVID-19 by car-free households (who do not own a car because they choose not to) versus car-less households (who do not own a car because of economic, health, age, or other reasons).

Car-free households in California are disproportionately white, have higher incomes, and are more highly educated compared to car-less households. Car-free travelers, despite not owning a car, often gain car access in another way: carsharing. For example, among car-free Californians who do not own a car because of environmental beliefs, about 39% held a carshare membership as of 2012, compared to just 3.6% of all zero-car households. Nonetheless, these other zero-car households (many of whom do not own cars due to economic circumstances) still take nearly one-quarter (22%) of trips by car, suggesting they still require at least intermittent car access and gain it in other ways ( 34 ). Previous research shows that immigrants, for example, are far more likely to have established a carpool—both within and between households—compared to native-born travelers ( 35 ).

COVID-19 brings additional pressures to bear on both car-free and car-less households, as people may be wary of riding public transit, lending their car, or carpooling with others. Research clearly and repeatedly documents the high financial barriers to car ownership including purchase, maintenance, and insurance costs ( 36 ). Those who cannot afford to own a car, and who cannot meet all of their travel needs through other modes of transportation, may instead purchase car access one trip at a time, a trend seen for years in the bimodal income distribution of taxi use ( 37 ) and more recently documented in ride-hailing ( 11 ).

Car-less and car-free households likely diverge in other important ways related to pandemic ride-hail use: specifically, the division between those workers who must be physically present at work versus those who can work remotely, and online consumer behaviors. Non-white and below-median wage workers are more likely to work jobs that cannot transition to remote work ( 38 ). Recent transit ridership data supports this: buses have retained riders at higher rates compared to rail services ( 3 ), a likely reflection of income disparities across these modes ( 39 ). Where transit service is now sporadic or no longer available, some workers may have turned to ride-hail for some or all of their commuting. On the consumption front, high-income earners have also shifted to making more essential and non-essential purchases online during the pandemic ( 40 ). Together, shopping and work statistics suggest that lower-income and non-white travelers–a greater share of whom do not own vehicles ( 41 )–continue to need to travel to/from work and essential destinations at rates exceeding car-free households, who largely earn higher incomes, are able to transition to remote work, and tend to buy more goods online during the pandemic. Additional individual-level research is needed to further investigate these possible explanations.

In short, the larger decrease in Uber travel during the first months of the pandemic in well-resourced neighborhoods, where many residents may work from home, walk, or drive their own cars for essential trips, and who order goods online and, thereby, avoid most travel, suggests that people have done just that: avoided travel. Yet in underserved areas, ride-hailing may serve as a bulwark in our transportation system, transporting people with fewer options who still need to travel. These findings present critical implications for ride-hailing specifically, and transportation planning and policy more broadly. First, cities and ride-hail companies alike should consider how to incentivize pooled rather than solo ride-hail services, which are more efficient and cheaper compared with non-pooled services and are disproportionately used by riders living in low-income communities ( 32 ). Both Uber and Lyft suspended pooled services during the pandemic amid public health concerns. Although face masks are currently required in all ride-hail vehicles, the surest way to resume pooled services and increase people's confidence in hailing a shared ride in the short term is to end the pandemic. Post-pandemic, cities considering ride-hail partnerships to, for example, bridge first-last mile transit gaps should require these services be pooled (as many are already doing, see, for e.g., Brown, Manville, and Weber [42]). Employers seeking to encourage employees to adopt alternatives to solo driving could likewise subsidize pooled services to encourage greater vehicle occupancy. Finally, ride-hail companies should continue to improve pooled services based on known rider perceptions. Ride-hail specific adjustments may include providing additional fare discounts for each additional passenger pickup to increase both the financial incentives and reduce dissatisfaction with added pickups, capping the total number of pickups per trip to increase travel time reliability, and continuing to optimize algorithms to reduce the excessive delay that some experience during pooled rides ( 43 ).

Findings also raise questions of transportation resiliency and system redundancy. While ride-hailing may serve as a fallback for populations that face a reduced transit service or other shocks in an extended crisis like the COVID-19 pandemic, the cost of ride-hailing for all essential travel may pose a financial hardship or impossibility for households that cannot afford a car. As such, the pandemic has highlighted existing transportation inequities that cannot be solved by ride-hailing alone. To be prepared for, withstand, and quickly recover from shocks or disruptions, resilient transportation systems must both identify vulnerabilities and improve redundancy ( 44 ). This would yield immediate benefits of boosting mobility options in underserved or disenfranchised communities while simultaneously meeting goals of environmental sustainability and racial justice. By continuing to invest in non-car modes of transportation (including robust public transit, micromobility, walkable infrastructure, and bike lanes) with a focus on disenfranchised and marginalized communities, cities can substantially improve resiliency in the long term and ensure that these communities have a broader suite of mobility options available during shocks and natural disasters. Targeting transportation investments in marginalized communities would start to redress the decades-long lack of investment in many of these same neighborhoods, while also recognizing the foundational role many of these residents play in the urban and economic ecosystem. Targeted investments in transportation infrastructure should begin immediately, but the system will not be improved overnight. In the immediate term, even as we roll out additional investments, cars still represent the most viable way to travel in the vast majority of U.S. urban areas, and they continue to provide superior access to economic opportunities compared to other modes of transportation in all but a handful of locations ( 5 , 45 , 46 ). Facilitating access to car travel via carsharing memberships and formal carpooling programs, and bolstering transit networks with ride-hail and taxi partnerships (see for example Miami-Dade County [ 47 ]) can help ensure that marginalized communities are not left behind while we build out a multimodal future.

Conclusion

Travel behaviors changed drastically during the first months of the COVID-19 shutdown. This research asked two questions about the potential equity implications of ride-hailing and the COVID-19 pandemic: first, how do neighborhood-level patterns of Uber trips compare before and during the pandemic? And second, how did the pandemic specifically affect patterns of identified essential travel? Using Uber data from four regions in California, we find that the total number of ride-hail trips fell by 82% between January–February and April–May. Trips serving identified essential destinations, however, fell by far less (62%), suggesting that ride-hail may serve an important role in providing access to essential services and goods like grocery stores and medical appointments. Changes in ride-hail use were unevenly distributed across neighborhoods, with the highest-income areas showing the largest decreases. Tracts with higher shares of transit commuters and higher shares zero-car households also showed steeper declines in trip making during the pandemic; however, neighborhoods with more older (45+) residents, and a higher share of Black, Hispanic/Latinx, and Asian residents still appear to rely more on ride-hail during the pandemic relative to other neighborhoods. These findings suggest divergent mobility experiences among car-less versus car-free travelers. In planning for the long term, cities should invest in robust and redundant transportation systems–including high-quality transit, walkable neighborhoods, and bike lanes–which together create a resilient mobility network.

Supplemental Material

sj-docx-1-trr-10.1177_03611981211037246 – Supplemental material for Equity Implications of Ride-Hail Travel during COVID-19 in California

Supplemental material, sj-docx-1-trr-10.1177_03611981211037246 for Equity Implications of Ride-Hail Travel during COVID-19 in California by Anne Brown and Rik Williams in Transportation Research Record

Footnotes

Author Contributions

The authors confirm contribution to the paper as follows: study conception and design: Brown and Williams; data collection: Williams; analysis and interpretation of results: Brown and Williams; draft manuscript preparation: Brown and Williams. All authors reviewed the results and approved the final version of the manuscript.

Declaration of Conflicting Interests

The author(s) declared the following potential conflicts of interest with respect to the research, authorship, and/or publication of this article: One of the authors is an employee and shareholder of Uber. No direct funding was provided by Uber for this study. The authors preregistered this study, which helps provide transparency with regard to (and track changes made to) the data and methodology used.

Funding

The author(s) received no financial support for the research, authorship, and/or publication of this article.

ORCID iDs

Anne Brown

Rik Williams

Supplemental Material

Supplemental material for this article is available online.

References

American Public Transportation Association. The Impact of the COVID-19 Pandemic on Public Transit Funding Needs in the U.S. American Public Transportation Association, Washington, D.C., 2020.

Domonoske

The Pandemic Emptied American Roads. But Driving is Picking Back Up. NPR, May20, 2020.

Puentes

COVID’s Differing Impact on Transit Ridership. Eno Center for Transportation, Washington, D.C., 2020.

Coveney

O’Dwyer

L. A.

Effects of Mobility and Location on Food Access. Health & Place, Vol. 15, No. 1, 2009, pp. 45–55.

Grengs

Job Accessibility and the Modal Mismatch in Detroit. Journal of Transport Geography, Vol. 18, No. 1, 2010, pp. 42–54.

Burdette

Larson-Friend

Transit Agencies Responses to COVID-19. University of Oregon, Eugene, OR, 2020.

Dong

Trade Uber for the Bus? An Investigation of Individual Willingness to Use Ride-Hail Versus Transit. Journal of the American Planning Association, Vol. 86, No. 2, 2020, pp. 222–235.

Gehrke

S. R.

Felix

Reardon

T. G.

Substitution of Ride-Hailing Services for More Sustainable Travel Options in the Greater Boston Region. Transportation Research Record: Journal of the Transportation Research Board, 2019. 2673: 438–446.

Conway

Salon

King

Trends in Taxi Use and the Advent of Ridehailing, 1995–2017: Evidence From the US National Household Travel Survey. Urban Science, Vol. 2, No. 3, 2018, p. 79.

10.

Young

Farber

The Who, Why, and When of Uber and Other Ride-Hailing Trips: An Examination of a Large Sample Household Travel Survey. Transportation Research Part A: Policy and Practice, Vol. 119, 2019, pp. 383–392.

11.

Brown

Redefining Car Access: Ride-Hail Travel and Use in Los Angeles. Journal of the American Planning Association, Vol. 85, No. 2, 2019, pp. 83–95.

12.

Deka

Fei

A Comparison of the Personal and Neighborhood Characteristics Associated with Ridesourcing, Transit Use, and Driving with NHTS Data. Journal of Transport Geography, Vol. 76, 2019, pp. 24–33.

13.

Lavieri

P. S.

Bhat

C. R.

Investigating Objective and Subjective Factors Influencing the Adoption, Frequency, and Characteristics of Ride-Hailing Trips. Transportation Research Part C: Emerging Technologies, Vol. 105, 2019, pp. 100–125.

14.

Henao

. Impacts of Ridesource—Lyft and Uber—on Transportation Including VMT, Mode Replacement, Parking, and Travel Behavior. PhD thesis. University of Colorado, 2017.

15.

Sabouri

Park

Smith

Tian

Ewing

Exploring the Influence of Built Environment on Uber Demand. Transportation Research Part D: Transport and Environment, Vol. 81, 2020, p. 102296.

16.

Gehrke

S. R.

Uber Service Area Expansion in Three Major American Cities. Journal of Transport Geography, Vol. 86, 2020, p. 102752.

17.

Jiao

Wang

Shared Mobility and Transit-Dependent Population: A New Equity Opportunity or Issue?

International Journal of Sustainable Transportation, Vol. 15, No. 4, 2021, pp. 294–305.

18.

Lavieri

P. S.

Dias

F. F.

Juri

N. R.

Kuhr

Bhat

C. R.

A Model of Ridesourcing Demand Generation and Distribution. Transportation Research Record: Journal of the Transportation Research Board, 2018. 2672: 31–40.

19.

Marquet

Spatial Distribution of Ride-Hailing Trip Demand and Its Association with Walkability and Neighborhood Characteristics. Cities, Vol. 106, 2020, p. 102926.

20.

San Francisco County Transportation Authority. TNCs Today: A Profile of San Francisco Network Company Activity. SFCTA, San Francisco, CA, 2017.

21.

Washington Metropolitan Area Transit Authority. Metro Announces Additional COVID-19 Changes, Including Reduced Service Beginning Monday. WMATA, Washington, D.C., 2020.

22.

Nelson

L. J.

Coronavirus Forces Deep Cuts to L.A. Metro Bus and Rail Service. Los Angeles Times, April, 17, 2020.

23.

Chapman

Port Authority Cuts Down on Service Due to COVID-19. Pittsburgh’s Action 4 News, March24, 2020.

24.

Tumlin

The Future of Transit Service through the Health and Budget Crisis. SFMTA, San Francisco, CA, 2020.

25.

District Department of Transportation. Shared Mobility for Essential Workers. https://ddot.dc.gov/page/shared-mobility-essential-workers. Accessed September 2020.

26.

Hamblin

Meyer

Millennials are Buying “COVID Cars”. Social Distance, The Atlantic, August 17, 2020.

27.

State of California. Executive Order N-33-20. State of California, Sacramento, CA, 2020.

28.

U.S. Census Bureau. 2018 5-Year American Community Survey. U.S. Census Bureau, Suitland, MA, 2018.

29.

Foursquare. Venue Categories. Foursquare, New York, NY, 2020.

30.

Crook

Uber Taps Foursquare’s Places Data So You Never Have To Type an Address Again. TechCrunch, May25, 2016.

31.

Transit App. How Coronavirus is Disrupting Public Transit. https://transitapp.com/coronavirus. Accessed September 2020.

32.

Brown

A. E.

Who and Where Rideshares? Rideshare Travel and Use in Los Angeles. Transportation Research Part A: Policy and Practice, Vol. 136, 2020, pp. 120–134.

33.

Joselow

There is Little Evidence That Mass Transit Poses a Risk of Coronavirus Outbreaks. E&E NEWS, July28, 2020.

34.

Brown

A. E.

Car-Less or Car-Free? Socioeconomic and Mobility Differences Among Zero-Car Households. Transport Policy, Vol. 60, 2017, pp. 152–159.

35.

Blumenberg

Smart

Getting By with a Little Help from My Friends… and Family: Immigrants and Carpooling. Transportation, Vol. 37, No. 3, 2010, pp. 429–446.

36.

Pendall

Hayes

George

McDade

Dawkins

C. J.

Jeon

J. S.

Knapp

Blumenberg

Pierce

Smart

Driving to Opportunity: Understanding the Links Among Transportation Access, Residential Outcomes, and Economic Opportunity for Housing Voucher Recipients. Urban Institute, Washington, D.C., 2014.

37.

Pucher

Renne

J. L.

Socioeconomics of Urban Travel: Evidence from the 2001 NHTS. Transportation Quarterly, Vol. 57, No. 3, 2003, pp. 49–77.

38.

Mongey

Pilossoph

Weinberg

Which Workers Bear the Burden of Social Distancing Policies?National Bureau of Economic Research, Cambridge, MA, 2020.

39.

Taylor

Morris

E. A.

Public Transportation Objectives and Rider Demographics: Are Transit’s Priorities Poor Public Policy?

Transportation, Vol. 42, No. 2, 2015, pp. 347–367.

40.

Charm

Coggins

Robinson

Wilkie

The Great Consumer Shift: Ten Charts That Show How US Shopping Behavior is Changing. https://www.mckinsey.com/business-functions/marketing-and-sales/our-insights/the-great-consumer-shift-ten-charts-that-show-how-us-shopping-behavior-is-changing. Accessed August 9, 2021.

41.

NHTS. 2017 National Household Travel Survey. NHTS, Washington, D.C., 2017.

42.

Brown

Manville

Weber

Mobility on Demand in the Los Angeles Region: Evaluation of the Use and Performance of the MOD Pilot. Eno Center for Transportation, Washington, D.C., 2021.

43.

Morris

E. A.

Pratt

A. N.

Zhou

Brown

Khan

S. M.

Derochers

J. L.

Campbell

Chowdhury

M. R.

Assessing the Experience of Providers and Users of Transportation Network Company Ridesharing Services. Center for Connected Multimodal Mobility, Clemson, SC, 2019.

44.

Litman

Pandemic-Resilient Community Planning: Practical Ways to Help Communities Prepare for, Respond to, and Recover from Pandemics and Other Economic, Social and Environmental Shocks. Victoria Transport Policy Institute, Victoria, British Columbia Canada, 2020.

45.

Shen

A Spatial Analysis of Job Openings and Access in a US Metropolitan Area. Journal of the American Planning Association, Vol. 67, No. 1, 2001, pp. 53–68.

46.

Kawabata

Shen

Commuting Inequality Between Cars and Public Transit: The Case of the San Francisco Bay Area, 1990-2000. Urban Studies, Vol. 44, No. 9, 2007, pp. 1759–1780.

47.

Miami-Dade County. Go Nightly. https://www.miamidade.gov/global/transportation/nightly-bus-service.page. Accessed September 2020.

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

0.00 MB

0.02 MB