Embracing Urban Micromobility: A Comparative Study of E-Scooter Adoption in Washington, D.C., Miami, and Los Angeles

Abstract

E-scooters have emerged as a popular micromobility option for short trips, with many cities embracing shared e-scooters to enhance convenience for travelers and reduce reliance on automobiles. Despite their rising popularity, there is a lack of clear understanding of how user preferences and adoption practices vary by location. This study aims to explore user and non-user attitudes towards e-scooter use in diverse urban settings. A meta-analysis of data from three surveys (N = 1197) conducted in Washington, D.C., Miami, FL, and Los Angeles, CA, was performed to compare e-scooter users and non-user profiles, mode choice factors, and attitudes and preferences towards e-scooter use. Additionally, machine learning (ML) and SHAP (SHapley Additive exPlanations) analysis were utilized to identify influential factors in predicting e-scooter use in each city. The results reveal that the majority of e-scooter users are 25 to 39 of age, male, with higher income and a bachelor’s degree, and 92% possess a driver’s license. Significant differences in attitudes between e-scooter users and non-users highlight the complexity of perceptions towards e-scooter usage. The ML model indicates that employment status negatively impacts the prediction of e-scooter users, while factors such as living without a car and using non-motorized modes positively influence e-scooter use. Educational background is a significant e-scooter mode choice factor in Washington, D.C. and Miami, whereas attitudinal questions on car and technology usage are influential in Los Angeles. These findings provide valuable insights into the factors shaping e-scooter adoption, informing urban transportation planning and policymaking and enhancing understanding of shared micromobility and its impact on urban mobility.

1. Introduction

The overreliance on automobiles to meet mobility needs has led to heightened congestion and increased pollutant levels, while public transportation struggles to cope with the growing demand for urban mobility [1]. In response to these challenges, various new mobility concepts have emerged, with shared mobility standing out as a rapidly expanding sector within the transportation-sharing economy. Shared mobility encompasses diverse services like car sharing and bike sharing, addressing the evolving needs of urban travelers [2]. Within the realm of shared mobility, the concept of shared micromobility has gained prominence. This involves temporarily renting small, low-speed vehicles designed for individual transportation, encompassing both station-based bike sharing and dockless alternatives [3]. A notable addition to this sector is the shared electric scooter (e-scooter), a service that has witnessed substantial global growth in recent years [4,5,6]. In the U.S. alone, 136 million shared micromobility trips were recorded in 2019, marking a substantial 60% increase from 2018. Notably, e-scooter trips accounted for 86 million of these journeys [7].

Given the exponential growth and increasing popularity of these emerging transportation services, it becomes important to document users’ preferences and how they impact their travel mode choices at locations where e-scooter services are available. Understanding the dynamics of these evolving mobility solutions is paramount for crafting effective strategies that cater to the diverse needs of urban travelers.

The examination of travel mode choice stands as a well-explored and captivating domain within travel behavior research. The existing literature highlights the multifaceted nature of factors influencing the selection of travel modes, encompassing sociodemographic characteristics of travelers [8,9,10,11], their attitudes towards modes [8,12,13,14], and external factors like comfort, convenience, safety, and overall travel satisfaction [12,15,16,17].

Most e-scooter studies traditionally rely on constructing user profiles of e-scooter users based solely on explanatory variables, such as demographic characteristics. Despite the rapidly growing research in this field, a knowledge gap exists regarding user and non-user preferences and attitudes toward e-scooter use while taking under consideration latent variables in different cities. This study aims to fill in this research gap by posing and addressing the following four research questions:

Research Question 1.

Are there any differences in the profiles of e-scooter users and non-users among different cities?

Research Question 2.

Are there any differences concerning mode choice factors and attitudes toward e-scooter and private vehicle use among different cities?

Research Question 3.

Can latent variables influence the prediction of mode choice of e-scooter users and non-users?

Research Question 4.

Do the influential factors on e-scooter mode selection vary across different cities?

At the core of this research lies the primary aim of investigating the potential effects of variables in predicting the utilization of e-scooters as a mode of transportation in urban environments. The study focuses on demographics, mode choice factors, and attitudinal and perceptual measures related to e-scooters and private vehicles. The geographical context provides depth to this exploration, encompassing three diverse cities—Washington, D.C., Miami, FL, and Los Angeles, CA. This deliberate selection enables a comparative analysis designed to unveil city-specific dynamics that may influence the adoption of e-scooters. Machine learning (ML) techniques serve as a valuable tool to support the objectives of this study through the utilization of prediction models and SHAP analysis.

2. Literature Review

Shared micromobility is recognized as a practical and sustainable transportation option for short-range trips that offers potential benefits, including alleviating local traffic congestion, enhancing air quality, and contributing to the decarbonization of the transportation fleet [18,19]. A swiftly expanding shared micromobility service is the e-scooter, which has been introduced in numerous cities and countries over the past five years [20].

Compared to other transportation modes, relatively little is known about the mode choice factors, travel patterns, perceptions and attitudes, and the potential modal shift associated with shared micromobility use. To comprehend the potential impact of this emerging service, it is essential to explore the varied usage patterns across different cities through user surveys and field data collection [21,22,23,24]. Collecting survey data involves distributing a questionnaire within a designated service area. Surveys are instrumental in gathering user demographics and preferences information, which are typically not captured through ridership data collection. A recent study suggests that factors such as land use, urban morphology, and socio-economic variables significantly influence travel mode choice [25]. Questions about user preferences include considerations related to the use of micromobility modes for first/last mile transport [22] and the intentions of survey participants to use the micromobility services [26]. For example, in their analysis, Degele et al. (2018) examined the utilization patterns of e-scooter sharing programs and contended that e-scooters are especially effective for short-distance travel [27]. This assertion was supported by their examination of data obtained from a German e-scooter provider. Surveys of e-scooter users also indicated that riders are attracted to e-scooters as they offer a quicker and more convenient alternative to automobile use but this is also due to their fun nature, affordability, and functionality as a convenient connector to transit options [28]. Research indicates that road infrastructure significantly influences the mode choice of e-scooter drivers, and bicycle paths are used by 60–90% of e-scooter users [29]. Furthermore, the availability of public transportation also impacts the usage patterns of shared e-scooters. Specifically, the number of tram departures positively affects shared e-scooter usage, especially during off-peak times, demonstrating an interdependence between public transportation and e-scooter usage [30].

Research findings propose that individuals utilizing shared micromobility services tend to have similar demographic characteristics. Previous research has examined the profiles of e-scooter users using survey analyses, uncovering a consistent pattern where the majority of e-scooter riders tend to be young, white (Caucasian) males, with higher household income levels and possessing a college degree [31,32,33,34]. Studies report that age is a significant factor influencing e-scooter usage and younger adults are overrepresented as e-scooter users [35,36,37]. Usage, in general, tends to decrease with age [38]; however, elderly individuals who begin e-scooter riding frequently evolve into consistent users [39]. Additionally, numerous studies indicate that e-scooters are perceived as attractive and receive positive evaluations from a significant portion of the population from diverse racial backgrounds and lower-income groups as a mode of transportation [32,40,41]. Murphy et al. (2021) examined surveys encompassing 18 cities in the U.S. and discovered a higher representation of people of color among e-scooter users. Additionally, there are minimal differences and variations in income levels among e-scooter riders and the broader population [3].

Apart from demographic and socioeconomic traits, research has highlighted that attitudinal factors significantly influence the likelihood of adopting and using e-scooters [42,43]. Earlier reports also noted that travelers’ attitudes toward shared e-scooters were crucial predictors of e-scooter use [33,44]. Certain authors explored the diversity of preferences in e-scooter usage through advanced modeling techniques, including structural equation modeling and latent class models [4,43,45]. There is continuing interest in identifying hidden factors that affect the acceptance or reluctance of shared micromobility use. It is also important to document the attitudes of respondents that significantly impact the adoption process. Reportedly, people with positive attitudes toward technology, including trust and interest in technologies [46,47,48], are more likely to adopt and embrace new emerging forms of transportation, such as shared e-scooters.

It is well recognized that modal choice factors such as safety, cost, comfort, reliability, and environmental impacts serve as underlying variables for adapting to new forms of transportation. Therefore, comprehending the factors that influence modal choice is essential in identifying the barriers that new users may face when considering the use of shared micromobility services [49]. A survey conducted in Greece has confirmed that safety concerns are a primary reason why non-users refrain from using e-scooters [39]. In another study, Sanders et al. (2020) emphasized that survey participants cited convenience and speed as benefits of e-scooter use and identified safety and reliability as their main concerns [50]. Other studies have noted that younger adults are motivated to use micromobility options primarily for their environmental advantages [51,52], while older adults are drawn to micromobility modes because of the potential health benefits they offer [53]. Results from other surveys underscored that factors such as convenience, ease of use, and the enjoyment of riding were the most significant positive aspects, whereas a lack of infrastructure, safety concerns, and regulatory issues emerged as the main barriers to embracing this mode of transportation [54].

Different geographical contexts provide an in-depth exploration of various urban settings. Previous studies have conducted comparative analyses in this context, including comparisons of four U.S. cities: Washington, D.C., Los Angeles, Miami, and Birmingham [55]. Other studies have compared Washington, D.C. and Los Angeles [56,57], as well as Chinese cities like Beijing and Guangzhou [58]. Additionally, in a broader geographical context, research has been conducted comparing five countries: Australia, Belgium, the Czech Republic, Norway, and Sweden [59], as well as comparing cities such as Guangzhou from China and Brisbane from Australia [60]. These studies highlight the significance of examining different urban environments to understand the variability in e-scooter usage and other mobility patterns.

This study expands on the current body of literature on e-scooter use by evaluating preferences, perceptions, and attitudes toward shared e-scooter use of both users and non-users in three US cities. The study further examines intentional and attitudinal variables in addition to the commonly used demographic and socioeconomic factors. Furthermore, this study considers variations due to geographical context by comparing users’ and non-users’ perspectives and identifying the observed and unobserved factors associated with e-scooter usage at different study locations.

3. Methodology

In this study, Likert scale visualization was employed to explore potential differences among study cities in terms of mode choice factors and travelers’ attitudes towards e-scooter and private vehicle use. Furthermore, the Kruskal–Wallis test [61] was utilized to assess variations across cities statistically, enhancing the robustness of the analysis. Another key objective of this study was to investigate the influence of latent variables as determinants of mode choice, particularly in distinguishing between e-scooter users and non-users. This objective necessitated the application of factor analysis in SPSS Statistics 26 [62] to identify latent variables. Subsequently, prediction models were employed to evaluate the impact of these latent variables on accurately predicting e-scooter usage patterns using machine learning techniques [63,64,65]. Moreover, the study aimed to determine if factors that influenced the use of e-scooters exhibited variability across the study cities. For this purpose, SHAP analysis was utilized to interpret the output of the prediction models [66]. The study steps are summarized in Figure 1.

Data

This study performed a meta-analysis of data gathered from an online questionnaire survey that included participants from three cities: Washington, D.C., Miami, FL, and Los Angeles, CA. These cities were chosen for analysis due to their status as some of the largest cities in the U.S. with early adoption of shared e-scooters. They also exhibit significant differences in the characteristics of their built environments, such as the size of street blocks and varying weather conditions. These differences could lead to varying preferences for and perceptions of e-scooter use. Additionally, all three cities have established permit programs for e-scooter operations, which include detailed terms and conditions regarding fleet management, parking, data reporting, payment options, and programs for low-income customers.

Data collection took place in April 2021 and May 2022 in Washington, D.C., from September to November 2021 in Miami, and in May 2022 in Los Angeles. Participants had to meet residency and age (18 years of age or above) eligibility criteria and provide their consent for participation in the study. The web-based survey was developed in Qualtrics, and Qualtrics managed the recruitment and compensation of participants in accordance with their standard business practices. Recruitment of participants involved leveraging personal networks, email, newsletters, and various social media platforms. The required approvals were secured from the Institutional Review Board offices in collaboration with the University of Florida and Florida International University [55,56,57]. The dataset containing the study participants’ responses was made available to this study through the STRIDE University Transportation Center, and additional details regarding the data collection effort are available in [67].

The survey asked participants about the frequency of using various travel modes in the preceding 30 days. Subsequent inquiries delved into attitudinal and perceptual measures concerning mobility, travel preferences, and trip purpose using Likert scale questions. Information about demographic and socioeconomic characteristics of participants was also gathered. More details regarding the content of the survey instrument are available in [67].

The study sample consisted of a total of 1197 complete responses. Some additional entries were excluded because of unanswered questions or in cases where responses to specific travel attitude questions were considered unreasonable. Respondents from Washington, D.C. contributed 414 valid responses (35% of total), those from Miami provided 408 responses (34%), and respondents from Los Angeles contributed 375 responses (31%). The sample size was deemed appropriate and consistent with that of earlier studies reported in the literature including [55,56,57].

4. Results

4.1. Descriptive Analysis

4.1.1. Comparison of E-Scooter Users’ and Non-Users’ Profiles

To address the first research question as stated in the introduction, the study investigated the overall characteristics of the study sample, distinguishing between e-scooter users and non-users (

Table 1. Sample size by city.

	Washington, D.C.		Miami		Los Angeles		Total
	n	% of Total	n	% of Total	n	% of Total	n	% of Total
E-scooter users	193	16%	101	8%	171	14%	465	39%
Non-users	221	18%	307	26%	204	17%	732	61%

). Participants who had never used shared e-scooters were categorized as “non-users”, whereas those who reported using a shared e-scooter at least once were categorized as “users”. Table 1 highlights intriguing differences between the proportions of e-scooter users and non-users across the three cities studied. The numbers of users and non-users in Washington, D.C. and Los Angeles were relatively balanced. Conversely, Miami reported a lower percentage of e-scooter users (8% of total survey respondents) and a notably higher percentage of non-users (26% of total).

Demographic trends among e-scooter users in the three cities studied are outlined in

Table 2. E-scooter users’ profile by city.

Variable	Category	Washington, D.C.		Miami		Los Angeles
		n	%	n	%	n	%
Gender:	Male	111	58	62	61	112	65
	Female	79	41	39	39	56	33
Age:	18–24	33	17	12	12	31	18
	25–29	51	26	19	19	44	26
	30–39	64	33	44	44	56	33
	40–49	28	15	17	17	28	16
	50–59	11	6	8	8	11	6
	60 or over	6	3	1	1	1	1
Income:	Less than USD 25,000	10	5	5	5	27	16
	USD 25,000–USD 49,999	29	15	26	26	28	16
	USD 50,000–USD 74,999	30	16	14	14	22	13
	USD 75,000–USD 99,999	30	16	21	21	23	13
	USD 100,000–USD 124,999	22	11	9	9	15	9
	USD 125,000–USD 149,999	16	8	13	13	6	4
	USD 150,000 or more	35	18	13	13	32	19
Vehicles:	0	72	37	8	8	37	22
	1	75	39	31	31	63	37
	2	35	18	29	29	50	29
	3	3	2	25	25	10	6
	4	7	4	4	4	9	5
	5	0	0	3	3	1	1
	6 or more	1	1	1	1	1	1
HousePop:	1	71	37	14	14	44	26
	2	76	39	20	20	67	39
	3	17	9	24	24	27	16
	4	16	8	25	25	18	11
	5	9	5	14	14	11	6
	6 or more	4	2	4	4	4	2
License:	Yes	178	92	100	99	153	89
Student:	Yes	26	13	28	28	35	20
Employment:	Employed	8	4	81	80	18	11
	Other or no answer	185	96	20	20	153	89
Education:	High school or less	11	6	17	17	29	17
	Associate’s degree	15	8	30	30	33	19
	Bachelor’s degree	95	49	42	42	78	46
	Post-graduate degree	72	37	13	13	31	18
Race:	White	127	66	68	67	87	51
	Black	17	9	27	27	11	6
	Asian	17	9	1	1	19	11
	Other (multicultural)	32	17	5	5	54	32

. The majority of e-scooter users are male (58% in Washington D.C., 61% in Miami, and 65% in Los Angeles), with a concentration in the 25–39 age group. Income distribution is diverse, with a significant portion falling into the higher brackets. Most e-scooter users own one vehicle; however, 37% in Washington, D.C. and 22% in Los Angeles users reported that they do not own an automobile. E-scooter users come from predominantly two-person households, and over 92% possess a valid driver’s license. Educational backgrounds vary, with a substantial number holding at least a bachelor’s degree, ranging from 86% in Washington, D.C. to 55% in Miami. The racial composition is diverse, with the majority being white but with multicultural representation. These profiles align with previous studies, indicating that e-scooter users are typically younger, educated white males, with higher incomes [27,29,63].

Table 3. E-scooter non-users’ profile by city.

Variable	Category	Washington, D.C.		Miami		Los Angeles
		n	%	n	%	n	%
Gender:	Male	111	50	172	56	112	55
	Female	104	47	135	44	88	43
Age:	18–24	31	14	28	9	22	11
	25–29	35	16	20	7	22	11
	30–39	55	25	102	33	44	22
	40–49	30	14	72	23	31	15
	50–59	30	14	54	18	34	17
	60 or over	40	18	31	10	51	25
Income:	Less than USD 25,000	18	8	53	17	49	24
	USD 25,000–USD 49,999	37	17	95	31	36	18
	USD 50,000–USD 74,999	29	13	57	19	42	21
	USD 75,000–USD 99,999	33	15	34	11	29	14
	USD 100,000–USD 124,999	21	10	29	9	13	6
	USD 125,000–USD 149,999	19	9	26	8	10	5
	USD 150,000 or more	59	27	13	4	25	12
Vehicles:	0	45	20	30	10	20	10
	1	91	41	126	41	95	47
	2	64	29	112	36	57	28
	3	19	9	30	10	22	11
	4	2	1	6	2	8	4
	5	0	0	3	1	2	1
	6 or more	0	0	0	0	0	0
HousePop:	1	59	27	44	14	44	22
	2	86	39	93	30	69	34
	3	35	16	85	28	45	22
	4	28	13	61	20	28	14
	5	6	3	18	6	10	5
	6 or more	7	3	6	2	8	4
License:	Yes	207	94	283	92	178	87
Student:	Yes	30	14	47	15	32	16
Employment:	Employed	154	70	186	61	106	52
	Other	67	30	121	39	98	48
Education:	High school or less	18	8	62	20	49	24
	Associate’s degree	39	18	112	36	63	31
	Bachelor’s degree	69	31	101	33	60	29
	Post-graduate degree	95	43	32	10	32	16
Race:	White	143	65	214	70	101	50
	Black	33	15	67	22	17	8
	Asian	24	11	4	1	33	16
	Other (multicultural)	21	10	22	7	53	26

summarizes distinctive characteristics of non-users and compares them with those of e-scooter users. Among non-users, gender distribution is balanced, with males constituting 50%, 56%, and 55% in Washington, D.C., Miami, and Los Angeles, respectively. The 30–39 age group is more prevalent among non-users. Notable income disparities exist, especially in Miami, where 31% of non-users earn an annual income ranging from USD 25,000 to USD 49,999, compared to e-scooter users that report income in higher income brackets. Vehicle ownership differences exist, with lower percentages of non-users having zero vehicles in Washington, D.C. (20%) and Los Angeles (10%) compared to e-scooter users (37% and 22%, respectively). Household sizes vary, as well, indicating distinctions in family structures. License ownership is consistently high among non-users. Employment status and student enrollment show differences, with a higher portion of non-users being employed compared to e-scooter users. Educational backgrounds vary from city to city, with a notable percentage of non-users holding at least a bachelor’s degree, similarly to e-scooter users. Racial composition is diverse, predominantly white, with significant representation from other racial and multicultural groups. These insights contribute to a comprehensive understanding, forming the foundation for developing strategies and targeted interventions to enhance e-scooter adoption and meet diverse urban mobility needs.

4.1.2. Variation in Mode Choices by City

Figure 2 presents the transportation mode choices of survey participants in Washington, D.C., Miami, and Los Angeles. The bar chart illustrates the percentage of participants utilizing each mode over a 30-day period and facilitates a clear comparison between the study cities. The respondents were allowed to select more than one transportation mode. Walking emerged as a mode of choice for completing trips for the majority of survey participants across all cities, notably soaring to 92% in Washington, D.C. Personal vehicle use was most prevalent in Miami, with 84% of respondents reporting use, and less common in Washington, D.C. (65% use). The trends were reversed with respect to the use of taxis or other ride-sharing services, public transit, and biking, with Washington, D.C. showing higher utilization rates than Miami. E-scooters showcased considerable popularity, particularly in Washington, D.C. and Los Angeles, indicating a burgeoning trend in urban mobility. As Figure 2 shows, approximately 46% of users reported riding e-scooters during the 30-day reference period in both Washington, D.C. and Los Angeles, compared to 25% in Miami. This similarity in e-scooter usage between Los Angeles and Washington, D.C., suggests a parallel adoption pattern of this emerging transportation mode, reflecting the cities’ supportive infrastructure and policies towards micromobility. The lower prevalence of car-sharing services (such as Zipcar) suggests that such services are a less common choice among respondents across the three cities studied.

4.2. Likert Scale Questions (Observed Variables)

This section summarizes responses related to mode choice factors and attitudes toward e-scooters and private automobiles. Participants rated the importance of six factors in their transportation mode selection: 1. cost, 2. time, 3. reliability, 4. comfort, 5. safety, and 6. environmental impacts. The Likert scale ranged from “not at all important” to “extremely important”. Figure 3 depicts participant responses, categorized by user group (e-scooter users versus non-users) and location (Washington, D.C., Miami, and Los Angeles).

Participants also shared their perspectives on e-scooters and their relationship with public transit, responding to six Likert scale questions ranging from “strongly disagree” to “strongly agree” with statements such as the following:

• Riding e-scooters is a safe way to get around.
• My city has enough bike lanes to accommodate e-scooter use.
• My city has enough space for proper e-scooter parking.
• The arrival of shared e-scooters is a good thing for the city.
• Shared e-scooters can strengthen the operations of public transit (e.g., facilitating last-mile transit connection).
• Shared e-scooters will make people use public transit less.

Figure 4 displays the range of responses for each question, distinguishing between e-scooter users and non-users, across the three study cities.

Participants were also asked to express their attitudes toward car use and technology using a Likert scale ranging from “strongly disagree” to “strongly agree” for the following statements:

• I hope to live without a car.
• I definitely want to own a car.
• I try to use public transit whenever I can.
• I try to travel with non-motorized modes (biking and walking) as much as I can.
• I am confident in my ability to use new technologies (e.g., a smartphone app).
• Learning how to use new technologies is often frustrating for me.
• As a general principle, I would rather own things than rent them.

Figure 5 visually represents respondents’ attitudes, providing insights into preferences and sentiments among different user groups (e-scooter users and non-users) across the three study cities.

Beyond the visual representation of mode choice factors and attitudes toward e-scooters and cars (Figure 2, Figure 3 and Figure 4), statistical tests were employed to systematically analyze and quantify these distinctions, offering a more comprehensive understanding of the variations in attitudes across the three cities. The findings are summarized next.

4.3. Kruskal–Wallis Test Results

The Kruskal–Wallis test assessed differences in mode choice factors and attitudes across Washington, D.C., Miami, and Los Angeles. The null hypothesis posits that Likert scale responses are similar across the cities, while a rejection of this hypothesis would suggest that variations exist in specific factors or attitudes among the cities. Subsequently, the significant findings were analyzed using posthoc tests to determine which of the categories were statistically different. The significance values were further adjusted by the Bonferroni correction for multiple tests, thereby reducing the normal alpha level to minimize the risk of type-I errors.

In terms of mode choice factors for e-scooter users (N = 464), Washington, D.C. and Los Angeles exhibited similar responses, while Miami had distinct viewpoints (

Table 4. Kruskal–Wallis test results for mode choice factors.

	E-Scooter Users N = 465						Non-Users N = 732
Factors	Sig	Decision	Pairwise Comparisons of City				Sig	Decision	Pairwise Comparisons of City
			S1–S2	Adj. Sig. a					S1–S2	Adj. Sig. a
Cost	0.005	Reject	D.C.-LA	1.000			0.000	Reject	D.C.-LA	0.027	*
			D.C.-Mi	0.005	**				D.C.-Mi	0.000	***
			LA-Mi	0.024	*				LA-Mi	0.000	***
Time	0.003	Reject	LA-D.C.	0.445			0.000	Reject	D.C.-LA	1.000
			LA-Mi	0.002	**				D.C.-Mi	0.000	***
			D.C.-Mi	0.074					LA-Mi	0.000	***
Reliability	0.000	Reject	LA-D.C.	1.000			0.000	Reject	D.C.-LA	1.000
			LA-Mi	0.000	***				D.C.-Mi	0.000	***
			D.C.-Mi	0.000	***				LA-Mi	0.000	***
Comfort	0.000	Reject	D.C.-LA	0.440			0.000	Reject	D.C.-LA	0.002	**
			D.C.-Mi	0.000	***				D.C.-Mi	0.000	***
			LA-Mi	0.000	***				LA-Mi	0.000	***
Safety	0.000	Reject	LA-D.C.	1.000			0.000	Reject	D.C.-LA	0.001	**
			LA-Mi	0.000	***				D.C.-Mi	0.000	***
			D.C.-Mi	0.000	***				LA-Mi	0.001	**
Environmental impacts	0.000	Reject	D.C.-LA	1.000			0.000	Reject	LA-D.C.	1.000
			D.C.-Mi	0.000	***				LA-Mi	0.000	***
			LA-Mi	0.000	***				D.C.-Mi	0.005	**

Notes: Significance levels for reference: * (p < 0.05), ** (p < 0.01), *** (p < 0.001). ‘Sig. ^a’ represents the adjusted significance values.

). Notably, time showed differences between Los Angeles and Miami. Non-users of e-scooters displayed varying views on mode choice factors across the three cities, particularly on cost, comfort, and safety. Washington, D.C. and Los Angeles shared similar perspectives among non-users regarding time, reliability, and environmental impacts.

As outlined in

Table 5. Kruskal–Wallis test results for e-scooter attitudes.

	E-Scooter Users N = 465						Non-Users N = 732
E-Scooter Attitudes	Sig	Decision	Pairwise Comparisons of City				Sig	Decision	Pairwise Comparisons of City
			S1–S2	Adj. Sig. a					S1–S2	Adj. Sig. a
Attitude 1.	0.125	Retain					0.006	Reject	LA-D.C.	0.337
									LA-Mi	0.004	**
									D.C.-Mi	0.372
Attitude 2.	0.000	Reject	LA-D.C.	1.000			0.000	Reject	LA-D.C.	1.000
			LA-Mi	0.000	***				LA-Mi	0.000	***
			D.C.-Mi	0.000	***				D.C.-Mi	0.000	***
Attitude 3.	0.284	Retain					0.030	Reject	LA-D.C.	0.954
									LA-Mi	0.029	*
									D.C.-Mi	0.362
Attitude 4.	0.001	Reject	Mi-LA	0.011	*		0.000	Reject	LA-Mi	0.000	***
			Mi-D.C.	0.001	**				LA-D.C.	0.000	***
			LA-D.C.	1.000					Mi-D.C.	0.178
Attitude 5.	0.000	Reject	Mi-LA	0.004	**		0.000	Reject	LA-Mi	0.020	**
			Mi-D.C.	0.000	***				LA-D.C.	0.000	***
			LA-D.C.	1.000					Mi-D.C.	0.020	**
Attitude 6.	0.007	Reject	D.C.-LA	1.000			0.000	Reject	D.C.-LA	1.000
			D.C.-Mi	0.006	**				D.C.-Mi	0.000	***
			LA-Mi	0.072					LA-Mi	0.001	**

Notes: Significance levels for reference: * (p < 0.05), ** (p < 0.01), *** (p < 0.001). ‘Sig. ^a’ represents the adjusted significance values. E-scooter use attitudes: 1. safe to ride; 2. bike-lanes are adequate; 3. enough e-scooter parking; 4. e-scooters are good for the city; 5. e-scooters support public transit; 6. e-scooters attract users from public transit.

, the Kruskal–Wallis test revealed notable variations in e-scooter users’ attitudes across Washington, D.C., Miami, and L.A. Attitudes 1 and 3 showed consistent agreement across all cities, indicating shared perspectives on e-scooter safety and parking availability, whereas attitudes 2, 4, and 5 exhibited differences in viewpoints between Washington, D.C. and Los Angeles users (similar stances) versus Miami users (distinct perspectives on bike lanes, impact on the city, and public transit). Regarding attitude 6 related to e-scooters reducing public transit use, a significant difference emerged between Washington, D.C. and Miami, with similar viewpoints found in comparisons involving Washington, D.C. and Los Angeles, as well as Los Angeles and Miami e-scooter users.

Kruskal–Wallis test results on car users’ attitudes revealed significant differences between Washington, D.C., Miami, and Los Angeles (

Table 6. Kruskal–Wallis test results for car use attitudes.

	E-Scooter Users N = 465						Non-Users N = 732
Car Attitudes	Sig	Decision	Pairwise Comparisons of City				Sig	Decision	Pairwise Comparisons of City
			S1–S2	Adj. Sig. a					S1–S2	Adj. Sig. a
Car attitude 1	0.000	Reject	D.C.-LA	1.000			0.008	Reject	D.C.-Mi	0.383
			D.C.-Mi	0.000	***				D.C.-LA	0.006	**
			LA-Mi	0.000	***				Mi-LA	0.194
Car attitude 2	0.000	Reject	LA-D.C.	0.004	**		0.000	Reject	LA-D.C.	0.001	**
			LA-Mi	0.000	***				LA-Mi	0.000	***
			D.C.-Mi	0.042	*				D.C.-Mi	0.000	***
Car attitude 3	0.047	Reject	LA-Mi	0.270			0.000	Reject	LA-Mi	0.284
			LA-D.C.	0.054					LA-D.C.	0.000	***
			Mi-D.C.	1.000					Mi-D.C.	0.000	***
Car attitude 4	0.000	Reject	Mi-LA	0.000	***		0.000	Reject	Mi-LA	0.000	***
			Mi-D.C.	0.000	***				Mi-D.C.	0.000	***
			LA-D.C.	1.000					LA-D.C.	0.000	***
Car attitude 5	0.000	Reject	LA-D.C.	0.318			0.000	Reject	LA-D.C.	0.000	***
			LA-Mi	0.000	***				LA-Mi	0.000	***
			D.C.-Mi	0.000	***				D.C.-Mi	0.000	***
Car attitude 6	0.000	Reject	Mi-D.C.	1.000			0.000	Reject	Mi-D.C.	0.000	***
			Mi-LA	0.000	***				Mi-LA	0.000	***
			D.C.-LA	0.000	***				D.C.-LA	0.000	***
Car attitude 7	0.020	Reject	D.C.-LA	0.069			0.000	Reject	D.C.-Mi	0.000	***
			D.C.-Mi	0.049	*				D.C.-LA	0.000	***
			LA-Mi	1.000					Mi-LA	0.048	*

Notes: Significance levels for reference: * (p < 0.05), ** (p < 0.01), *** (p < 0.001). ‘Sig. ^a’ represents the adjusted significance values. Car use attitudes: 1. prefer no car; 2. want to have a car; 3. try to use public transit; 4. try to use non-motorized modes; 5. use tech and apps; 6. tech and apps are hard to use; 7. prefer to own than rent.

). While the initial test suggested rejecting the hypothesis for car use attitude 3 (i.e., trying to use public transit whenever I can), subsequent posthoc tests with Bonferroni correction found no significant differences among cities, supporting the null hypothesis. E-scooter users in Washington, D.C. and Los Angeles shared similarities in car use attitudes 1, 4, and 5, with Miami differing. Additionally, Miami and Washington, D.C. showed differences in car use attitude 7 related to new technologies, with low significance. Among non-users, significant differences were observed in all car use attitudes, except for attitudes 1 and 3, where only Washington, D.C. and Los Angeles exhibited differing perspectives for attitude 1, and Los Angeles and Miami shared similar viewpoints for attitude 3.

4.4. Travel Behavior Characteristics of E-Scooter Users

In earlier work, Yang et al. (2023) conducted an analysis of the frequency of shared e-scooter usage; however, their study did not consider both trip purpose and the frequency of e-scooter usage together [55]. The present study, on the other hand, included a comprehensive comparison among the three cities studied by considering both trip purpose and frequency of e-scooter use.

Figure 6 presents the frequency distribution of e-scooter use across Washington, D.C., Miami, and Los Angeles, offering insights into shared trends and distinctive patterns. It was found that in Washington, D.C., e-scooters were notably favored for occasional use in recreational activities, shopping, and social events, with a heightened frequency for commuting and for attending social events (23), recreation (19), and running errands (19). Miami residents reported varied e-scooter usage, emphasizing social activities and commuting as primary trip purposes. Notably, e-scooter use was particularly popular for recreation purposes (20 of respondents using them 1–2 times) and commuting (29, 3–4 times) in Miami during the past 30 days. Residents also utilized e-scooters for shopping, errands, and social activities. Los Angeles reported e-scooter use for various purposes, including recreational activities (17, occasionally, and 14, 1–2 times) and shopping or errands (15, 1–2 times). E-scooters in Los Angeles were also employed for commuting and social activities. These findings underscore the varied roles of e-scooter use across the three cities, serving both leisure and practical needs. Such distinctions highlight the unique preferences and diverse applications of e-scooters among residents in each city, underscoring the significance of tailoring urban mobility solutions to local needs.

4.5. Factor Analysis Results (Dimensionality and Reliability of Latent Variables)

Factor analysis aims to assess connections between observed variables and latent constructs while reducing dimensionality, utilizing Principal Component Analysis, Varimax rotation, and Kaiser Normalization. The results from this case study are summarized in

Table 7. Factor analysis results: Rotated Component Matrix (KMO and Bartlett’s test).

Observed Variables	Mean	Factor (Latent Constructs)
		1	2	3	4
E-scooter Attitudes_1	3.350	0.727
E-scooter Attitudes_2	2.944	0.712
E-scooter Attitudes_3	3.150	0.774
E-scooter Attitudes_4	3.835	0.705
E-scooter Attitudes_5	3.848	0.664
E-scooter Attitudes_6	3.185	0.625
Mode Choice Factors_1	3.921		0.534
Mode Choice Factors_2	4.085		0.699
Mode Choice Factors_3	4.269		0.789
Mode Choice Factors_4	3.654		0.720
Mode Choice Factors_5	4.127		0.732
Car use Attitudes_2	3.259			0.826
Car use Attitudes_3	3.086			0.553
Car use Attitudes_5	3.607			0.710
Car use Attitudes_6	3.223			−0.767
Car use Attitudes_1	2.305				−0.721
Car use Attitudes_4	3.973				0.642
Eigenvalues		3.6	2.7	2.0	1.4
of variance explained		21	16	12	8
Kaiser–Meyer–Olkin Measure of Sampling Adequacy.			0.755
Bartlett’s Test of Sphericity	Approx. Chi-Square		6120.813
	df		136
	Sig.		0.000

Notes: Extraction method: Principal Component Analysis. Rotation method: Varimax with Kaiser Normalization.

, and demonstrate that reliability and internal consistency for factors surpass accepted standards. The Kaiser–Meyer–Olkin Measure of Sampling Adequacy indicated satisfactory sampling adequacy, and Bartlett’s Test of Sphericity affirmed data suitability for factor analysis. The analysis successfully reduced dimensionality, revealing latent constructs for mode choice factors, e-scooter use attitudes, and car use attitudes. Notably, e-scooter attitudes and mode choice factors displayed substantial factor loadings, signifying robust associations. Car use attitudes exhibited a more intricate pattern, grouped into Factors 3 and 4. The exclusion of certain variables was based on factor loadings below 0.5, consistent with earlier studies [68,69]. The output highlighted distinct latent constructs (Factors 1 to 4) capturing a significant portion of data variability (56.6). Research that followed utilized these factors as latent variables for prediction models, offering insights into underlying factors influencing observed variables.

4.6. Model Results

4.6.1. Comparison and Evaluation of Different Models

Three ML models were considered in this study, namely Binary Logistic Regression [55,57,70], Decision Tree [39,71], and Random Forest [71,72]. The three models selected for comparison in

Table 8. Models performance.

Model	Class	Precision	Recall	F1-Score	Overall Accuracy
Binary Logistic Regression	0	0.81	0.87	0.84	0.74
	1	0.74	0.65	0.69
Decision Tree	0	0.82	0.76	0.79	0.79
	1	0.63	0.71	0.67
Random Forest	0	0.83	0.90	0.86	0.82
	1	0.79	0.70	0.74

Notes: Class 1: shared e-scooter users; Class 0: non-users.

were chosen based on their proven effectiveness and popularity in predictive modeling for transportation studies, as well as their ability to handle diverse data types and provide robust, accurate predictions. Table 8 presents performance metrics for these three ML models in the classification of e-scooter users (Class 1) and non-users (Class 0) across all datasets encompassing the three study cities. Binary variables were generated, assigning a value of 1 to individuals who were male, white, possessed a driver’s license, and were employed or students. Correspondingly, a value of 0 was assigned for the opposite scenarios. Ordinal scales were applied for age, education, income, and frequency of taxi/Uber and transit usage.

The evaluation of the findings showed that Random Forest achieved the highest accuracy (0.82) while Binary Logistic Regression showed the lowest accuracy (0.74). Precision values favored non-users, and recall performance mirrored precision for all methods. Notably, Random Forest had the highest F1-score, while Decision Tree had the lowest F1-score among the methods considered.

4.6.2. Evaluating Feature Impacts on e-Scooter Usage

After selecting the optimal prediction model, it is important to assess the impact of each feature on e-scooter usage. Logit models, while simple, may not match the effectiveness of more complex machine learning models [73]. Ensemble models, though intricate, pose challenges in interpretation [74]. To overcome these limitations, the study utilized SHAP analysis with Violin plots. Figure 7a illustrates the distribution of SHAP values for each variable in the Random Forest model. Wider sections indicate a higher concentration of SHAP values for a specific feature. Positive SHAP values imply features that positively contribute to predictions, while negative values indicate features with a negative impact. Notable findings include employment having a significant negative impact, consistent with previous research [55]. Factors 1 and 4 (living without a car and using non-motorized modes, respectively) are representative of positive attitudes toward e-scooters and perceptions and exhibit positive impacts. Increase in age shows an association with a decline in usage, while higher education, income, and being male positively impact e-scooter usage predictions.

4.6.3. Uncovering Key Predictors through SHAP Analysis across Study Cities

SHAP analysis and Random Forest were also used to evaluate the important features of e-scooter users in each city separately, offering a clearer understanding of the partial influences of explanatory variables and improving the interpretability of the prediction model. The stacked bar chart presents the mean SHAP values for each feature derived from the SHAP analysis to illustrate the differences in influential factors based on size and color. The three colors represent each city—Washington, D.C., Miami, and Los Angeles—and the size of each color represents the higher impact on the prediction.

According to Figure 7b, employment status stands out as the most crucial feature for predicting e-scooter use in Miami and Washington, D.C., aligning with the results obtained from analyzing all data combined. However, it is noteworthy that Factor 3 that represents attitudinal questions on car and technology usage is the most important predictor for e-scooter use in Los Angeles; however, it did not emerge as a significant factor for e-scooter use in the aggregate data analysis. The second most important feature is Factor 4, which represents living without a car and using non-motorized modes, and it holds the highest importance for predicting Miami e-scooter users. The education feature, ranking as the second highest in importance for predicting e-scooter use in Washington, D.C. after employment, stands out as an influential factor in the model. Factor 2 and the number of vehicles in the household also influence predictions, while other features have a comparatively lower impact on the predictive outcomes.

5. Discussion and Conclusions

This study aimed to compare profiles of e-scooter users (N = 465) and non-users (N = 732), and document mode choice factors and attitudes towards e-scooter use. The study also compared influential factors for predicting e-scooter use between three cities (Washington, D.C., Miami, and Los Angeles) using machine learning and SHAP analysis. Likert scale visualization and the Kruskal–Wallis test were employed to compare observed variables, while factor analysis was utilized to extract latent constructs. The study explored four research questions.

To address Research Question 1 (see Introduction), the study initially investigated variations in profiles between e-scooter users and non-users across three cities. The data analysis revealed that the e-scooter user profile showed male dominance, a concentration of travelers in the 25–39 age group, higher income representation, and varied educational backgrounds, with a substantial number holding at least a bachelor’s degree. The profile aligns with previous studies at other locations [27,29,53,63]. On the other hand, gender variation was balanced among non-users, with a concentration of non-users in the 30–39 age group. The ownership of vehicles among non-users was notably lower (compared to e-scooter users). Compared to e-scooter users, a higher portion of non-users were employed and while their education profile was similar to e-scooter users (bachelor’s or higher) as also reported in [55].

In addressing Research Question 2, the study conducted a comparative analysis of attitudes and perceptions among e-scooter users and non-users across Washington, D.C., Miami, and Los Angeles. By employing Likert scale visualizations, the research emphasized distinctions in responses to three specific attitudinal questions. The Kruskal–Wallis test revealed significant differences in mode choice factors, e-scooter attitudes, and car use attitudes. In terms of mode choice factors, e-scooter users in Washington, D.C. and Los Angeles consistently expressed similar views, whereas non-users exhibited divergent attitudes, particularly with respect to cost, comfort, and safety.

Furthermore, variations in e-scooter attitudes were identified among shared e-scooter users in Washington, D.C., Miami, and Los Angeles While shared perspectives were observed related to the safety of e-scooters and the availability of e-scooter parking, distinct viewpoints emerged on issues such as the adequacy of bike lanes, the impact of e-scooters on the city, and their potential influences on public transit usage. Interestingly, non-users displayed significant differences in attitudes across all three cities, reflecting diverse perspectives on the mentioned e-scooter attitudes.

Moreover, the study uncovered significant differences in car use attitudes between e-scooter users and non-users across the three cities, with variations in specific attitudes. Washington, D.C. and Los Angeles users exhibited similarities, while Miami differed, and significant distinctions among non-users were observed in all attitudes except attitudes 1 (living without a private car) and 3 (using public transit).

In the realm of travel behavior, the study unveiled diverse utilization patterns of e-scooters across Washington, D.C., Miami, and Los Angeles. In Washington, D.C., e-scooters were predominantly used for occasional recreational activities, shopping, and social events, with an increased frequency observed for commuting and attending social events. Miami residents showcased varied e-scooter usage, with a focus on social activities and commuting. Notably, e-scooters were favored for commuting, with 29 utilizing them 3–4 times in Miami. In Los Angeles, e-scooters serve a range of purposes, including recreational activities, shopping, errands, commuting, and social events, with the majority of usage occurring occasionally.

The factor analysis effectively reduced dimensionality, exposing latent constructs associated with mode choice factors, e-scooter attitudes, and car use attitudes. The findings revealed the extraction of four factors, each with Eigenvalues exceeding one, emphasizing the categorization of variables into clear latent constructs. This outcome enhanced our comprehension of the underlying factors that impact the observed variables.

Research Question 3 focused on whether these latent factors could impact prediction models. Three predictive models were considered, namely Binary Logistic Regression, Decision Tree, and Random Forest, for two classes of users, i.e., e-scooter users (Class 1) and non-users (Class 0) across datasets from three cities. The evaluation revealed that Binary Logistic Regression exhibited the least effective performance, while Random Forest showcased the highest overall accuracy. SHAP analysis results indicated that following implementation, latent Factors 4 and 1 from the factor analysis emerged as the most important variables, emphasizing the influential role of latent variables in impacting prediction models in addition to employment status. Additionally, age and the frequency of Uber usage were identified as influential variables across all datasets.

To compare important variables affecting mode choice across three cities for e-scooter users and address Research Question 4, the study utilized mean SHAP values. The results revealed that employment was the most crucial factor for e-scooter users in Miami and Washington, D.C., which is consistent with the model for all datasets, and Factor 3 was the most important variable for e-scooter users in Los Angeles. This indicates that influential variables differ across the three cities. Additionally, the study found that the second most important feature was Factor 4, which held significant importance for predicting Miami e-scooter use but had a less pronounced effect on predicting Washington, D.C. users. Surprisingly, the education feature ranked as the second highest in importance for predicting Washington, D.C. use (following employment), and it emerged as the seventh most influential factor in the model for differentiating e-scooter users and non-users. This finding underscores that each city has its own influential factors in predicting e-scooter users, influenced by participants’ demographics, perceptions, attitudes, and other contextual factors; thus, localized studies are important in order to captures local dynamics.

While this study successfully identified influential factors among e-scooter users, it did not comprehensively examine all variables that could potentially impact predictions. Addressing this issue is crucial for advancing research in this field. Subsequent studies could investigate broader datasets that incorporate additional explanatory variables such as last trip distance, trip duration, and public transit satisfaction, which may influence mode choice. Examining how these variables shape shared micromobility usage patterns would be a valuable avenue for future research. The scope of this research was limited to consideration of survey responses from three U.S. cities. Expansion of the dataset to include additional study locations is expected to refine the study findings and further increase the reliability of estimates on certain factors. Furthermore, including additional psychological and social factors in future surveys is expected to improve the current understanding of people’s perceptions of using e-scooters and the motivations behind their mode choices. Future studies should also consider environmental conditions (e.g., weather conditions, temperature variation), and specific city characteristics (e.g., being a capital city; a tourist destination) and address their likely impacts on e-scooter usage.