Examining Post-pandemic Ride-Hailing Use in California and Its Determinants

Abstract

The prevalence and utilization of ride-hailing grew substantially in the 2010s; however, this growth was halted by the onset of the COVID-19 pandemic. In light of the disruptive impacts of the pandemic, it remains to be seen whether ride-hailing use in the post-pandemic period will be similar to that of the pre-pandemic period. Given the negative externalities associated with ride-hailing before the pandemic (including increases in vehicle miles traveled, induced travel, and the replacement of trips made using more sustainable modes), it is crucial to understand the potential long-term impacts of the pandemic on ride-hailing use. This study uses data from two waves of a repeated cross-sectional survey of California residents to examine ride-hailing use during the post-pandemic period and its determinants. Specifically, the data are used to compare pre- and post-pandemic ride-hailing use, estimate hurdle models of post-pandemic ride-hailing frequency, and estimate a zero-inflated negative binomial regression model of the frequency of post-pandemic shared ride-hailing use. The results suggest that the use of ride-hailing for commuting trips is more common post-pandemic than it was pre-pandemic. Moreover, commuting frequency was positively associated with how often ride-hailing was used for commuting trips. The results also indicate that the uptake of shared ride-hailing is relatively low during the post-pandemic period, suggesting that additional strategies (such as encouraging vehicle electrification) may be needed to help address the negative externalities associated with ride-hailing. Overall, this information can inform policies that aim to mitigate the negative externalities associated with ride-hailing in the post-pandemic era.

In the 2010s, the prevalence and use of ride-hailing services (such as those offered by Uber and Lyft in the United States) grew substantially. These services, which are also referred to as ride-sourcing or ridesharing services, allow users to request a ride using a smartphone application. In the midst of this growth, significant effort was dedicated to understanding the potential impacts of ride-hailing, with studies highlighting the potential benefits and negative externalities of these services. For example, ride-hailing can improve the mobility and accessibility of those who do not (or cannot) travel by private vehicle and can help users overcome spatial or temporal gaps in the transit network ( 1 – 3 ). However, studies have also found that ride-hailing can attract demand from more sustainable modes, enable trips that would not have been made otherwise, and contribute to increases in vehicle miles traveled (VMT), which can contribute to congestion and emissions ( 1 , 4 – 6 ). Prior studies have noted that shared ride-hailing services (i.e., the category of ride-hailing services that offer customers discounted fares in exchange for the potential to share their ride with customers traveling to a similar destination) have the potential to help address the negative externalities associated with ride-hailing. However, the potential benefits of shared ride-hailing services (such as UberX Share) are heavily influenced by the uptake and willingness to use these services ( 7 – 9 ).

The growing utilization of ride-hailing services was halted in 2020 due to the onset of the COVID-19 pandemic, which significantly disrupted activity participation and travel behavior in cities around the world ( 10 , 11 ). In response to the pandemic, numerous studies have examined the near-term impacts of the pandemic on travel behavior. However, there is currently a limited number of studies on travel behavior during the post-pandemic period (defined in this study as the period after the World Health Organization declared an end to COVID-19 as a public health emergency in May 2023 [ 12 ]). Aggregate statistics suggest that VMT in the United States has rebounded to pre-pandemic levels, while transit ridership and ride-hailing use have not ( 13 – 15 ). Given the impacts of ride-hailing before the pandemic, and in light of the disruptive impacts of the pandemic on travel behavior, it is crucial to understand the potential long-term impacts of the pandemic on ride-hailing use.

This study uses data from two waves of a repeated cross-sectional survey of California residents to examine the determinants of ride-hailing frequency in the post-pandemic period and to compare ride-hailing use during the pre- and post-pandemic periods. As part of this study, hurdle models are estimated to examine the determinants of ride-hailing frequency, and a zero-inflated negative binomial (ZINB) regression model is estimated to investigate the factors influencing the use of shared ride-hailing services among ride-hailing users. The results offer insights into the similarities and differences between ride-hailing use during the pre- and post-pandemic periods and shed light on the determinants of shared ride-hailing use during the post-pandemic period. This information can help inform strategies aiming to address the negative externalities associated with ride-hailing services in the post-pandemic period.

The remainder of the paper is organized as follows. First, a review of studies on the determinants of ride-hailing frequency is presented. Next, information concerning the data used in this study is provided and the formulations of the hurdle and ZINB regression models are presented. This is followed by a summary of the key findings of the study and a discussion of their implications.

Literature Review

Determinants of Ride-Hailing Frequency during the Pre-pandemic Period

The growing utilization of ride-hailing prompted investigations into the factors influencing how often an individual uses these services. At an aggregate level, it appears that there is a subset of ride-hailing users that make a disproportionately large percentage of all ride-hailing trips ( 16 , 17 ). At the user level, studies on the topic have noted that how often an individual uses ride-hailing services can vary based on personal and household characteristics. For example, studies have consistently found a negative association between age and ride-hailing frequency ( 18 – 20 ). Conversely, education, employment, and household income tend to be positively associated with ride-hailing frequency ( 21 – 26 ). Notable exceptions to these trends include: 1) Mitra et al. ( 27 ), who noted that education was negatively associated with ride-hailing frequency among older adults in the Unites States, and 2) Abouelela et al. ( 28 ), who noted that income can negatively affect ride-hailing frequency in their study on the use of a shared ride-hailing service in Mexico City. In addition, previous studies have underscored the potential for factors such as gender, ethnicity, race, and household vehicle ownership to influence ride-hailing frequency ( 21 , 25 , 28 – 30 ).

Prior studies have also highlighted the potential for the ride-hailing frequency of an individual to differ based on the attributes of the area in which they live. For example, living in an urban area tends to be positively associated with ride-hailing frequency ( 19 , 24 , 31 ), while the opposite has been observed for those living in a suburb or small town ( 20 ). In addition, living in an area with greater population ( 22 , 27 ), job ( 21 ), and activity ( 25 , 30 ) density tends to be associated with greater ride-hailing frequency. Conversely, studies using data collected from residents of California have noted that living in an area with greater land-use mix is negatively associated with ride-hailing frequency ( 19 , 25 ).

Impacts of the COVID-19 Pandemic on Ride-Hailing Use

The onset of the pandemic resulted in a significant decline in ride-hailing use in the United States, with studies utilizing data from New York City, Chicago, and California suggesting that ride-hailing trips decreased by approximately 80% ( 15 , 32 – 34 ). This decline in ride-hailing use has been attributed to two key factors. First, the non-pharmaceutical interventions that were implemented in an effort to reduce the spread of COVID-19 contributed to substantial decreases in out-of-home activity participation, particularly during the early stages of the pandemic ( 35 – 37 ). Second, the pandemic resulted in a shift in travel mode preferences, with the preference for individual modes (e.g., private vehicles, active modes) increasing and the preference for shared modes (e.g., public transit, ride-hailing) decreasing ( 38 – 43 ). This shift in preferences appeared to stem from differences in the perceived risk of infection associated with different travel modes ( 44 – 48 ). Moreover, the number of daily reported COVID-19 cases has been found to negatively affect ride-hailing demand during the pandemic ( 49 – 51 ).

Studies exploring the impacts of the pandemic have noted that declines in ride-hailing demand varied based on zone-level attributes. For example, Brown and Williams ( 32 ) found that decreases in ride-hailing demand were greater in neighborhoods with higher median incomes in California, while Debnath et al. ( 50 ) noted a similar finding for census tracts in Chicago. Similarly, the percentage of residents between the ages of 20 and 44 and between the ages of 25 and 34 were positively associated with declines in ride-hailing demand in California ( 32 ) and Chicago ( 33 ), respectively. Conversely, Brown and Williams ( 32 ) noted that neighborhoods with a greater share of residents who identify as Hispanic, Asian, or Black or African American exhibited smaller declines in ride-hailing demand. Built environment variables have also been found to influence the extent to which ride-hailing demand declined at the onset of the pandemic. In particular, population density and land-use mix were associated with lower declines, while employment density and walkability were associated with greater declines ( 32 , 33 , 50 ). In addition, the extent to which ride-hailing demand rebounded from the impacts of the pandemic appears to vary based on the attributes of the residents and built environment of a given zone ( 52 , 53 ).

A limited number of studies have also explored the impacts of the pandemic on how often ride-hailing services are used. Overall, studies on the topic suggest that ride-hailing services were used less often during the pandemic than they were before the pandemic, with certain individuals refraining from using ride-hailing during the pandemic ( 54 ). However, there is also evidence that the onset of the pandemic resulted in a small subset of individuals using ride-hailing services more often. Specifically, the results of studies using data collected through web-based surveys of residents of the Greater Toronto Area (9.5%) ( 55 ), Dhaka (5.8%) ( 56 ), and Indonesia (1.9%) ( 57 ) suggest that a subset of respondents used ride-hailing more often during the pandemic than they did during the pre-pandemic period. Similar results were reported by Kiriazes and Watkins ( 58 ) in their study of the Atlanta metropolitan area, where 4% of respondents who were classified as “occasional and active” ride-hailing users indicated that they increased their use of these services. In their examination of the impacts of the pandemic on ride-hailing use in the Greater Toronto Area, Loa et al. ( 59 ) found that the most common motivations for using ride-hailing more often were related to concerns about the reliability of public transit, an aversion to using public transit, and the desire to avoid crowds.

Ride-Hailing Use in the Post-pandemic Era

A relatively limited number of studies have examined the use of ride-hailing services during the latter stages of the pandemic and beyond. For example, Zhao et al. ( 60 ) estimated multi-scale geographically weighted regression models to explore the determinants of ride-hailing demand in Nanjing during April 2022. The results suggest that the density of roads, bus stops, and metro stations were positively associated with demand, while the density of bike lanes had the opposite effect. Similarly, Jin et al. ( 61 ) estimated a generalized additive mixture model to understand the determinants of ride-hailing demand in Chicago and explore whether their impacts were altered by the pandemic. Notably, the authors found that the impacts of factors such as population density, bus accessibility, and land-use mix differed between the pre- (March to June 2019) and post-pandemic (i.e., March to June 2022) periods. Conversely, the authors also found that the impacts of factors such as median household income, the percentage of residents who identify as Hispanic, employment density, road density, and rail accessibility remained unchanged.

Additionally, there have been efforts to understand the factors influencing how often an individual uses ride-hailing services during this period. In one of the few studies on the topic, Sadeghvaziri et al. ( 62 ) estimated an ordered logit model to examine the determinants of ride-hailing frequency among US residents. Using data from the 2022 National Household Travel Survey (NHTS), the authors found that ride-hailing was used more often by those who identified as Black or African American, those with a college or bachelor’s degree, and those from higher-income households. Conversely, the authors noted that ride-hailing was used less often by older individuals and those who live in rural areas. Applying the same approach, Du et al. ( 63 ) used data from a web-based survey conducted in November 2022 to investigate the determinants of ride-hailing frequency among residents of Nanjing. The results suggest that ride-hailing tends to be used less often by those who typically travel by private vehicle and those who expressed concerns about the security of their personal information when using ride-hailing. Finally, Tran et al. ( 64 ) used data collected through face-to-face surveys conducted during March 2023 in Hanoi to understand the factors influencing the use of car- and motorcycle-based ride-hailing services. Through the estimation of a bivariate ordered probit model, the authors found that education and vehicle ownership were both positively associated with the probability of using car-based ride-hailing services on a frequent basis.

While previous studies offer important insights into the factors influencing ride-hailing demand and frequency, there are three notable gaps in the literature. First, there are only a handful of studies that have examined the determinants of ride-hailing frequency during the post-pandemic period. Given the negative externalities associated with ride-hailing before the pandemic, insights into the factors influencing the use of these services can contribute to efforts aiming to address the negative externalities associated with these services during the post-pandemic period. Second, studies investigating the determinants of post-pandemic ride-hailing frequency do not distinguish between the use of these services for commuting and non-commuting trips. In light of the changes in the prevalence of remote work and online activities that occurred during the pandemic ( 65 , 66 ), distinguishing between commuting and non-commuting trips can contribute to a more holistic understanding of differences between pre- and post-pandemic ride-hailing use. Third, there has yet to be a study examining the determinants of how often shared ride-hailing services are being used during the post-pandemic period. Despite their potential to help address the negative externalities associated with ride-hailing services ( 28 , 67 ), the utilization of shared ride-hailing services is substantially lower during the post-pandemic period than it was during the pre-pandemic period ( 15 ). Consequently, understanding the determinants of shared ride-hailing use during the post-pandemic period can offer insights into the attributes of those who are currently using these services. To help address these gaps, this study uses data from two waves of a repeated cross-sectional survey of California residents to: 1) understand the determinants of post-pandemic ride-hailing frequency for commuting and non-commuting trips, 2) explore how post-pandemic ride-hailing use compares with pre-pandemic use, and 3) examine the determinants of shared ride-hailing use in the post-pandemic period. The results can offer insights into the potential long-term impacts of the COVID-19 pandemic on ride-hailing use and how the uptake of shared ride-hailing in the post-pandemic period can vary among different segments of the population.

Data and Methodology

Data Description

Survey Design and Conduct

The data used in this study were collected through the fall 2020 and fall 2023 waves of the California Mobility Panel (CMP) survey, which was first conducted in 2018. Both waves of the survey were implemented using a web-based survey platform. Several methods were used to recruit participants for the fall 2020 and fall 2023 waves of the survey. First, those who completed a prior wave were invited to participate in the survey; these individuals were given a $10 (fall 2020) or $5 (fall 2023) electronic gift card after completing the survey. Second, quota sampling was applied to recruit participants from an online opinion panel. As part of this process, quotas based on factors such as age, gender, race, ethnicity, employment, income, and home location were used to help improve the extent to which the sample represented the study area. Upon completion of the survey, respondents recruited through this method were provided with non-monetary compensation by the company that maintains the panel.

During the fall 2020 wave, participants were also recruited through social media advertisements and professional email lists (e.g., subscribers to newsletters). Respondents recruited through these methods were entered into a random draw to win one of 10 $100 electronic gift cards or one of 200 $10 electronic gift cards after completing the survey. As part of the fall 2023 wave, stratified random sampling was applied to select households that would receive mailed invitations to participate in the survey. As part of this process, households in areas designated as equity priority ( 68 ) or disadvantaged ( 69 ) communities were oversampled. On completion of the survey, respondents recruited through this method were entered into a random draw to win one of 10 $100 electronic gift cards or one of 500 $10 electronic gift cards. See Ozbilen et al. ( 66 ) for more information about the design and conduct of the CMP.

In both the fall 2020 and the fall 2023 waves, respondents were asked to provide information about their personal and household characteristics, how often they worked from various locations, and their use of shared mobility services in the month before the survey. In addition, respondents were asked to indicate how often they used various travel modes for commuting and non-commuting (e.g., shopping, social, and leisure) trips. Moreover, respondents of the fall 2020 wave were asked to report their use of various travel modes for commuting and non-commuting trips during fall 2019. There are two notable differences in the wording used for questions pertaining to the use of travel modes for commuting and non-commuting trips between the two waves of the survey. First, fall 2020 respondents were asked to indicate how often they used various modes during fall 2019, whereas fall 2023 respondents were asked to indicate how often they used various modes in July and August 2023. Second, the response options provided to fall 2020 respondents were characterized by the number of times that they used each mode, whereas the response options provided to fall 2023 respondents were characterized by the number of days that they used each mode.

Data Processing and Cleaning

Before the data were used for the analysis, several steps were taken to clean, process, and supplement the data. First, responses from those who selected the incorrect response to the attention-check questions, incomplete responses, and responses from those that completed the survey in a relatively short amount of time were removed. Second, responses from respondents who live outside of California were removed from the dataset. Following this process, 4,969 responses from the fall 2020 wave and 4,369 responses from the fall 2023 wave remained. These sample sizes include 630 respondents who completed both waves of the survey (386 of whom were employed at the time when they completed the surveys).

Next, weights were developed to help improve the extent to which the data represented the population of the study area. An iterative procedure was used to develop weights through the application of iterative proportional fitting, with information from both the American Community Survey (ACS) and NHTS being used as control variables. Specifically, variables pertaining to age, gender, race, ethnicity, education, employment, household income, and frequency of remote work were used as the control variables in the weighting procedure. The weights developed for respondents from the fall 2020 wave of the survey used data from the 2020 ACS five-year estimates, while the weights developed for respondents from the fall 2023 wave used data from the 2022 ACS five-year estimates (unfortunately, data from the 2023 ACS was not available at the time the weights were developed). The iterative weighting procedure continued until the maximum difference between the weights produced in successive iterations satisfied the convergence criterion. The weighing procedure was applied using the mipfp package that was written for the R programming language ( 70 ) and the weighting procedure was based on the procedure applied by Wang et al. ( 71 ).

Following the weighting procedure, responses from those who did not complete the question concerning the use of various travel modes for non-commuting trips were removed. Responses from those who did not complete the question concerning their use of various travel modes for commuting trips despite being employed or students were also removed from the dataset. Following this process, 3,990 responses from the fall 2020 wave and 4,124 responses from the fall 2023 wave remained. Finally, information from the Smart Location Database (SLD) maintained by the US Environmental Protection Agency was used to supplement the survey data based on the Census block group (CBG) in which a respondent resides. The SLD contains a variety of information on land use and built environment attributes, including density, diversity, design, and destination accessibility. See Chapman et al. ( 72 ) for more information concerning the SLD.

Sample Statistics

The weighted distributions of key personal and household attributes in the fall 2020 sample, the fall 2023 sample, and five-year estimates from the 2022 ACS are compared in Table 1. In both samples, the percentages of respondents aged 18 to 29 and aged 60 and over are lower than that of the ACS. Similarly, the percentages of respondents who identify as Hispanic are lower in the samples compared to the ACS, whereas the percentages of respondents who identify as non-Hispanic White are higher. Discrepancies in educational attainment are also observed, as the percentages of individuals with a high school education or below are lower in the samples compared to the ACS, while the percentages of those with a bachelor’s degree or above are higher. Moreover, the percentages of individuals from households earning over $100,000 annually are lower in the samples compared with the ACS, whereas the percentages of individuals from households earning between $50,000 and $99,999 are higher.

Table 1.

Comparison of Weighted Distributions of Sociodemographic Attributes: Fall 2020, Fall 2023, 2022 ACS

Category	Fall 2020 (%)	Fall 2023 (%)	2022 ACS (%)
Age
18 to 29	15.6	19.5	21.8
30 to 39	22.1	20.7	18.7
40 to 49	22.0	17.3	16.6
50 to 59	16.1	16.1	16.2
60 and over	24.2	26.4	26.7
Gender
Male	51.5	49.9	50.1
Female	48.5	50.1	49.9
Race and ethnicity
Hispanic alone	33.3	37.6	39.7
Non-Hispanic White alone	43.3	36.6	35.2
Non-Hispanic Black or African American alone	3.9	4.9	5.3
Asian or Pacific Islander alone	13.0	15.0	15.2
Native American alone	0.4	0.3	0.3
Other	6.1	5.7	4.3
Education
High school or below	22.0	33.1	36.0
Some college	31.6	29.2	28.1
Bachelor’s degree or above	46.4	37.6	35.9
Household income
Less than $50,000	26.2	28.9	27.9
$50,000 to $99,999	30.5	25.9	25.7
$100,000 and above	43.3	45.3	46.4
Home location
San Francisco Bay Area	23.3	16.1	20.1
Los Angeles Region	44.1	47.5	47.6
San Diego County	12.2	9.6	8.5
Sacramento Region	13.9	10.8	6.5
Central Valley	3.3	9.7	10.1
Other	3.3	6.4	7.1
Sample size	3,990	4,124	-

Note: ACS = American Community Survey.

Model Formulation

Overview

Two types of models were estimated to gain insights into the determinants of ride-hailing use during the post-pandemic period: hurdle models and ZINB regression models. Both types of models are applied when examining the factors influencing how often an outcome of interest occurs. In addition, both types of models include a component that distinguishes between those who have never exhibited the outcome of interest and those who did not exhibit the outcome during the survey period. The inclusion of this component allows for the distinction between the factors that influence how often the outcome of interest occurs and the factors influencing whether the outcome occurs. Avoiding the conflation of these factors helps contribute to a more comprehensive understanding of the factors influencing whether an individual has used ride-hailing services during the post-pandemic period and the factors influencing how often these services are used.

The primary distinction between the hurdle and ZINB regression models is that the former is applied in situations where the outcome of interest is an ordinal variable, while the latter is applied in situations where the outcome of interest is a count variable. This distinction results in the hurdle and ZINB regression models being formulated based on different set of assumptions. As part of the fall 2023 wave of the CMP survey, respondents were asked to report how often they used ride-hailing using a seven-point ordered scale (with response options ranging from not available to 5 or more days per week) and the number of shared ride-hailing trips that they made in the month before the survey. Consequently, hurdle models were used to gain insights into the factors influencing post-pandemic ride-hailing use and a ZINB regression model was used to examine the determinants of shared ride-hailing use. The formulation of the hurdle and ZINB regression models that were estimated as part of this study are summarized in the following subsections.

Hurdle Model

Hurdle models were estimated to examine the determinants of ride-hailing use for a given trip purpose during the post-pandemic period. The hurdle models used in this study consisted of an availability component and a frequency component. The former was used to examine the factors influencing whether a respondent regarded ride-hailing as an available mode of travel for the given trip purpose, while the latter was used to explore the determinants of ride-hailing frequency. Let $y_i$ represent the ride-hailing use of individual $i$ for a given trip purpose, where $y_i=0$ corresponds to ride-hailing not being regarded as an available travel mode for a given trip purpose. A binary logit model was used for the availability component of the hurdle model. Let $U_i$ represent the utility obtained by individual $i$ when regarding ride-hailing as an available mode for a given trip purpose. This utility is comprised of a deterministic ( $V_i$ ) component, which can be modelled as a function of explanatory variables, and a random ( ${\epsilon }_i$ ) component

$$U_i=V_i+{\epsilon }_i={\mathbf{\beta }}^{\prime }{\mathit{x}}_{\mathit{i}}+{\epsilon }_i$$

(1)

where $\mathbf{\beta }$ = vector of parameters, and ${\mathit{x}}_{\mathit{i}}$ = vector of explanatory variables corresponding to individual $i$ .

Under the assumption that ${\epsilon }_i$ follows the Type I Extreme Value distribution, the probability of individual $i$ concerning ride-hailing as an available mode is ( 73 ):

$$P(y_i>0)=\frac{\exp (V_i)}{1+\exp \left(V_i\right)}=\frac{\exp \left({\mathbf{\beta }}^{\prime }{\mathit{x}}_{\mathit{i}}\right)}{1+\exp \left({\mathbf{\beta }}^{\prime }{\mathit{x}}_{\mathit{i}}\right)}$$

(2)

The frequency component of the hurdle model was defined based on the formulation of the ordered logit model. In this model, it is assumed that observed ordinal outcomes are influenced by the value of a continuous latent variable ( 74 )

$$y_i^{*}={\mathbf{\theta }}^{\prime }{\mathit{z}}_{\mathit{i}}+{\xi }_i$$

(3)

where $y_i^{*}$ = value of the continuous latent variable corresponding to individual $i$ , $\mathbf{\theta }$ = vector of parameters, ${\mathit{z}}_{\mathit{i}}$ = vector of explanatory variables corresponding to individual $i$ , and ${\xi }_i$ = continuous random disturbance term corresponding to individual $i$ .

Based on the value of the latent variable, the value of the observed outcome is determined based on a censoring mechanism ( 74 )

$$y_i=j\mathrm{if}{\tau }_{j-1}<y_i^{*}\le {\tau }_j\mathrm{for}j=1,\dots ,J$$

(4)

where $y_i^{*}$ = value of the continuous latent variable for individual $i$ , and ${\tau }_j$ = threshold parameter corresponding to ordered outcome j.

Under the assumption that ${\xi }_i$ follows the standardized logistic distribution, the probability of individual $i$ displaying ride-hailing frequency $j>0$ is

$$P(y_i=j|j>0)=\frac{\exp ({\tau }_j-{\mathbf{\theta }}^{\prime }{\mathit{z}}_{\mathit{i}})}{1+\exp ({\tau }_j-{\mathbf{\theta }}^{\prime }{\mathit{z}}_{\mathit{i}})}-\frac{\exp ({\tau }_{j-1}-{\mathbf{\theta }}^{\prime }{\mathit{z}}_{\mathit{i}})}{1+\exp ({\tau }_{j-1}-{\mathbf{\theta }}^{\prime }{\mathit{z}}_{\mathit{i}})}$$

(5)

Combining Equations 2 and 4, the unconditional probability of individual $i$ displaying ride-hailing frequency $j$ is defined as

$$P(y_i=j)=\{\begin{array}{c}\begin{array}{c}1-P(y_i>0)\mathrm{if}j=0\hfill \end{array}\hfill \\ \multicolumn{1}{c}{\begin{array}{c}P(y_i>0)*P(y_i=j|j>0)\mathrm{if}j>0\hfill \end{array}}\end{array}$$

(6)

Two sets of models were estimated—one using data collected through the fall 2023 wave of the CMP survey and another only using data from respondents who completed both the fall 2020 and fall 2023 waves of the survey (referred to here as the panel respondents). The latter set of models are used to explore the impacts of pre-pandemic ride-hailing use on post-pandemic ride-hailing frequency. The hurdle models were estimated through maximum likelihood estimation using the apollo package written for the R programming language ( 75 ).

ZINB Regression Model

The factors influencing the use of shared ride-hailing services in the month before the survey were examined using a ZINB model. This model is typically used in situations where an observed count outcome contains excess zeros and exhibits overdispersion. The use of a zero-inflated model allows for the distinction between respondents who will never use shared ride-hailing (i.e., structural zeros) and those who did not use shared ride-hailing in the month before the survey (i.e., random zeros) ( 76 ). The ZINB regression model consists of two components: the zero-inflation model, which distinguishes between structural and non-structural zeros, and the count model.

Let $r_i$ be the number of shared ride-hailing trips made by individual $i$ in the month before the survey. Define $P_1(r_i=0)$ as the probability that individual $i$ has never used shared ride-hailing services and $P_2(r_i)$ as the probability of individual $i$ reporting that they made $r_i\ge 0$ trips using shared ride-hailing. The probability that individual $i$ made $r_i=k$ trips is ( 77 )

$$P(r_i=k)=\{\begin{array}{c}\begin{array}{c}{P}_1(r_i=0)+(1-P_1(r_i=0))*P_2(r_i=0)\mathrm{if}k=0\hfill \end{array}\hfill \\ \multicolumn{1}{c}{\begin{array}{c}(1-P_1(r_i=0))*P_2(r_i=0)\mathrm{if}k>0\hfill \end{array}}\end{array}$$

(7)

In the ZINB model, binary logistic regression is typically used for the inflation model. The probability that individual $i$ will never use shared ride-hailing is

$$P_1(r_i=0)=\frac{\exp (\mathbf{\Upsilon }\prime {\mathbf{w}}_{\mathbf{i}})}{1+\exp (\mathbf{\Upsilon }\prime {\mathbf{w}}_{\mathbf{i}})}$$

(8)

where $\mathbf{\Upsilon }$ = vector of parameters, and ${\mathit{w}}_{\mathit{i}}$ = vector of explanatory variables corresponding to individual $i$ .

The count model is defined based on the formulation of negative binomial regression. The probability of individual $i$ displaying frequency $r_i=k$ is ( 77 )

$$\begin{array}{c}\multicolumn{1}{c}{P_2(r_i=k)=\frac{\mathrm{\Gamma }({\mathrm{\alpha }}^{-1}+k)}{\mathrm{\Gamma }\left({\mathrm{\alpha }}^{-1}\right)\mathrm{\Gamma }(k+1)}*{\left(\frac{{\mathrm{\alpha }}^{-1}}{{\mathrm{\alpha }}^{-1}+\mathrm{\mu }}\right)}^{{\mathrm{\alpha }}^{-1}}}\\ \multicolumn{1}{c}{*{\left(\frac{\mathrm{\mu }}{\mathrm{\mu }+{\mathrm{\alpha }}^{-1}}\right)}^k}\end{array}$$

(9)

where $\mathrm{\Gamma }$ = the Gamma integral, which reduces to a factorial when the argument is an integer, $\mathrm{\alpha }$ = dispersion parameter, and $\mathrm{\mu }$ = deterministic value.

The value of $\mathrm{\mu }$ in Equation (9) is modeled as a function of explanatory variables

$$\mathrm{\mu }=\exp \left({\mathbf{\psi }}^{\prime }{\mathit{q}}_{\mathit{i}}\right)$$

(10)

where $\mathbf{\psi }$ = vector of parameters, and ${\mathit{q}}_{\mathit{i}}$ =a vector of explanatory variables corresponding to individual $i$ .

The ZINB regression model was estimated through maximum likelihood estimation using the zeroinfl function that is available through the pscl package written for the R programming language ( 78 ). To help understand the use of shared ride-hailing among respondents with previous ride-hailing experience, the ZINB regression model was estimated using only responses from those who indicated that they have used ride-hailing at least once. This decision was motivated by shared ride-hailing being a specific category of ride-hailing services.

Results

Descriptive Analysis

Weighted data from the fall 2020 and fall 2023 waves of the CMP survey were used to compare ride-hailing use during the pre- and post-pandemic periods. As presented in Table 2, fall 2023 respondents were more likely to indicate that they have used ride-hailing for commuting trips compared with fall 2020 respondents. Moreover, the results of the ${\chi }^2$ test for proportions suggest that this difference is statistically significant. Given the relatively higher cost of using ride-hailing compared with other modes, this shift could stem from a reduction in commuting frequency among certain individuals because of the greater prevalence of hybrid work arrangements in the post-pandemic period compared with the pre-pandemic period. This trend is somewhat consistent with recent media reports suggesting that the use of ride-hailing for commuting trips may be more prevalent during the post-pandemic period than it was during the pre-pandemic period ( 79 , 80 ). In contrast, the difference in the percentage of respondents who indicated that they used ride-hailing for non-commuting trips was not statistically significant.

Table 2.

Percentage of Respondents Using Ride-Hailing for Commuting and Non-commuting Trips, by Time Period

Time period	% of respondents using ride-hailing	${\chi }^2$ statistic (p-value)
Commuting
Pre-pandemic	21.6	17.207 (<0.001)
Post-pandemic	25.8
Non-commuting
Pre-pandemic	24.5	0.256 (0.613)
Post-pandemic	25.0

Note: N (pre-pandemic) = 3,990; N (post-pandemic) = 4,124.

Weighted data from the fall 2023 wave was also used to examine ride-hailing frequency during the post-pandemic period. As shown in Figure 1, ride-hailing tends to be used on a relatively infrequent basis for both commuting and non-commuting trips, which is consistent with the findings of pre-pandemic studies. In addition, there is a percentage of respondents who do not regard ride-hailing as an available travel mode, even for non-commuting trips (e.g., those for social and recreational purposes). This could stem from the need to download a smartphone application and create an account before ride-hailing services can be used.

Figure 1.

Post-pandemic ride-hailing frequency among fall 2023 respondents, by trip purpose (N = 4,124).

Additionally, weighted data from the fall 2020 and fall 2023 waves were used to compare ride-hailing frequency in the pre- and post-pandemic periods. In light of the differences in how the questions were worded in the two waves, four frequency categories were defined: 1) not available, 2) available but not used, 3) less than once per month, and 4) at least once per month. As illustrated in Table 3, ride-hailing was more likely to be regarded as an available mode for commuting trips during the post-pandemic period compared to the pre-pandemic period. Moreover, ride-hailing was more likely to be used less than once per month for commuting trips during the postpandemic period compared to the pre-pandemic period. The results of the ${\chi }^2$ test for proportions suggest that these differences are statistically significant. Conversely, differences in ride-hailing frequency for non-commuting trips during the pre- and post-pandemic periods were not statistically significant.

Table 3.

Comparison of Ride-Hailing Frequency by Trip Purpose and Time Period

Frequency	Pre-pandemic (%)	Post-pandemic (%)	${\chi }^2$ statistic (p-value)
Commuting
Not available	26.8	22.2	20.857 (<0.001)
Available but not used	51.5	52.0	0.187 (0.666)
Less than once per month a	7.7	11.5	28.995 (<0.001)
At least once per month b	13.9	14.4	0.299 (0.585)
Non-commuting
Not available	25.2	24.1	1.584 (0.208)
Available but not used	50.2	50.9	0.420 (0.517)
Less than once per month a	10.5	10.2	0.186 (0.666)
At least once per month b	14.0	14.8	0.992 (0.319)

Note: N (pre-pandemic) = 3,990; N (post-pandemic) = 4,124.

a

This outcome corresponds to the “less than once a month” response option in the fall 2020 wave and the “less than one day a month” response option in the fall 2023 wave.

b

This outcome corresponds to the “1 to 3 times a month” and “at least once per week” response options in the fall 2020 wave and the “1 to 3 days a month” and “at least one day per week” response options in the fall 2023 wave.

Finally, weighted data from the fall 2023 wave of the CMP survey was used to explore how often shared ride-hailing services were used among respondents. Given that shared ride-hailing is a specific category of ride-hailing services, the frequency of shared ride-hailing use was only examined among respondents who indicated that they have used ride-hailing at least once. As shown in Figure 2, almost half of the respondents indicated that they have never used shared ride-hailing services. In addition, roughly 46% of respondents reported that they did not use these services in the month before the survey. These findings are consistent with trends observed in publicly available information from New York City and Chicago, which suggest that the use of shared ride-hailing is much less prevalent in the post-pandemic period than it was during the pre-pandemic period ( 15 ). Moreover, these findings somewhat echo the results of studies on the potential long-term impacts of the pandemic on ride-hailing use, which suggest that certain individuals may use ride-hailing less often ( 81 ) or refrain from using shared ride-hailing ( 82 ) during the post-pandemic period. This decline could stem from the adverse impacts of the pandemic on the willingness to use shared modes ( 47 ) or the suspension of these services during the pandemic. These findings call into question whether encouraging shared ride-hailing use is a viable strategy for addressing the negative externalities associated with ride-hailing services in the post-pandemic period.

Figure 2.

Shared ride-hailing use among respondents with previous ride-hailing experience (N = 2,697).

Post-pandemic Ride-Hailing Frequency

The final specifications of the hurdle models for commuting and non-commuting trips are summarized in Tables 4 and 5, respectively. Variables corresponding to personal and household characteristics and built environment attributes were tested during the process of finalizing the specifications of the hurdle models. In addition, variables pertaining to the frequency of working from one’s primary work location were tested in the hurdle models corresponding to commuting trips, and variables pertaining to pre-pandemic ride-hailing use were tested in the hurdle models that were estimated using data from the panel respondents. The decision of whether to retain a variable in the model was made based on the sign and t-statistic of the corresponding parameter. In this study, a variable was retained in the model if it was statistically significant at the 90% confidence level. This relatively lenient threshold was chosen because of the relative dearth of studies on post-pandemic ride-hailing frequency. The adjusted ${\mathrm{\rho }}^2$ values of the hurdle models exceeded 0.300, which is indicative of a reasonable goodness-of-fit ( 83 ). Descriptive statistics for the variables included in the final specifications of the hurdle models are summarized in online Appendix A. In addition, the factors influencing post-pandemic ride-hailing use for commuting and non-commuting trips are summarized in online Appendix B.

Table 4.

Final Specifications of the Hurdle Models for Commuting Trips

	Full sample		Panel respondents
Variable	Estimate	t-stat.	Estimate	t-stat.
Availability model
Constant	1.593	6.748	2.590	4.336
Age (years)	−0.020	−5.356	−0.025	−2.333
Person identifies as Black or African American [0/1]	0.419	1.632	NS	NS
Person identifies as Asian or Pacific Islander [0/1]	NS	NS	0.550	1.384
Education: bachelor’s degree or above [0/1]	0.535	4.680	NS	NS
Household income: less than $50,000 annually [0/1]	−0.311	−2.353	NS	NS
Person is renting their current residence [0/1]	0.199	1.645	NS	NS
Home location: other region [0/1]	−0.555	−4.314	−1.129	−2.390
Characterization of home location: Urban [0/1] a	0.256	2.151	NS	NS
Activity density in home CBG (jobs and housing units per acre)	NS	NS	−0.005	−2.091
Road network density in home CBG (links per sq. mi.)	0.008	1.477	NS	NS
Frequency of working from primary workplace: less than once per week [0/1]	0.260	2.155	0.738	2.201
Frequency of working from primary workplace: 1 or 2 days per week [0/1]	0.258	1.430	NS	NS
Frequency of using bus for commuting trips: at least once per week [0/1]	0.363	1.410	NS	NS
Frequency of using subway or rail for commuting trips: at least once per week [0/1]	0.756	2.110	NS	NS
Frequency of using bikes, e-bikes, or e-scooters for commuting trips: at least once per week [0/1]	0.447	1.808	NS	NS
Frequency of walking for commuting trips: at least once per week [0/1]	0.594	2.652	1.412	1.847
Person used ride-hailing for commuting trips pre-pandemic [0/1]	NS	NS	0.994	4.334
Frequency model
Age (years)	−0.024	−6.222	−0.021	−1.971
Gender: female [0/1]	−0.301	−3.084	NS	NS
Person identifies as Black or African American [0/1]	0.469	2.380	NS	NS
Person identifies as Hispanic [0/1]	0.453	4.095	NS	NS
Education: bachelor’s degree or above [0/1]	NS	NS	0.531	1.309
Household income: $100,000 or more annually [0/1]	NS	NS	0.597	2.021
Home location: Los Angeles Region [0/1]	NS	NS	0.835	2.919
Home location: other region [0/1]	−0.438	−2.748	NS	NS
Characterization of home location: urban [0/1] a	0.725	7.116	0.691	2.489
Number of household vehicles: 0 [0/1]	0.753	4.211	NS	NS
Population density in home CBG (people per acre)	0.008	3.195	NS	NS
Employment density in home CBG (jobs per acre)	0.003	3.315	NS	NS
Land-use mix based on 8-tier employment entropy in home CBG	NS	NS	−1.024	−1.479
Aggregate frequency of transit service per capita in home CBG (service frequency per capita)	0.451	1.771	NS	NS
Frequency of working from primary workplace: 1 or 2 days per week [0/1]	0.497	2.883	NS	NS
Frequency of working from primary workplace: 3 or more days per week [0/1]	0.463	4.117	NS	NS
Pre-pandemic ride-hailing frequency for commuting trips: 1 to 3 times per month [0/1]	NS	NS	2.085	5.802
Pre-pandemic ride-hailing frequency for commuting trips: At least once per week [0/1]	NS	NS	1.967	2.983
Threshold parameters
Available but not used | Less than a day per month	0.586	2.894	0.997	1.295
Less than a day per month | 1 to 3 days per month	1.418	6.894	2.335	2.992
1 to 3 days per month | At least one day per week	2.757	12.492	4.106	4.868
Goodness-of-fit measures
Number of observations	2,639		386
Null loglikelihood	−4,780.64		−705.62
Final loglikelihood	−3,122.90		−396.22
${\mathrm{\rho }}^2$ (Adjusted ${\mathrm{\rho }}^2$ )	0.347 (0.340)		0.438 (0.412)

Note: CBG = Census block group; [0/1] = binary variables; NS = parameter was not statistically significant.

a

Respondents were asked to choose from the following response options: urban, suburban, small town, and rural area.

Table 5.

Final Specifications of the Hurdle Models for Non-commuting Trips

	Full sample		Panel respondents
Variable	Estimate	t-stat.	Estimate	t-stat.
Availability model
Constant	1.336	6.428	0.498	0.853
Age (years)	−0.015	−5.935	−0.009	−1.314
Person identifies as Hispanic [0/1]	−0.234	−2.353	NS	NS
Education: bachelor’s degree or above [0/1]	0.186	2.177	NS	NS
Household income: less than $50,000 annually [0/1]	−0.298	−3.361	NS	NS
Household income: $100,000 or more annually [0/1]	NS	NS	0.433	2.055
Home location: other region [0/1]	−0.504	−5.200	−0.621	−1.670
Characterization of home location: urban [0/1] a	0.217	2.379	NS	NS
Road network density in home CBG (links per sq. mi.)	0.015	2.569	NS	NS
National Walkability Index b	0.031	2.362	0.080	2.882
Frequency of using car-sharing for non-commuting trips: at least once per week [0/1]	0.941	3.406	NS	NS
Frequency of using bikes, e-bikes, or e-scooters for non-commuting trips: At least once per week [0/1]	0.655	3.358	NS	NS
Frequency of walking for non-commuting trips: at least once per week [0/1]	0.326	2.832	0.635	2.013
Person used ride-hailing for non-commuting trips pre-pandemic [0/1]	NS	NS	0.750	2.555
Frequency model
Age (years)	−0.031	−12.925	−0.013	−1.833
Gender: female [0/1]	−0.387	−4.894	−0.447	−2.057
Person identifies as Black or African American [0/1]	0.438	2.670	NS	NS
Person identifies as Asian or Pacific Islander [0/1]	−0.498	−4.303	NS	NS
Education: bachelor’s degree or above [0/1]	0.159	1.752	NS	NS
Household income: $100,000 or more annually [0/1]	0.259	2.902	NS	NS
Home location: San Francisco Bay Area [0/1]	0.205	2.086	NS	NS
Home location: other region [0/1]	−0.329	−2.651	NS	NS
Characterization of home location: urban [0/1] a	0.345	3.883	0.867	3.890
Number of household vehicles: 0 [0/1]	1.053	8.092	0.572	1.755
Population density in home CBG (people per acre)	0.010	4.090	NS	NS
Employment density in home CBG (jobs per acre)	0.002	2.421	NS	NS
Land-use mix based on 8-tier employment entropy in home CBG	−0.424	−1.814	NS	NS
Road network density in home CBG (links/ sq. mi.)	NS	NS	0.020	1.615
Aggregate frequency of transit service per capita in home CBG (service frequency per capita)	0.492	2.083	NS	NS
National Walkability Index b	0.018	1.464	NS	NS
Pre-pandemic ride-hailing frequency for non-commuting trips: 1 to 3 times per month [0/1]	NS	NS	1.049	3.060
Pre-pandemic ride-hailing frequency for non-commuting trips: at least once per week [0/1]	NS	NS	2.634	5.711
Threshold parameters
Available but not used | Less than a day per month	−0.530	−2.237	1.125	2.238
Less than a day per month | 1 to 3 days per month	0.278	1.159	2.294	4.272
1 to 3 days per month | At least one day per week	1.831	7.300	4.045	6.756
Goodness-of-fit measures
Number of observations	4,124		630
Null loglikelihood	−7,298.84		−1,124.28
Final loglikelihood	−4955.80		−667.07
${\mathrm{\rho }}^2$ (Adjusted ${\mathrm{\rho }}^2$ )	0.321 (0.317)		0.407 (0.392)

Note: CBG = Census block group; [0/1] = binary variables; NS = parameter was not statistically significant.

a

Respondents were asked to choose between urban, suburban, small town, and rural area.

b

Values range from 1 (lowest) to 20 (highest).

Commuting Trips

The hurdle models corresponding to commuting trips were estimated based on responses from those who were employed and/or students at the time of the respective surveys. The specification of the availability component of the hurdle model for commuting trips suggests that older respondents and those from households earning less than $50,000 annually were less likely to regard ride-hailing as an available mode for commuting trips. The former echoes the findings of studies on ride-hailing adoption, which have consistently reported that ride-hailing use is more prevalent among younger individuals ( 1 ). The latter could stem from the relatively high cost of these services compared with other modes. In addition, respondents living outside of major metropolitan areas were less likely to regard ride-hailing as an available mode, while the opposite was true for those who indicated that they live in an urban area. This is likely because of the tendency for the availability of ride-hailing vehicles to be greater in denser and more populated areas.

Notably, the probability of ride-hailing being regarded as an available mode for commuting trips appears to differ based on how often an individual commutes. Specifically, respondents who worked from their primary workplace less than three days per week were more likely to regard ride-hailing as an available mode. This is likely because of the impact of the relatively high per-trip cost of ride-hailing on the viability of using these services on a frequent basis. In addition, the use of public transit and micromobility (i.e., bicycles and electric scooters) was found to influence whether ride-hailing was regarded as an available mode. Specifically, respondents who used these modes at least once per week were more likely to regard ride-hailing as an available mode. Similar results were observed among respondents who walked to work at least once per week, which could be related to public transit trips typically involving walking as an access or egress mode. Finally, the results of the panel respondents model suggest that respondents who used ride-hailing for commuting trips in the pre-pandemic period were also more likely to do so in the post-pandemic period.

The specification of the frequency component of the hurdle model sheds light on the determinants of ride-hailing frequency for commuting trips in the post-pandemic period. Similar to the findings of pre-pandemic studies, older and female respondents were less likely to use ride-hailing on a frequent basis. However, respondents who identified as Black or African American or Hispanic were more likely to use ride-hailing on a frequent basis, which contradicts the results of pre-pandemic studies such as Deka et al. ( 21 ) and Sikder ( 29 ). A possible explanation for this discrepancy is that the aforementioned studies did not consider how often ride-hailing is used for specific trip purposes.

In addition, living outside of a major metropolitan area was negatively associated with ride-hailing frequency, while living in an urban area was positively associated with ride-hailing frequency. Moreover, respondents from zero-vehicle households were more likely to use ride-hailing on a frequent basis for commuting trips. Ride-hailing frequency for commuting trips was also found to differ based on built environment and land-use attributes. For example, population and employment density were both associated with more frequent ride-hailing use. Transit service frequency was also positively associated with ride-hailing frequency, which is somewhat consistent with the findings of Shamshiripour et al. ( 18 ).

Notably, respondents who worked from their primary workplace at least once per week were more likely to use ride-hailing on a frequent basis. This is very likely because of these individuals making more commuting trips overall than those who work from their primary workplace on a less frequent basis. Finally, the results of the panel respondents model suggest that the use of ride-hailing for commuting trips at least once per month during the pre-pandemic period was positively associated with ride-hailing frequency in the post-pandemic period. This suggests that the frequency of using ride-hailing for commuting trips during the post-pandemic period may be influenced by how often these services were used during the pre-pandemic period.

Non-commuting Trips

The specification of the availability component of the hurdle model for non-commuting trips was largely consistent with that of the hurdle model for commuting trips (as summarized in Appendix B). One key difference was that respondents who identified as Hispanic were less likely to regard ride-hailing as an available mode for non-commuting trips. This is somewhat consistent with the work of Alemi et al. ( 84 ), who found that California residents who identified as Hispanic were less likely to be ride-hailing users. In addition, living in a more walkable area and an area with a greater density of roadway facilities both increased the probability of regarding ride-hailing as an available mode for non-commuting trips. These results could stem from the impacts of these attributes on the feasibility of using modes other than private vehicles. In addition, the use of car-sharing services for non-commuting trips was also associated with a greater likelihood of regarding ride-hailing as an available mode. This could stem from the potential for the adoption of car-sharing services to influence the adoption of ride-hailing services ( 25 ). Similar to the results of the commuting model, the results of the panel model suggest that those who regarded ride-hailing as an available mode for non-commuting trips in the pre-pandemic period were also more likely to do so during the post-pandemic period.

Several similarities between the specifications of the frequency components of the hurdle models for commuting and non-commuting trips were also observed. For example, older and female respondents were less likely to use ride-hailing on a frequent basis for non-commuting trips. In addition, respondents who identified as Black or African American were also more likely to use ride-hailing on a frequent basis. Although relatively few studies have noted differences in ride-hailing frequency based on race or ethnicity, this result differs from the findings of Sikder ( 29 ) who used data from the 2017 NHTS. In addition, living in an urban area and in areas with greater population or employment density increased the probability of an individual using ride-hailing on a frequent basis. Moreover, living in an area where transit service is provided on a more frequent basis was also associated with using ride-hailing on a more frequent basis.

Educational attainment and income were positively associated with ride-hailing frequency for non-commuting trips. In addition, living in the San Francisco Bay Area was found to increase the probability of using ride-hailing on a frequent basis for non-commuting trips. Furthermore, living in an area with a greater mix of land uses (defined based on employment entropy) was associated with lower ride-hailing frequency, which is consistent with the findings of other studies that have used data from California residents ( 19 ). Conversely, living in an area with greater walkability was associated with more frequent ride-hailing use for non-commuting trips. Finally, the results of the panel respondents model suggest that the use of ride-hailing at least once per month during the pre-pandemic period increased the probability of using ride-hailing on a frequent basis for non-commuting trips during the post-pandemic period. Interestingly, unlike the results for commuting trips, the use of ride-hailing on a weekly basis during the pre-pandemic period had a greater impact on ride-hailing frequency than the use of ride-hailing on a monthly basis.

Post-pandemic Shared Ride-Hailing Frequency

The final specification of the ZINB regression model is presented in Table 6. Similar to the hurdle models, variables pertaining to personal and household characteristics and built environment attributes were tested in the process of determining the final specification of the model. Moreover, the decision of whether to retain a variable in the model was made based on the sign and t-statistic of the corresponding parameter. Similar to the hurdle models, a variable was retained in the final specification of the model if it was statistically significant that the 90% confidence level. The final McFadden’s R² of the ZINB model was 0.116, which is indicative of an adequate goodness-of-fit. The distributions of the variables included in the final specifications of the ZINB regression model are presented in Appendix A.

Table 6.

Final Specification of the ZINB Regression Model for Shared Ride-Hailing Frequency

Variable	Estimate	t-stat.
Zero-inflation model
Constant	2.465	5.805
Age: 18 to 34 years old [0/1]	−1.138	−6.513
Age: 65 and older [0/1]	1.813	5.193
Gender: female [0/1]	1.364	8.099
Person identifies as White [0/1]	−0.987	−4.347
Person identifies as Black or African American [0/1]	−0.950	−2.605
Person identifies as Hispanic [0/1]	0.421	2.016
Education: bachelor’s degree or above [0/1]	−1.128	−5.686
Number of household vehicles: 0 [0/1]	−0.498	−1.951
Land-use mix based on 8-tier employment entropy in home CBG	0.674	1.629
Count model
Constant	−0.098	−0.326
Person is employed [0/1]	0.415	1.723
Household income: less than $50,000 annually [0/1]	0.767	2.992
Household income: $100,000 or more annually [0/1]	0.541	3.214
Number of household vehicles	−0.102	−1.392
Home location: other region [0/1]	−0.511	−2.115
Normalized proportional accessibility of regional destinations by public transit in home CBG a	1.864	1.900
$\mathrm{\alpha }$	0.529	2.242
Goodness-of-fit measures
Number of observations	2,697
Null loglikelihood	−1,286.28
Final loglikelihood	−1,137.00
McFadden’s R2	0.116

Note: ZINB = zero-inflated negative binomial; CBG = Census block group; [0/1] = binary variables.

a

Values were normalized by dividing each value by the maximum value in the sample.

The specification of the zero-inflation model suggests that older respondents were more likely to have never used shared ride-hailing services, while the opposite was true for younger respondents. This is consistent with the findings of studies on the adoption of shared ride-hailing services, such as Lavieri and Bhat ( 31 ) and Loa and Habib ( 24 ). Similarly, female respondents were more likely to indicate that they have never used shared ride-hailing, which contradicts the findings of Bansal and Kockelman ( 85 ). A possible explanation for this discrepancy is that the aforementioned study used data collected from respondents who reside in areas in the United States where ride-hailing services are offered, whereas this study only used data from residents of California. Conversely, individuals who identify as White or Black or African American were less likely to indicate that they have never used shared ride-hailing, which somewhat echoes the findings of Gehrke et al. ( 86 ) in their study of the Greater Boston Region. Similarly, educational attainment was negatively associated with the probability of an individual indicating that they have never used shared ride-hailing, which contradicts the findings reported by Lavieri and Bhat ( 31 ) in their study of the Dallas–Fort Worth metropolitan area.

Experience using shared ride-hailing also appears to differ based on household attributes. Specifically, those from households that do not own a vehicle were less likely to have never used shared ride-hailing. Conversely, living in an area with a greater mix of land uses increased the probability that an individual has never used shared ride-hailing. This result could be a reflection of the negative association between living in an area with a greater mix of land uses and ride-hailing frequency that has been reported in other studies that have used data from California residents ( 19 , 25 ).

The specification of the count model offers insights into the factors influencing the frequency of shared ride-hailing use. Employment was positively associated with the number of shared ride-hailing trips made by an individual, which is consistent with the results of Abouelela et al. ( 28 ). In addition, individuals from both lower- and higher-income households were more likely to have made more trips using shared ride-hailing. This is consistent with the bimodal distribution of the incomes of ride-hailing and taxi users that has been noted in previous studies ( 32 , 87 ). The positive association between belonging to a household that earns less than $50,000 annually and the number of trips made using shared ride-hailing likely stems from the relatively lower costs of shared services and ride-hailing allowing users to circumvent the costs associated with vehicle ownership ( 3 ). In addition, individuals living in areas with greater accessibility by transit were more likely to have made more shared ride-hailing trips. This finding could stem from the greater feasibility of using modes other than private vehicles in areas with greater accessibility by transit.

Discussion

The results of this study offer insights into the nature of ride-hailing use during the post-pandemic period and how it compares with that of the pre-pandemic period. Notably, the use of ride-hailing for commuting trips appears to be more prevalent during the post-pandemic period than it was during the pre-pandemic period. Moreover, the results suggest that how often a person works from their primary workplace is positively associated with how often they use ride-hailing for commuting trips. Amid the implementation of return-to-office (RTO) mandates, it remains to be seen whether ride-hailing use for commuting trips will increase if those with hybrid work arrangements are required to commute more often. Based on the results of the hurdle model for commuting trips, those who worked from their primary workplace less than three days per week were more likely to regard ride-hailing as an available mode for their commuting trips. However, anecdotal evidence suggests that the enforcement of RTO policies may have influenced certain workers to use ride-hailing for some of their commuting trips ( 88 ). In addition, programs that allow ride-hailing users to temporarily avoid surge pricing (such as the price lock program offered by Lyft [ 89 ]) could contribute to increases in the number of ride-hailing trips made during peak commuting hours. Such increases would contribute to increases in VMT, which could worsen congestion and emissions (particularly in dense urban areas). In light of these trends, planning agencies should monitor traffic conditions, ride-hailing trip volumes, and the performance of public transit routes in areas with a relatively high density of jobs. This information can help identify areas that could benefit from the implementation of congestion management strategies or transit priority measures.

The results of this study also have implications for initiatives that aim to address the negative externalities associated with ride-hailing. In particular, the results suggest that the uptake of shared ride-hailing services is relatively low during the post-pandemic period. Given that the utilization of shared ride-hailing influences the extent to which these services can address the negative externalities associated with ride-hailing ( 90 ), it is imperative to explore the reasons behind the reluctance to use shared ride-hailing in the post-pandemic period. Prior studies have noted that the use of these services can potentially be influenced by factors such as privacy concerns ( 31 , 91 ), perceptions of risk ( 92 , 93 ), concern for the environment ( 94 ), the desire for convenience ( 91 ), and the discount associated with these services ( 95 , 96 ). Moreover, in light of the results of the ZINB regression model, it is possible that the factors influencing the reluctance to use shared ride-hailing differ based on sociodemographic characteristics. This information can help inform planning and policy efforts to increase the willingness to use these services, which could reduce the distance traveled by ride-hailing vehicles while not serving passengers (i.e., deadheading) ( 1 ). Moreover, understanding the barriers to the use of shared ride-hailing services can help inform strategies to encourage the use of pooled automated vehicles and shuttles if they were to become available in the market.

Moreover, the results underscore the need to pursue other approaches to mitigate the negative externalities associated with ride-hailing services. For example, policies that aim to encourage the electrification of ride-hailing fleets (such as the Clean Miles Standard program in California) can reduce the tailpipe emissions associated with ride-hailing. However, it is important to note that ride-hailing services that are delivered using electric vehicles will still contribute to VMT, which is positively associated with congestion and emissions from gasoline-powered vehicles ( 97 ). The contributions of ride-hailing to VMT could be mitigated through initiatives and policies that aim to encourage users to travel shorter distances. In the near term, improving the integration of ride-hailing services with public transit (such as offering discounts to those who use ride-hailing as an access mode) could contribute to reducing the length of ride-hailing trips. In the longer term, policies that encourage dense and mixed-use development in areas that are well-served by public transit can help reduce the need to travel longer distances.

Finally, the results suggest that the determinants of post-pandemic ride-hailing frequency are relatively consistent with those identified by studies using data collected during the pre-pandemic period. While this finding may seem somewhat trivial, the results of studies published using data collected during the early stages of the pandemic suggest that short-term changes in the determinants of travel behavior may have occurred. For example, Loa et al. ( 59 ) found that the impacts of factors such as age, income, and vehicle ownership on ride-hailing frequency during the early stages of the pandemic in the Greater Toronto Area differed from the impacts identified in pre-pandemic studies. In addition, the results of Abdullah et al. ( 98 ) suggest that factors that typically influence mode choice decisions (such as travel time and cost) were less important during the early stages of the pandemic than they were during the pre-pandemic period. Despite the similarity in the determinants of ride-hailing frequency, publicly available data on ride-hailing trip volumes from New York City and Chicago suggest that post-pandemic ride-hailing use has yet to reach the highs observed during the pre-pandemic period, requiring continued research on this topic.

Conclusion

This study presented the results of an investigation into the determinants of ride-hailing frequency during the post-pandemic period and a comparison of pre- and post-pandemic ride-hailing use. Data from two waves of a repeated cross-sectional survey of California residents were used for descriptive analysis and to estimate two hurdle models and a ZINB regression model. The results suggest that the use of ride-hailing for commuting trips is more common during the post-pandemic period than it was during the pre-pandemic period. Moreover, the results of the hurdle model for commuting trips suggest that the frequency of using ride-hailing for commuting is positively associated with how often an individual commutes. This suggests that RTO policies could affect the number of ride-hailing trips that are made during peak commuting hours. Despite evidence of peak spreading during the post-pandemic period ( 99 ), increases in the number of ride-hailing trips made during these periods have the potential to worsen both congestion and emissions. The results also shed light on the relatively low uptake of shared ride-hailing services during the post-pandemic period. This suggests that efforts to mitigate the negative externalities associated with ride-hailing should consider options aside from encouraging the use of shared ride-hailing, such as the electrification of the ride-hailing fleet. However, it will be crucial to explore methods to promote the use of shared ride-hailing in the post-pandemic era, as it can help reduce the distance traveled by ride-hailing vehicles both with and without passengers.

While the results of this study offer insights into the use of ride-hailing in the post-pandemic period and its determinants, there are several notable limitations. First, changes in the response options that were used to understand how often respondents used a given mode for commuting and non-commuting trips precludes a more detailed comparison of pre- and post-pandemic ride-hailing frequency. Second, information concerning the use of shared ride-hailing services during the post-pandemic period was limited to the month before the respondents completed the survey. This information could have been used to identify the attributes of individuals who were more likely to have used these services after they were reintroduced following the pandemic. Third, despite the application of weights to the fall 2020 and fall 2023 samples, certain demographic groups (e.g., younger adults, those who did not attend college or university, and those who identify as Hispanic) remained under-represented in the weighted samples. This has the potential to adversely affect the extent to which the results of the study can be generalized to the population of California, and in particular, the generalizability of the results to members of segments of the population that were under-represented in the weighted datasets. Finally, the fall 2020 wave of the CMP survey did not include questions pertaining to the use of shared ride-hailing services before the pandemic. This information could have facilitated a comparison of the uptake of shared ride-hailing during the pre- and post-pandemic periods.

There are several approaches that future studies could take to build on the work presented in this paper. First, future studies should investigate the factors influencing the decision to ride-hailing for commuting trips during the post-pandemic period. For example, exploring the motivations for using ride-hailing can shed light on whether the use of these services was driven more by factors such as flexible work schedules or the shortcomings of other modes of travel. Moreover, understanding the areas where those who commute via ride-hailing services live and work can offer insights into the potential impacts of increases in ride-hailing trips made during commuting hours. Second, studies that use data containing information on how often ride-hailing was used by the same set of respondents during the pre- and post-pandemic periods should consider applying temporal shift analysis to examine changes in the determinants of ride-hailing frequency. Because of differences in the response options that were used to collect information on how often various modes of travel were used, the application of temporal shift analysis was not possible in this study. Finally, future studies should also explore the factors influencing the reluctance to use shared ride-hailing during the post-pandemic period. In particular, future studies should examine barriers to the use of shared ride-hailing and the factors influencing the reluctance to use these services. This information can help inform policy interventions aiming to encourage shared ride-hailing use and offer insights into whether it is possible for such initiatives to influence the use of these services.

Supplemental Material

sj-docx-1-trr-10.1177_03611981251380589 – Supplemental material for Examining Post-pandemic Ride-Hailing Use in California and Its Determinants

Supplemental material, sj-docx-1-trr-10.1177_03611981251380589 for Examining Post-pandemic Ride-Hailing Use in California and Its Determinants by Patrick Loa, Yongsung Lee and Giovanni Circella in Transportation Research Record