What Makes the Route More Traveled? Optimizing U.S. Suburban Microtransit for Sustainable Mobility

Abstract

Microtransit services that provide pooled on-demand transportation with dynamic routing have been used in low-density areas since the 1970s, but improvements to routing technology have led to a resurgence of interest in the past decade. Questions remain about the effectiveness of microtransit to serve riders in low-density, car-dependent suburban areas. Better understanding of the factors underlying microtransit ridership can improve usage of these services and shift travelers to more sustainable modes in suburban areas. We compile a database of suburban microtransit programs from 32 public transit agencies in the U.S. to study internal factors (e.g., operating hours, service area) and external factors (e.g., population density, vehicle ownership) impacting ridership using a random effects model. We find that internal agency factors have a greater effect on microtransit ridership than external factors. The most impactful factor is operating a point deviation service, where vehicles have scheduled stops at one or more checkpoints within the service area (e.g., transit center or shopping center), rather than zone-based services, where vehicles pick up and drop off passengers at any time within a service area. There is high potential to convert some zone-based services to point deviation services; 52% of zone-based service areas contain a transit center that could be used as a checkpoint. For the remaining zone-based service areas, maximizing ridership may not be feasible, and using ridership as an evaluation metric can be misleading. Instead, metrics that capture the accessibility, safety, or customer satisfaction impacts of microtransit may be more appropriate for these services.

1. Introduction

Public transit is easier to operate in denser urban areas due to a higher density of people, employment, and amenities, allowing transit agencies to design routes connecting population centers and activity generators. Accordingly, across the extensive literature on factors influencing public transit ridership, the variables that explain the greatest variation in transit ridership are environmental factors outside of transit agency control, such as car access, household income, population and employment density, and traffic congestion [1]. In the suburbs, public transit performance has historically been poor, which coincides with high levels of vehicle ownership, abundant free parking, and low population and employment density. Compared to urban public transit systems, suburban transit services have a lower farebox recovery ratio, require a higher subsidy per passenger, and serve fewer passengers per mile [2]. Meanwhile, suburbs continue to grow in population, presenting an issue for transportation sustainability in the auto-oriented suburban environment. In response to the difficulties of operating public transit in suburban areas, a growing suburban population, and need to provide more sustainable transportation options, general public demand response transit (DRT), or microtransit, has been proposed as a mobility strategy.

General public DRT as a concept is not new. Currie and Fournier (2020) [3] identify three distinct periods of DRT development tracing back to the 1970s: (1) early DRT “Dial-a-Bus” services (1970–1984), (2) Americans with Disabilities Act (ADA) paratransit (1985–2009), and (3) information and communications technology (ICT) DRTs (2010–2019). This latest wave of ICT DRTs coined the term “microtransit”, defined by SAE International as a “technology-enabled transit service that typically uses shuttles or vans to provide pooled on-demand transportation with dynamic routing” [4]. Microtransit is the term we use in this paper to refer to all types of DRT services available to the general public.

Interest in microtransit has been growing in the public sector. In 2019, there were about 60 on-demand partnerships between public and private agencies in the U.S., including urban, suburban, and rural areas. This also includes numerous partnerships between microtransit companies and companies such as Uber and Lyft [5]. In our research, we identified 144 microtransit programs operating in 2022 in U.S. suburban areas alone. Many microtransit programs in the U.S. are currently situated in low-density suburban or rural areas, as public transit agencies hope that microtransit can provide transit access for residents at a lower operating cost and with higher customer satisfaction compared to traditional fixed-route transit (FRT) [6]. Furthermore, a growing number of low-income suburban households with inconsistent or no vehicle access highlights the need for transportation alternatives in suburban areas. When using public transit in suburban areas, low-income residents experience long travel times, especially when traveling between two suburbs, as public transit systems have been designed to optimize trips between suburbs and central cities [7] Low-income individuals are also more likely to travel at off-peak times when fixed-route transit has limited availability [8]. Microtransit services that operate on more flexible schedules in defined zones that serve trips within suburbs are a potential strategy to address these challenges for low-income suburban residents.

As the number of microtransit programs in the U.S. expands, there is uncertainty about the effectiveness of these services and the extent to which microtransit is a viable alternative to FRT in suburban areas, in particular to serve the transportation needs of low-income residents and provide more sustainable transportation options for suburban residents. In car-dominant suburbs, microtransit may provide a more sustainable transportation option, and increase public transit access for low-income households with no or limited access to a personal vehicle. However, reviews of microtransit programs in the U.S. have shown low ridership and high costs [6,9,10]. There has not been a comprehensive, quantitative evaluation of the factors impacting microtransit ridership and what public transit agencies can do to improve ridership of microtransit services. Addressing this issue could be an important step to improve public transit access in suburban areas and transportation sustainability by providing an alternative to private vehicle travel in the suburbs. Specifically, our research questions include:

(1)

What are the characteristics of suburban microtransit programs in the U.S. (e.g., service area demographics, operating characteristics, service hours, etc.)?

(2)

What are the main factors influencing suburban microtransit ridership?

(3)

How can public transit agencies optimize service characteristics to improve microtransit services?

Literature Review

The factors affecting public transit ridership have been studied extensively and have typically been separated into “external” factors, or factors outside of the transit agency’s control, and “internal” factors. A 2013 review of the public transit ridership literature found that variation in transit ridership across agencies in the U.S. is mostly explained by “external” factors, including employment and population density, car ownership, parking and gas prices, and median household income [1]. “Internal” factors have comparatively less impact on ridership although they are still important to increase public transit ridership. In general, improvements to service quality (e.g., reducing delay time, increasing service frequency) have a greater impact on ridership than fare changes [1].

More recent research has instead found that internal factors, such as vehicle revenue hours or vehicle revenue miles, are a main contributor to variation in transit ridership across public agencies in the U.S. and Canada. Boisjoly et al. (2018) conclude that public transit agencies can increase ridership through increasing investment in transit operations; however, this study does not account for the potential endogeneity of transit supply and demand, which may affect the relative impact of internal and external agency factors on ridership [11,12]. Diab et al. (2020) do account for this endogeneity and found that both internal (e.g., vehicle revenue hours) and external factors (e.g., size and population of the metropolitan area, car ownership) affect ridership [12]. Meanwhile, research in Southern California, a region dominated by sprawl and an auto-oriented built environment, shows the continued significance of vehicle ownership on transit ridership. Increased auto access among low-income households, the demographic most likely to ride public transit, has significantly contributed to the decline in transit ridership across Southern California since 2013 [13].

The relationship between built environment factors and public transit use has been explored by many researchers, and a meta-analysis of these studies found that land use mix, pedestrian connectivity, and jobs–housing balance have the highest relative impact on transit ridership [14]. In suburban areas that lack these built environment characteristics, general public DRT has been proposed as an alternative. Surveys from the 1990s indicated that a meaningful number of suburban agencies adopted some form of general public DRT in low-density suburban neighborhoods [2,15]. A more recent survey of agencies in 2019 indicates that this is still the case, and operators are continuing to use general public DRT, now commonly referred to as microtransit, to retain or provide transit service in areas and times of low demand [6].

There have been a number of studies on the factors influencing ridership and demand for DRT, with recent studies focusing on ADA paratransit and rural areas [16,17,18]. Early studies of DRT services also estimated demand using regression modeling and ridership surveys [19,20,21]. Authors found that DRT demand was influenced by fare, trip time, and population density of the service area [19]. A review of eight DRT implementations in the U.S. showed the potential of some DRT services to attain higher ridership than fixed-route services, though these results varied significantly between public transit agencies [21].

Recent studies using regression modeling to study determinants of suburban microtransit ridership have been conducted as case studies of a single transit agency or single service area [22,23,24]. Overall, these studies found that DRT demand and ridership were higher with higher levels of service (i.e., more vehicles in the DRT fleet) and in places with a low population density, higher percentage of elderly and people with disabilities, and low vehicle ownership. The call-n-Ride performance review, conducted by the Denver Regional Transit District (RTD), also investigated the impact of service design on ridership. The RTD operates three main service types: (1) simple DRT, where vehicles have no established route or scheduled stops; (2) point deviation, where vehicles pick up and drop off passengers on demand while also arriving at designated checkpoints at scheduled times; and (3) route deviation, where vehicles have a designated route and stops but may deviate to pick up or drop off passengers up to a certain distance away from the route. The RTD found that more structured services (i.e., point or route deviation) have higher ridership than simple DRT services.

Simulation studies have also compared traditional fixed-route transit (FRT) and microtransit in suburban settings. Overall, these studies have shown that microtransit is more effective than FRT to minimize travel distances, perceived travel time, and operational costs when demand density is low [25,26,27]. An additional benefit of microtransit is its flexibility, making microtransit better suited than FRT to handling changes in passenger demand throughout the day [26]. A more extensive review of DRT simulation studies is found in [28].

There have been a number of real-world microtransit program evaluations; we focus on U.S. programs given our research scope. A review of 18 microtransit programs in the U.S. found that ridership per hour varied between 0.4 to four rides per hour with an average of 2.7 rides per hour [10]. Two other microtransit evaluations, using case study analysis and a survey of public transit agencies, also found microtransit ridership is low, with a ridership between three and four passengers per hour, and the expense to the agency is high [6,9]. However, an evaluation of a microtransit pilot operated through Dallas Area Rapid Transit found that the cost of microtransit was competitive with low-ridership fixed-route buses operating in the region [29].

The high cost of implementing microtransit programs has led to a high failure rate of these programs. Researchers compiled a database of 120 microtransit systems across 19 countries and found that 48% of programs failed between 1970 and 2019 [3]. Meanwhile, microtransit costs have not decreased in the new era of technology-enabled systems in the 2010s [3]. However, the success of some microtransit programs can also be revealing. The Denver Regional Transit District (RTD) has operated some form of general public DRT since the 1990s. The RTD carefully chooses where to deploy DRT based on service area characteristics and is able to maintain low costs by strictly controlling where and how to operate the DRT service. The RTD found that more structured DRT services, with service zones based around scheduled pick up/drop off locations, can generate higher productivity than pure demand response operation [30]. A report surveying 22 public transit agencies similarly found that the most efficient way for DRT to operate is by loading as many passengers as possible at designated “checkpoints” [6].

While simulations show the promise of microtransit to provide a better service than FRT in low-density areas, evaluations of real-world applications have shown mixed results and high failure rates. There have been some studies analyzing factors impacting ridership and demand for DRT, but to our knowledge, there is no comprehensive study examining specifically general public DRT services across multiple agencies. While studies of ADA paratransit and rural DRT exist, these services are specific use cases and have unique problems to general public DRT in suburban areas. ADA paratransit services serve only qualifying passengers with disabilities and have longer dwell times due to the time needed to provide a door-to-door service and attend to the specific needs of passengers. Rural services have a much lower population density and demand than suburban services. Thus, our research aims to fill this literature gap with an analysis of microtransit ridership across the suburban agencies in the U.S., comparing the relative influence of factors internal and external to public transit agencies.

2. Materials and Methods

In this research, we developed a dataset of ridership, internal, and external program characteristics for 32 microtransit programs operating in the U.S., as of 2022. We used these data to perform a linear random effects analysis of the relationship between internal and external characteristics and microtransit ridership. In this section, we detail our data collection process, model selection and specification, and limitations and potential omitted variables.

2.1. Data Collection

We compiled a list of active microtransit programs in the U.S. using the National Transit Database (NTD) from the Federal Transit Administration. We searched for Full Reporter agencies that have a “Demand Response” mode, which includes microtransit services along with ADA paratransit and agency partnerships with taxis or companies, such as Uber and Lyft. (Full Reporter agencies are defined by the FTA as agencies with at least 30 vehicles operating during maximum service (VOMS). Agencies that are not Full Reporter agencies are smaller and generally rural transit agencies. We exclude these agencies from our analysis given our focus on suburban areas). To filter out only those agencies with general public microtransit, we referenced transit agency websites and planning documents. We identified 144 total microtransit programs that were operational in 2022.

For each of these agencies, we searched through public planning documents, such as annual reports and board meeting minutes, for 2022 ridership data. For programs without published ridership data, we contacted representatives from the agency for ridership data. We were able to obtain microtransit ridership data from 32 agencies (Figure 1). All programs with ridership data were in the Western, Midwest, and Southern regions of the U.S. While there are several microtransit programs in the Northeast, we were not able to obtain ridership data for these programs. We used the population density of the microtransit service areas to verify that programs were in suburban areas and excluded programs with a population density of less than 500 people per square mile (i.e., rural programs).

Microtransit programs often have multiple service areas, or zones, within which a small number of microtransit vehicles operate. Riders are typically able to travel using microtransit within a zone but not between two zones. Figure 2 shows an example of two different microtransit programs. In 2022, CapMetro Pickup (left), operated by the Capital Metropolitan Transportation Authority in Austin, Texas, had a total of 10 microtransit zones throughout the greater Austin metropolitan area. Area Regional Transit On-Demand (right), operating in St. Lucie County, Florida, had only one microtransit zone.

With many zones spread out across a metropolitan area, we expect zones to have different built environment characteristics, population demographics, and transportation accessibility. Thus, we conducted our analysis at the zone level when possible. Of the 32 agencies with ridership data, 17 agencies had ridership data by zone, eight agencies reported ridership aggregated across multiple zones, and seven agencies operated only one microtransit zone. Our final dataset size included 145 microtransit programs or zones with ridership data.

We collected data on internal and external factors for each microtransit zone. For programs with ridership data at the zone level, we collected these data for each zone, and for programs that provided only aggregate program-level ridership data, we took the average or sum of variables across all zones in the microtransit program service area, as appropriate. To obtain data on internal factors, we searched public agency websites and news articles. We also categorized programs as either many-to-many “zone-based” or “point deviation” services organized around checkpoints based on the description of the microtransit service provided by agencies on their website or in planning documents. These service types are summarized in

Table 1. Description of microtransit service types.

Service Type	N	Description
Zone-based	106	Many-to-many (i.e., many origins, many destinations) services that pick up and drop off passengers anywhere within a specified zone
Point deviation	39	Same as zone services, except with a scheduled departure time at a specific location within the zone (e.g., scheduled departure from Walmart every 60 min)

Source: own compilation based on descriptions of microtransit services provided by public transit agencies on websites or in planning documents.

.

Examples of the difference between zone-based and point deviation services are shown in Figure 3. On the left is an example of a typical zone-based microtransit service where passengers can be picked up and dropped off anywhere within the zone. On the right is an example of a point deviation service where passengers can also be picked up and dropped off anywhere within the zone, but there is also a scheduled departure every hour from the Walmart shopping center.

To collect data on external factors impacting ridership, we first digitized microtransit service area maps using ArcGIS. We then averaged the 2021 American Community Service (ACS) Five-Year estimate data for each variable over all Census block groups that overlapped with the microtransit service area. For the variable “# of points of interest (POIs) per square mile”, we used the OSMnx package in python to retrieve points of interest within the microtransit service area [31]. POIs included places such as restaurants, schools, healthcare facilities, fitness centers/gyms, retail shopping, and grocery stores. We clustered POIs using the DBSCAN algorithm, as we considered many densely populated POIs to be a shopping center, strip mall, or other similar location to count as one major POI, rather than many discrete POIs. To measure the coverage of existing, fixed-route public transit services in the area, we used Google Transit Feed Specification (GTFS) data to identify public transit stops within the service area. We then drew a 400-meter buffer around each public transit stop and divided the total area covered by public transit stops by the total area of the service zone. A 400 m or roughly quarter mile buffer is commonly used as the “walkshed” or distance that most people are willing to walk to reach a public transit stop. A public transit coverage value of 0 indicates that there are no public transit stops in the service zone, while a value of 1 indicates that the entire service zone is within 400 m of a public transit stop.

The summary and description of all variables is provided in

Table 2. Summary and description of variables considered for model specification.

Variable	Source	Variable Description	Mean	St. Dev.	Min	Max
Dependent variables
Total passengers, 2022	Agency website or personal communication	Total passengers in 2022	27,904.7	73,524.5	291	615,975
Passengers per capita, 2022	Agency website or personal communication	Total passengers in 2022/service area population	0.7	1.6	0.03	15.5
Independent variables
Internal factors
Program launch year	Agency website	Year that program began service	2016	4.925	2001	2022
Hours of operation per week	Agency website	Hours of operation per week	77.4	19.2	20	122
Fare	Agency website	Fare	2.2	0.9	0.0	5.0
Service area	Agency website	Area of microtransit zone (square miles)	17.2	26.6	0.7	200.3
Service type dummy [1: point deviation]	Agency website	Dummy variable for point deviation service type 1 if point deviation, 0 otherwise	0.3	0.4	0	1
External factors
Population density	2021 ACS Five-Year Estimates	Total population/geographic area of Census block groups	3439.1	2228.5	638.5	15,279.8
Average median income	2021 ACS	Average median income of Census block groups contained in zone	86,225.1	28,577.0	37,021.4	177,478.2
% households in poverty	2021 ACS	Population under poverty line/total population	0.05	0.04	0.01	0.3
% zero-vehicle households	2021 ACS	Number of zero-vehicle households/total households	0.08	0.06	0.00	0.3
% college graduates	2021 ACS	Number of college graduates/total population	0.4	0.2	0.03	0.8
% over 65	2021 ACS	Number of residents over age 65/total population	0.1	0.06	0.03	0.4
% African American	2021 ACS	Number of African American residents/total population	0.1	0.1	0.00	0.7
% public transit commuters	2021 ACS	Number of public transit commuters/total population of employed workers	0.02	0.03	0.00	0.2
Employment density	EPA Smart Location Database	Number of jobs/geographic area	2308.7	2865.2	79.7	17,553.4
# POIs per square mile	OpenStreetMap	Number of points of interest contained in microtransit zone area + 250 m buffer/microtransit zone area	5.1	4.6	0.2	32.9
Public transit coverage	GTFS	Sum of 400 m buffer areas around public transit stops/microtransit zone area	0.28	0.22	0.0	0.94

Data sources: 2021 ACS data from https://data.census.gov/, EPA Smart Location Database from https://www.epa.gov/smartgrowth/smart-location-mapping, GTFS from https://mobilitydatabase.org/. OpenStreetMap accessed through the OSMnx Python package. Sources accessed 1 August 2023.

. Dependent variables include total passengers and passengers per capita in 2022. We chose independent variables based on our literature review of the most important factors influencing public transit ridership. We include hours of operation per week as an independent variable, although prior research has noted the issue of endogeneity between transit supply (e.g., vehicle revenue hours) and ridership (i.e., that transit supply and ridership are correlated) [32]. As ridership increases, public agencies may accordingly increase transit supply, which further increases ridership, and so on. However, since most of the microtransit programs in our sample have relatively few years of service, we assume that operating hours are determined independently of ridership, i.e., agencies have not yet adjusted operating hours in response to ridership levels. For service types, we coded a dummy variable for point deviation services compared to zone-based services.

2.2. Model Specification

Our dataset is comprised of a mix of zone-level and agency-level ridership for suburban microtransit programs in the U.S. We explore the relationship between internal and external agency factors and microtransit ridership using a linear random effects regression model. Other studies of public transit ridership have used ordinary least squares (OLS) regression to identify the significant factors influencing ridership [33,34,35]. A key assumption of OLS regression is that observations are independent; however, our dataset includes zone-level ridership observations within the same public agency. Ridership observations within an agency are likely correlated, as each public agency may choose to operate microtransit services differently in ways that are unobserved. The random effects model assumes that the error term for each cluster, in this case, public agencies, is drawn from a random distribution, and as such, observations within a cluster are no longer treated as independent. The random effects model is summarized as:

$$y_{i,j}={\beta }_0+{\beta }_1{X}_{i,j}+{\mu }_i+{\epsilon }_{i,j}$$

(1)

where $y_{i,j}$ is the total ridership or ridership per capita for microtransit zone j under public transit agency i, $X_{i,j}$ is a vector of independent variables for zone j under agency i, ${\mu }_i$ is the agency-specific error term, and ${\epsilon }_{i,j}$ is the stochastic error. In contrast to OLS regression, this model includes the error term ${\mu }_i$, which varies for each public agency.

Another strategy to address the correlation between ridership within agencies is to include a dummy variable for each public agency as a fixed effect. However, we decided against this approach for a number of reasons. First, we are interested in the overall variance in ridership associated with the public agency, rather than the effect of each individual public agency on microtransit ridership. Second, our dataset is a sample of agencies that have microtransit programs rather than an exhaustive list of all microtransit programs in the U.S. Therefore, we have chosen to use a random effects model for our analysis.

We used the lme4 package in R to perform the random effects model [36]. We transformed highly skewed variables using the natural logarithm function to obtain a normal distribution. We also scaled and centered all variables to have a mean of 0 and standard deviation of 1. We arrived at the final model specification following the procedure in Zuur et al. (2009) [37], which recommends tuning the random effects part of the model using all possible covariates (summarized in Table 2) and interaction terms, then adjusting the fixed effects part of the model by systematically removing highly correlated and insignificant terms. We identified correlated terms through a correlation matrix and insignificant terms using likelihood ratio tests.

Random effects models can have either random intercepts (i.e., allows the intercept to vary for each agency) or random intercepts and slopes (i.e., allows both intercept and slope to vary for each agency). To select the best performing model, we used the Akaike information criteria (AIC) statistic, which is a relative measure of model quality. Using the transit agency as the random effect, we compared the AIC between random intercept and random intercept and slope and found that the random intercept had a better model fit.

2.3. Limitations and Potential Omitted Variables

There were some limitations to the data available for this study. We were only able to obtain zone-level ridership for about half of the microtransit programs and had to take average values for independent variables of the microtransit programs without zone-level data. We tested the model using a mixture of zone-level and agency-level data against a model using only zone-level data and did not find significant differences. Thus, we decided to keep both zone- and agency-level data in this analysis due to the larger sample size. There were other variables that we would have liked to include in the model, but the data were not consistently available for all programs. For example, more direct measures of service supply, such as number of vehicles or vehicle revenue hours, could have captured the productivity and efficiency of microtransit services. The number of vehicles used in each microtransit zone also likely affects total ridership. However, these data were not consistently reported for all programs. Cost per passenger or cost per hour is also a frequently used measure of operational efficiency. However, we were only able to find cost data for 14 of the 32 agencies, and none of the agencies reported costs at the zone level. We summarize these cost data in the results, but we do not use the data for further modeling.

3. Results

In this section, we summarize the general characteristics of the suburban microtransit programs in our dataset, including the demographics of microtransit service areas and key operating characteristics such as service hours, off-peak and weekend service availability, and average wait time. We then summarize the results from random effects models of microtransit ridership and microtransit ridership per capita.

3.1. Suburban Microtransit Characteristics

To assess the extent to which microtransit can serve suburban transportation needs, we examined microtransit service area demographics and service characteristics, specifically off-peak and weekend service availability and average wait time.

3.1.1. Microtransit Service Area Demographics

The average median income across microtransit service areas in our dataset was USD 86,225, which is higher than the median household income in the U.S. (USD 69,021 from 2021 ACS Five-Year Estimates). The average poverty rate across microtransit service areas was 8%, lower than the U.S. average (11.5% from 2021 ACS Five-Year Estimates). This indicates that most suburban microtransit programs are not operating in low-income neighborhoods.

A study of microtransit user demographics across 14 public agencies in the U.S. also found that microtransit service areas had a higher income than the U.S. average [38]. However, survey respondents from these 14 microtransit programs had a lower income than both the microtransit service area and the U.S., with 60.9% of survey respondents having an annual household income less than USD 50,000 (compared to 27.9% of households in the service area and 36.5% of households in the U.S.). Another survey of microtransit riders in Los Angeles found that 46.8% of riders have an income less than USD 50,000, though this study did not report on the average income within the microtransit service area [39]. These findings from rider surveys suggest that while the median income of microtransit service areas is higher than the U.S. average, users within these service areas tend to have a lower income. The disparity between service area demographics and rider survey demographics points to a limitation of our analysis, and microtransit riders may be more diverse than the service areas where microtransit operates.

3.1.2. Microtransit Service Characteristics

In less dense suburban areas where fixed-route public transit is more difficult to operate, microtransit may offer an improved customer experience with better wait times. Suburban public transit services often have headways of 30 or 60 min and limited service on weekends. While we were not able to collect actual wait time data from suburban microtransit programs, we looked at estimated wait times published on agency websites. For example, the microtransit program website for LA Metro states that the maximum wait time from time of reservation to time of pickup is 15 min. We were able to find estimated wait times for 20 programs, and the average wait time was 14 min. This wait time is much shorter than the expected wait times for public transit in suburban areas where headways are often 30 min or even 60 min or longer.

In addition, about half of microtransit service areas have service hours before the morning peak (i.e., before 7 am) and after the evening peak (i.e., after 7 p.m.), and 44% have service on weekends. This indicates the potential of microtransit to provide service during times of day and days of the week when low-income residents are more likely to travel. Off-peak periods and weekends are also times when fixed-route public transit service is more limited with longer headways, and the shorter wait times of microtransit represent an additional service improvement compared to fixed-route transit.

High public transit operating costs in suburban areas is also a concern but one that microtransit may not be able to address. Using NTD data from 2021, we found that the cost per bus trip for suburban agencies ranged from USD 6.05 to USD 38.25, with an average cost per trip of USD 17.05. Meanwhile, the cost per suburban microtransit trip in the U.S. ranged between USD 7.57 and USD 68.60 per trip, with an average of USD 44.30. Thus, the cost-saving potential of microtransit is an area for further research. However, given that microtransit can provide better off-peak, weekend, and local accessibility for low-income suburban residents compared to existing fixed-route services, higher service costs may be justified. Metrics for assessing microtransit beyond cost per passenger is also an area for future research.

3.2. Findings from Microtransit Ridership Model

To further investigate opportunities to improve microtransit service, we used a random effects model to examine the impact of internal (e.g., operating hours) and external (e.g., population density) agency factors on suburban microtransit ridership. The results from the final model specification for total 2022 ridership is shown in

Table 3. Final model specification for 2022 total ridership.

	Total Ridership, 2022 n = 145 Microtransit Service Zones Operated by 32 Public Transit Agencies
Predictors	Estimates	CI	p
(Intercept)	0.04	−0.32–0.39	0.840
Internal factors
Hours of operation/week	0.40	0.27–0.53	<0.001
Service type [point deviation]	0.69	0.27–1.11	0.001
Service area (sq mile)	0.45	0.30–0.60	<0.001
External factors
Population density (persons/sq mile)	0.06	−0.08–0.20	0.416
% African American	0.11	−0.01–0.22	0.075
% Public transit commuters	−0.12	−0.24–0.00	0.057
# POIs per sq mile	0.02	−0.16–0.20	0.835
Employment density (jobs/sq mile)	0.11	−0.04–0.26	0.148
Random Effects
Residual	0.30
Agency-specific	0.75
ICC	0.72
Nagency	32
Observations	145
Marginal R2/Conditional R2	0.304/0.803

Bold p-values indicate significance at the p < 0.05 level. Source: own compilation, based on model output of lme4 package in R version 4.2.0.

. The final conditional R² value of 0.803 indicates a solid model fit. For each covariate, p-values were obtained through likelihood ratio tests of the full model against a model without that covariate. For random effects, our model finds that the intraclass correlation coefficient (ICC) is 0.72, indicating high correlation between multiple ridership observations within a public agency. Comparing the marginal R² (only considers fixed effects; R² = 0.304), and the conditional R² (considers fixed and random effects; R² = 0.803) shows that including random effects greatly increases the model fit. Thus, our choice of the random effects model is validated.

In the next subsections, we summarize findings on the impacts of internal and external factors on suburban microtransit ridership.

3.2.1. Impacts of Internal Agency Factors on Microtransit Ridership

All internal agency factors (hours of operation per week, service type (point deviation or zone-based), and service area) were found to be significant in the model and had positive effects on ridership. Increasing the hours of operation per week and service area increases the potential number of riders that could be served, which results in higher ridership. In terms of service type, point deviation services had overall higher ridership compared to zone-based systems. Point deviation services operate by having a scheduled departure time at a specific location, which is different from zone-based services where riders can be picked up and dropped off at any location in the zone at any time. Thus, point deviation services are more similar to fixed-route public transit, as the service picks up many riders at the same time who have congregated at a pickup point, explaining the higher ridership compared to zone-based services.

3.2.2. Impacts of External Agency Factors on Microtransit Ridership

Contrary to the prior literature on determinants of public transit ridership, external agency characteristics did not have strong relationships with ridership. The only significant variable was the percentage of public transit commuters within the microtransit service area, which exhibited a negative relationship with microtransit ridership. One potential implication of this finding is that there is minimal overlap of riders of traditional public transit modes (e.g., subway, public bus) and microtransit passengers. The other potential implication is that microtransit passengers may not be using the service to connect to other public transit modes, but rather they are using microtransit in isolation of other public transit.

The model coefficient for the percentage of African American residents was positive and weakly significant (p < 0.1). This finding is in line with the extensive prior literature that has found that African Americans use public transit more than other racial/ethnic groups [40,41]. However, this finding is surprising if app-based microtransit services are considered as innovative shared mobility services, where research has found that African American and other minority racial/ethnic groups are less likely to use these services [42,43]. Given that the coefficient on the percentage of African American residents in our model was only weakly significant (p < 0.1), there is insufficient evidence to definitively support the relationship between the percentage of African American residents and microtransit ridership.

Two service area characteristic variables were not significant, but we included them in the model for theoretical reasons. We would expect population density to have an effect on ridership as this variable represents the number of potential riders and proximity of potential riders within a given microtransit service area. Regarding the number of POIs within a service area, we learned through the process of collecting ridership data from public transit agency planners and through communications with other microtransit operators that having POIs within a service area was an indicator of high ridership. For example, superstores, such as Walmart or Kroger, have been anecdotally given as examples of strong microtransit ridership drivers. However, in our model, the number of POIs per square mile was not found to have a significant relationship with ridership.

3.3. Findings from Microtransit Ridership per Capita Model

One of the ridership model findings was that as the size of the service area increases, ridership increases, which makes sense because a larger service area can capture more potential riders. However, we were also interested in the overall share of microtransit passengers in the population. We divided microtransit ridership by the total population of the microtransit zone to determine ridership per capita and used the same methodology to arrive at the final model summarized in

Table 4. Final model specification for 2022 ridership per capita.

	Ridership per Capita, 2022 n = 145 Microtransit Service Zones Operated by 32 Public Transit Agencies
Predictors	Estimates	CI	p
(Intercept)	0.03	−0.36–0.41	0.898
Internal factors
Hours of operation/week	0.43	0.29–0.58	<0.001
Service type [point deviation]	0.84	0.36–1.31	0.001
Service area (sq mile)	−0.33	−0.50–−0.16	<0.001
External factors
Population density (persons/sq mile)	−0.40	−0.56–−0.24	<0.001
% African American	0.13	0.00–0.26	0.052
% Public transit commuters	−0.19	−0.33–−0.05	0.008
# POIs per sq mile	0.10	−0.06–0.26	0.231
Random Effects
Residual	0.39
Agency-specific	0.88
ICC	0.70
Nagency	32
Observations	145
Marginal R2/Conditional R2	0.304/0.788

Bold p-values indicate significance at the p < 0.05 level. Source: own compilation, based on model output of lme4 package in R.

.

Similar to the total ridership model, the ridership per capita model has a solid model fit (R² = 0.788). The intraclass correlation coefficient is also similar (ICC = 0.70), indicating a high correlation between ridership per capita measurements within the same public transit agency. The following subsections summarize the findings on the impact of internal and external factors on suburban microtransit ridership per capita.

3.3.1. Impact of Internal Agency Factors on Microtransit Ridership per Capita

In the ridership per capita model, internal agency factors continued to have a significant impact. Hours of operation per week and service type had positive effects on ridership per capita, and the magnitude of impact was similar to the ridership model. One difference is that the coefficient for service area changed in sign from positive to negative. While increasing the service area may increase total ridership, the share of microtransit passengers within the service area will decrease.

3.3.2. Impact of External Agency Factors on Microtransit Ridership per Capita

The population density became a significant variable in the ridership per capita model and had a negative coefficient. This finding is counter-intuitive given that we would assume that having a higher population density would result in more potential passengers and increase ridership per capita. However, taken together with the negative coefficient on the percentage of public transit commuters, it is possible that a higher population density is associated with higher availability of traditional public transit services, which draw riders away from microtransit services.

The model coefficient for the percentage of African American residents was positive and significant at the p < 0.05 level. As discussed earlier, this finding is supported by the literature on the higher use of public transit by African American riders, but counter to the literature on the low usage of shared mobility by African Americans. The use of microtransit by different racial/ethnic groups could be explored further in future research.

Finally, although the number of POIs per square mile was still not significant, the p-value was higher in the ridership per capita model compared to the total ridership model, which suggests a stronger relationship between number of POIs per square mile and ridership per capita.

4. Discussion

We found that internal agency factors (e.g., service design, operating hours) had a greater impact on microtransit ridership and microtransit ridership per capita than external agency factors (e.g., population density, percentage public transit commuters).

This finding suggests that, while public transit agencies have discretion over the design and optimization of both fixed-route services and microtransit, changes to microtransit operations can have a higher impact on ridership than prior research has found on fixed-route transit ridership, which is more highly impacted by external factors such as: vehicle ownership, parking prices, and population density.

Based on the results of our random effects model, we found that the most significant factor to increase microtransit ridership was through operating point deviation models rather than zone-based services. The case study evaluation of Denver Regional Transportation District’s (RTD’s) general public DRT system can shed light on why point deviation models can gain more ridership than zone-based services [30]. The Denver RTD case found that carefully planning service areas based on population and employment density, street layout characteristics, and public transit connections could result in more productive microtransit services. The Denver RTD found that more structured, point deviation services achieved higher productivity (five to nine passengers per hour) compared to many-to-many zone-based services (three to four and a half passengers per hour) [30]. This is because, with point deviation services, microtransit vehicles can pick up larger groups of passengers that are concentrated at a scheduled location, rather than having to travel to pick up individuals from many different locations.

To what extent is this approach feasible for other public transit agencies, i.e., can some zone-based services be re-designed as point deviation services to increase ridership and productivity? We compared the service area characteristics of point deviation and zone-based systems and found minimal differences in population density, percentage of public transit commuters, and availability of a public transit service. Point deviation services had a significantly higher density of POIs per square mile (6.7 POIs per square mile vs. 4.3 POIs per square mile for zone-based systems). Since we found that 13 of the 38 point deviation services had a scheduled stop at a POI (e.g., shopping center, college/university, medical center), this indicates the importance of POIs in selecting where to locate the scheduled stop for point deviation services. Public transit centers were also used as scheduled stops for point deviation services, with 19 point deviation services having a scheduled stop at a transit center or park-and-ride lot connecting to bus and/or rail. Since the density of public transit stops was similar between point deviation and zone-based systems and 52% of zone-based systems had at least one transit center within service area boundaries, there is potential for some zone-based systems to set anchor points at a transit center to increase ridership and act as a first mile–last mile connection to public transit.

For the remaining 48% of zone-based systems without a public transit center or a significant number of POIs, designing a point deviation service to increase ridership may not be feasible. In these places, using ridership as an evaluation metric can be misleading, as it is naturally difficult to attain ridership in low-density, low-demand regions. Our other findings on microtransit services—lower wait times and operating hours during off-peak times and weekends—show the potential for zone-based services to increase accessibility for residents in low-density suburban areas. Surveys of microtransit passengers have also found that customer satisfaction with the service is high, particularly among low-income travelers, passengers without a personal vehicle, and women [38]. Another report on microtransit pilot programs also noted the issue with only using ridership as a performance metric because in contrast to fixed-route systems that are designed to optimize ridership and cost, microtransit is aiming to deliver a fundamentally different service [9]. Instead, metrics oriented around mobility, safety, and customer experience can show additional microtransit benefits for low-density suburban areas, and for low-income or other transportation-underserved groups [9].

5. Conclusions and Policy Strategies

This research contributes to the understanding of the performance of suburban microtransit programs at a time of great interest in the ability of general public DRT to support existing public transit in low-density suburban areas. We used a random effects model to analyze the internal and external agency factors impacting suburban microtransit ridership using ridership data from 32 microtransit programs in the U.S. We found that internal agency factors, such as service type (point deviation vs. zone-based service), operating hours, and service area have a more significant impact on microtransit ridership than external agency factors, such as population density and number of POIs per square mile.

The main policy strategy that public agencies could take to increase microtransit ridership is by converting zone-based services to point deviation services. The results of our random effects models found that the most significant variable influencing ridership and ridership per capita was service design; specifically, that point deviation services had higher ridership than zone-based services. Zone-based services operate by picking up and dropping off passengers anywhere within a service area, while point deviation services operate similarly but have a timed departure at a certain point of interest (e.g., shopping center, public transit center). In our sample of microtransit programs, 52% of zone-based microtransit service areas had at least one public transit center within service area boundaries. This shows there is great potential for these zone-based service areas to be converted to first mile–last mile point deviation services, with scheduled departures set at public transit centers. This type of microtransit service design can also strengthen connections to fixed-route public transit in suburban areas, a more sustainable transportation mode than private vehicle use.

For places without clear activity generators, operating a point deviation service and optimizing for higher ridership may not be possible. This can be seen in lower density suburban environments with more dispersed activity areas or more residential areas, where there are not as many pick-up or drop-off hot spots. In these environments, it is difficult for microtransit services to serve trips with efficiency since origins and destinations are generated with lower concentrations. This is in contrast to other suburban areas with a public transit, shopping, or employment center that can generate a higher concentration of trip requests. Instead, as other research has shown, public transit agencies should consider performance metrics other than ridership to assess the effectiveness of microtransit at serving transportation needs. For example, metrics such as safety, customer satisfaction, or changes to job access, healthcare access, etc., could show the relative merits of microtransit over other transportation options.

Further Research

Our analysis was limited by the data available for microtransit programs in the U.S. However, there are examples of microtransit programs in different geographic contexts such as Europe and Asia where determinants of microtransit ridership may differ. Further research could explore the topic of microtransit ridership in other places outside of the U.S.

The data used in this analysis were collected manually from publicly available sources. This means that several key variables were left out due to data quality or availability issues including the number of microtransit vehicles per zone, vehicle revenue hours, and operating costs. The Federal Transit Administration (FTA) requires that public transit agencies report key performance metrics (e.g., ridership, revenue hours, operating costs, etc.) for all public transit modes. However, under current FTA definitions, microtransit performance metrics are reported under “Demand Response” services, which include ADA paratransit, a mode with higher wait times, low ridership, and intended to serve a specific population (individuals with disabilities). As the number of microtransit programs grows in the U.S., creating a separate category for microtransit within the FTA’s National Transit Database reporting system would allow for more studies like this one to be conducted, with a larger and more consistent sample of microtransit programs.

In this quantitative analysis, we used only quantitative sources. One of our findings was that point deviation services can attain higher ridership than zone-based services, which we infer is due to operating similarities between point deviation microtransit and fixed-route transit. This is supported by the case study of the Denver RTD. However, passenger behavior motivations or differences in ride-matching algorithms could also explain the higher ridership of point deviation services, which we do not address in this paper. This is an area of further research.

Finally, future research could explore the potential benefits of microtransit programs outside of operating characteristics, which focus on the impacts on suburban residents. As the low-income suburban population continues to grow, our findings show the potential for microtransit to provide enhanced mobility compared to fixed-route public transit in areas and times when low-income residents need to travel. Qualitative analysis employing focus groups and/or interviews with microtransit passengers and non-users can help to reveal the transportation needs that microtransit can meet in suburban areas. Expert interviews with public agency representatives could reveal how agencies see microtransit fitting in as part of the public transit landscape and how it can support suburban transportation systems.