Willingness to Pay for Station Access Transport: A Mixed Logit Model with Heterogeneous Travel Time Valuation

Abstract

This study estimates a willingness-to-pay (WTP) space mixed logit model to evaluate user valuations of travel time, safety, and comfort attributes associated with common access modes in Bangkok, including walking, motorcycle taxis, and localized minibuses. The model accounts for preference heterogeneity by specifying random parameters for travel time. Results indicate that users—exhibiting substantial variation in preferences—place higher value on reducing motorcycle taxi travel time, particularly in time-constrained contexts such as peak-hour commuting, whereas walking is more acceptable in less pressured settings. Safety and comfort attributes—such as helmet availability, smooth pavement, and seating—significantly influence access mode choice. Notably, the WTP for helmet availability is estimated at THB 8.04 per trip, equivalent to approximately 40% of the typical fare for station access, underscoring the importance of safety provision. Women exhibit stronger preferences for motorized access modes, reflecting heightened sensitivity to environmental and social conditions. This study represents one of the first applications of WTP-space modeling for valuing informal station access transport in Southeast Asia, offering context-specific and segment-level estimates. These findings support targeted interventions—including differentiated pricing, safety regulations, and service quality enhancements—to strengthen first-/last-mile connectivity. The results provide policy-relevant evidence to advance equitable and sustainable transport, particularly in rapidly urbanizing contexts aligned with SDG 11.2.

1. Introduction

Advancing the United Nations Sustainable Development Goals (SDGs)—particularly Goal 11 on “Sustainable Cities and Communities”—requires improving access to safe, affordable, and inclusive transport systems [1]. Urban rail plays a vital role in this agenda by enhancing mobility and reducing congestion in rapidly growing metropolitan regions. However, the effectiveness of rail networks hinges not only on in-vehicle service quality, but also on the ease, cost, and reliability of station access. In many cities, especially across the Global South, first- and last-mile access remains a persistent challenge that constrains overall transit ridership [2,3,4].

In Bangkok, the ongoing expansion of the metro system has improved network coverage, yet many neighborhoods still lack formal feeder services [5]. As a result, travelers frequently rely on walking, motorcycle taxis, and localized minibus to reach metro rail stations. These access modes vary significantly in travel time, comfort, and service quality, making the choice of access mode a key component of the overall user experience.

While many studies have applied discrete choice models to estimate user preferences, most focus on trunk-line or formal modes of travel [6,7]. Relatively few studies directly examine users’ willingness to pay (WTP) for improvements in informal or semi-formal access services—such as cleaner helmets, shaded walking paths, or available seating—despite the importance of such features in shaping access behavior. Furthermore, conventional preference-space models often require post-estimation conversion to WTP metrics, and may obscure individual-level valuation differences.

This study addresses these gaps by estimating a WTP-space mixed logit model to quantify user valuations of travel time, safety, and comfort attributes associated with common access modes in Bangkok. The model directly estimates marginal rates of substitution in monetary terms and captures preference heterogeneity through random parameter specifications. This approach offers more intuitive, policy-relevant outputs that can guide infrastructure investment, pricing mechanisms, and service design.

By focusing on informal and semi-formal station access modes—walking, motorcycle taxis, and localized minibuses—this study contributes to a more behaviorally grounded understanding of access decisions in a rapidly urbanizing city context. The findings provide empirical evidence for the economic value users assign to access attributes and offer actionable insights for improving first-/last-mile connectivity.

The remainder of this paper is structured as follows: Section 2 reviews the relevant literature on access mode valuation and modeling approaches. Section 3 outlines the methodological framework. Section 4 presents the estimation results. Section 5 discusses the key findings and policy implications, followed by concluding remarks in Section 6.

2. Literature Review

2.1. Transport Modes for Short-Distance Access to Rail Stations

The effectiveness of urban rail systems depends not only on in-vehicle service quality, but also on the ease with which users can access the network. For many travelers, the first or last segment of a journey—typically covering a short distance—plays a decisive role in whether rail is a viable option [8]. A variety of transport modes are available for these access trips, each offering a distinct combination of travel time, cost, flexibility, and service characteristics [9].

Walking is often the default mode for short-distance access, due to its simplicity and zero direct cost. However, its viability depends on the local context, including distance, urban form, and user-specific constraints. When walking is not feasible or desirable [10], short-haul motorized alternatives such as motorcycle taxis and shared minibuses (e.g., Songthaews in Thailand) become important. Motorcycle taxis are valued for their speed and door-to-door convenience, especially in congested or low-connectivity areas [11]. Songthaews, operating on semi-fixed routes, offer low fares and broad coverage, albeit with potential downsides in terms of waiting times and comfort [12].

Other modes—such as conventional buses, taxis, and ride-hailing services—can serve station access trips but are often less suitable for very short distances, due to higher costs, lower availability, or inefficiencies in local routing.

Given the diversity of available options, commuters make access decisions by trading off observable mode attributes such as travel time, cost, and service availability. While unobserved influences such as perceived safety, comfort, or reliability are not explicitly modeled, they are implicitly reflected in alternative-specific constants and their interactions with socio-demographic variables. Understanding how users evaluate and trade off key attributes is essential for analyzing access mode preferences and informing policy decisions.

2.2. Valuing Transport Attributes Using Discrete Choice Modeling

An expanding body of research has applied discrete choice models to quantify users’ preferences and trade-offs among transport attributes, including cost, travel time, reliability, and comfort. As summarized in Table 1, most studies adopt a preference-space specification, wherein utility coefficients are directly estimated and subsequently used to derive willingness-to-pay (WTP) values [13]. Examples include applications to Mobility-as-a-Service [14], informal motorcycle taxis in Uganda and Bangladesh [15], and electric vehicle charging preferences in Norway [16]. Similar modeling frameworks have been extended to emerging transport contexts, such as shared autonomous vehicles [17,18] and public transit services during emergency conditions [19].

More recent studies have shifted toward WTP-space modeling, where marginal rates of substitution are estimated directly, rather than being derived post-estimation from utility coefficients [20]. This approach produces outputs that are more intuitive and policy relevant, particularly when evaluating user valuations of time, safety, or comfort attributes. For instance, one study estimated WTP for access time and real-time information in the context of shared auto-rickshaws [21], while another applied a stated choice experiment to examine commute preferences, revealing significant variation in WTP for both travel and work time [22]. Although recent efforts have incorporated preference heterogeneity using latent class models and hybrid choice frameworks [23,24], most of them were estimated in preference space, where variation is captured at the utility coefficient level, rather than directly in monetary terms.

2.3. Valuation of Helmet Safety

Several studies have evaluated users’ willingness to pay (WTP) for motorcycle helmets as a means of reducing injury or fatality risk, offering useful benchmarks for safety-related valuation in access transport. Using contingent valuation methods, studies in Thailand, Vietnam, and Myanmar consistently report positive WTP for helmets that reduce severe injury or death risk by 50%, with mean estimates ranging from approximately USD 6 to USD 29 [25,26,27]. These valuations often exceed the prevailing market price of standard helmets, indicating users’ recognition of the safety benefit. Key determinants of WTP include income, prior accident experience, habitual helmet use, and perceived effectiveness of safety equipment. More recent work in Thailand applying behavioral modeling under the Theory of Planned Behavior (TPB) further highlights the influence of safety attitudes and behavioral intentions on helmet-related WTP [28]. These findings provide empirical support for incorporating helmet provision as a valued safety attribute in station access mode choice, particularly for motorcycle taxi users.

2.4. Valuation of Service Quality in Access Transport

Service quality attributes—particularly seating availability, ride comfort, and crowding—are consistently valued by users, as shown in willingness-to-pay (WTP) estimates. Although most evidence comes from mainline rail and bus systems, the findings are relevant to short-distance access modes such as motorcycle taxis, feeder buses, and informal services [29,30].

Comfort-related factors such as avoiding standing or crowding carry high disutility [31], with estimated WTP ranging from USD 0.87 to USD 2.43 per trip. For example, users were reportedly willing to pay up to EUR 5.88 per trip to avoid bus overcrowding, and as much as 12.05 pence per minute to avoid standing—more than twice the standard value of time [31].

These findings indicate that seat availability and comfort are not marginal preferences, but core components of perceived service quality. In the context of access transport—often characterized by small, unregulated, and crowded vehicles—enhancing comfort and regulating passenger loads can yield meaningful user benefits and support targeted improvements.

Table 1. Literature Reviews: Willingness to Pay (WTP) for Transport Attributes. Existing studies focus largely on formal transport systems in high-income settings, with limited attention to informal access modes and safety-service attributes in developing contexts. Most rely on preference-space models, limiting direct policy interpretation.

Study	Location	Alternative(s)	Attributes Valued	Model Type	Key Findings/WTP Range
Sillano and Ortúzar (2005) [13]	Santiago, Chile	Residential location (SP)	Travel time to work/study, air pollution, rent	Mixed Logit (preference space)	WTP (CLP): TTW = 36–51/min; TTS = 22–31/min
Eboli and Mazzulla (2008) [32]	Cosenza, Italy	Urban bus	Overcrowding, reliability, cleanliness, stop amenities, personnel attitude, info	Mixed Logit (preference space)	WTP to avoid overcrowding: EUR 5.88; reliability: EUR 12.24; cleanliness: EUR 7.18; personnel: EUR 5.07; info: EUR 4.83
Wang et al. (2018) [19]	New York City, NY, USA	Urban rail and transit	Disaster-recovery time	Mixed–Mixed Logit (preference space)	WTP (USD): 15–775/year depending on recovery time and latent class
Ardeshiri (2021) [24]	New South Wales, Australia	Shared parking	Space size, security, restrooms, walking distance	Latent Class Choice Model (preference space)	WTP (AUD): CCTV = AUD 1.01; restroom = AUD 0.54; full-day access = AUD 0.85, etc.
Tsouros et al. (2021) [14]	Greater Manchester, UK	MaaS (PT, Car-share, Bike, Taxi)	Service access	Multinomial Logit (preference space)	WTP: PT = EUR 97–119/month; Taxi = EUR 9.9–12/trip; Bike = EUR 23.5–29/month
Wang et al. (2021) [18]	Greater Toronto, ON, Canada	Private AV, Shared AV, Conventional Car	Automation level, error risk, socio-demographics	Mixed MNL/Nested Logit (preference space)	WTP for Level 4 AV (CAD): CAD 10,800–29,800; varies by user group
Bwambale et al. (2023) [15]	Uganda, Bangladesh	Public transport, motorcycle taxis	Safety (helmet), hygiene, social distancing	Multinomial Logit (preference space)	WTP (USD): helmet = USD 0.22–0.42; transparent shield = USD 0.62–0.63; safety significant
Das et al. (2023) [21]	Mumbai, India	Shared auto-rickshaw	Travel time reliability, access time, waiting time, real-time info	Mixed Logit (WTP space)	WTP (INR): Reliability = 0.28/min; Access = 0.26/min; Real-time info = 21.56/trip
Hoen et al. (2023) [16]	Norway	EV charging	Wait time, availability, cost, urgency	Mixed Logit (preference space)	WTP: EUR 30/h to reduce wait time; varies by group
Liu and Liu (2023) [33]	Wuhan, China	Last-mile shared AV	Travel time, waiting time, in-vehicle time, AV attitudes	Hybrid Choice Model (preference space)	WTP (CNY): walk/wait = CNY 11.0–441.3/h; in-vehicle = CNY 20.4–63.0/h; strong heterogeneity in AV preference
Guan et al. (2024) [17]	China	Subscription AV, Self-owned AV, Shared AV	Subscription cost, ownership cost, service quality	Mixed Logit (preference space)	WTP (CNY): AVS = CNY 3000/month; Self-owned AV = +CNY 130,000; SAV = CNY 418/month
Kang et al. (2024) [23]	Seoul, South Korea	Shared e-scooter	Exclusive lane, driving info, maintenance, docking	Latent Class Logit (preference space)	WTP: exclusive lane +47.1%, info +33.8%, maintenance +32.2%; docking not significant
Poudel and Singleton (2024) [22]	United States	Commute (mode-agnostic)	Travel time, work time, income	Mixed Logit (WTP space)	WTP: Travel = USD 18.56/h; Work = USD 5.77/h; travel time valued more by most users
This study	Bangkok, Thailand	Walk, Motorcycle Taxi, Songthaew	Travel time, walkway, helmet, seating	Mixed Logit (WTP space)	WTP (THB): Helmet = 8.04; Seat = 23.65; Pavement = 15.72; Walk time = 3.35/min; MCT time = 7.43/min ST time = 1.23/min

Table 1. Literature Reviews: Willingness to Pay (WTP) for Transport Attributes. Existing studies focus largely on formal transport systems in high-income settings, with limited attention to informal access modes and safety-service attributes in developing contexts. Most rely on preference-space models, limiting direct policy interpretation.

Study	Location	Alternative(s)	Attributes Valued	Model Type	Key Findings/WTP Range
Sillano and Ortúzar (2005) [13]	Santiago, Chile	Residential location (SP)	Travel time to work/study, air pollution, rent	Mixed Logit (preference space)	WTP (CLP): TTW = 36–51/min; TTS = 22–31/min
Eboli and Mazzulla (2008) [32]	Cosenza, Italy	Urban bus	Overcrowding, reliability, cleanliness, stop amenities, personnel attitude, info	Mixed Logit (preference space)	WTP to avoid overcrowding: EUR 5.88; reliability: EUR 12.24; cleanliness: EUR 7.18; personnel: EUR 5.07; info: EUR 4.83
Wang et al. (2018) [19]	New York City, NY, USA	Urban rail and transit	Disaster-recovery time	Mixed–Mixed Logit (preference space)	WTP (USD): 15–775/year depending on recovery time and latent class
Ardeshiri (2021) [24]	New South Wales, Australia	Shared parking	Space size, security, restrooms, walking distance	Latent Class Choice Model (preference space)	WTP (AUD): CCTV = AUD 1.01; restroom = AUD 0.54; full-day access = AUD 0.85, etc.
Tsouros et al. (2021) [14]	Greater Manchester, UK	MaaS (PT, Car-share, Bike, Taxi)	Service access	Multinomial Logit (preference space)	WTP: PT = EUR 97–119/month; Taxi = EUR 9.9–12/trip; Bike = EUR 23.5–29/month
Wang et al. (2021) [18]	Greater Toronto, ON, Canada	Private AV, Shared AV, Conventional Car	Automation level, error risk, socio-demographics	Mixed MNL/Nested Logit (preference space)	WTP for Level 4 AV (CAD): CAD 10,800–29,800; varies by user group
Bwambale et al. (2023) [15]	Uganda, Bangladesh	Public transport, motorcycle taxis	Safety (helmet), hygiene, social distancing	Multinomial Logit (preference space)	WTP (USD): helmet = USD 0.22–0.42; transparent shield = USD 0.62–0.63; safety significant
Das et al. (2023) [21]	Mumbai, India	Shared auto-rickshaw	Travel time reliability, access time, waiting time, real-time info	Mixed Logit (WTP space)	WTP (INR): Reliability = 0.28/min; Access = 0.26/min; Real-time info = 21.56/trip
Hoen et al. (2023) [16]	Norway	EV charging	Wait time, availability, cost, urgency	Mixed Logit (preference space)	WTP: EUR 30/h to reduce wait time; varies by group
Liu and Liu (2023) [33]	Wuhan, China	Last-mile shared AV	Travel time, waiting time, in-vehicle time, AV attitudes	Hybrid Choice Model (preference space)	WTP (CNY): walk/wait = CNY 11.0–441.3/h; in-vehicle = CNY 20.4–63.0/h; strong heterogeneity in AV preference
Guan et al. (2024) [17]	China	Subscription AV, Self-owned AV, Shared AV	Subscription cost, ownership cost, service quality	Mixed Logit (preference space)	WTP (CNY): AVS = CNY 3000/month; Self-owned AV = +CNY 130,000; SAV = CNY 418/month
Kang et al. (2024) [23]	Seoul, South Korea	Shared e-scooter	Exclusive lane, driving info, maintenance, docking	Latent Class Logit (preference space)	WTP: exclusive lane +47.1%, info +33.8%, maintenance +32.2%; docking not significant
Poudel and Singleton (2024) [22]	United States	Commute (mode-agnostic)	Travel time, work time, income	Mixed Logit (WTP space)	WTP: Travel = USD 18.56/h; Work = USD 5.77/h; travel time valued more by most users
This study	Bangkok, Thailand	Walk, Motorcycle Taxi, Songthaew	Travel time, walkway, helmet, seating	Mixed Logit (WTP space)	WTP (THB): Helmet = 8.04; Seat = 23.65; Pavement = 15.72; Walk time = 3.35/min; MCT time = 7.43/min ST time = 1.23/min

2.5. Research Gap and Objectives

Despite increasing attention to improving urban rail access, several research gaps remain. First, most prior studies rely on preference-space models, which do not directly yield users’ monetary valuation of service attributes. Although WTP-space models provide more intuitive and policy-relevant outputs, they have rarely been applied to the context of informal access modes—such as walking, motorcycle taxis, and Songthaews—particularly in developing cities.

Second, empirical evidence on preference heterogeneity in station access mode choice remains limited. Understanding how users value travel time, safety, and infrastructure quality across socio-demographic groups and trip contexts is essential for designing inclusive and responsive mobility strategies.

To address these gaps, this study estimates users’ willingness to pay for key access attributes—including travel time, helmet availability, seating, and pedestrian infrastructure—using a WTP-space mixed logit model. The model incorporates unobserved preference heterogeneity to reflect variation in user sensitivities. By doing so, the study aims to generate behaviorally realistic and policy-relevant insights that support more effective infrastructure investment and pricing strategies for enhancing metro station accessibility in Bangkok.

3. Methodology

3.1. Stated Preference Survey Design and Attribute Specification

A stated preference (SP) survey was designed to examine commuters’ mode choices for accessing metro stations in Bangkok under varying service and infrastructure conditions. The survey targeted regular metro users who frequently access the station by either walking or using motorcycle taxis or Songthaews. Each SP scenario presented respondents with a choice among three station access mode alternatives—Walk, Motorcycle Taxi (MC), and Songthaew (ST). The survey limited the alternatives to these three modes to reflect the most commonly used options in the study context while ensuring that respondents could make realistic and informed trade-offs without fatigue.

The alternatives are described by five key attributes: travel time, travel cost, and three qualitative factors reflecting walkability, safety, and comfort. For the walking alternative, these included pavement condition (smooth vs. uneven), availability of shading (e.g., trees or awnings), and the presence of obstacles (e.g., parked vehicles or street vendors). For the motorcycle taxi, a safety indicator was included to denote whether the passenger would be provided with a helmet. For Songthaews, the availability of a seat was included to capture comfort and service quality.

Although attributes such as lighting and pollution such as noise and dust are potentially relevant concerns for walking or, similarly, driver reliability, vehicle condition, and insurance coverage for motorcycle taxis and Songthaew, they were excluded from the SP design to reduce respondent burden and minimize cognitive overload. Similarly, for motorcycle taxis and Songthaew, factors such as waiting time, driver reliability, vehicle condition, or insurance coverage were omitted to maintain a manageable number of attributes per alternative. Instead, the SP design prioritized a parsimonious set of attributes that capture the most salient aspects of walkability and service quality for short-distance station access, which were obtained from the pre-observation survey.

The SP design comprised two blocks, each containing six choice tasks. Each respondent completed one block, resulting in six hypothetical scenarios covering access distances of 1 km, 1.5 km, and 2 km (two tasks per distance). In each task, respondents chose their preferred access mode based on randomized combinations of mode-specific attribute levels. These levels included walking time, pavement quality, shading, and obstacle presence for walk; travel time, cost, and helmet availability for motorcycle taxis; and travel time, cost, and seating availability for Songthaews.

The final design achieved a D-error of 0.2082, which serves as a useful benchmark for assessing internal consistency and suggests that the design is suitably informative for parameter estimation under the assumed priors [34]. The attribute structure is summarized in

Table 2. Access Mode Alternatives and Attribute Levels Used in the Stated Preference Survey. Three access modes—walking, motorcycle taxi, and Songthaew—are described by service attributes related to travel time, cost, safety, and comfort. Attribute levels were designed to reflect realistic access conditions to metro stations in Bangkok.

Mode	Factor	Level
Walk	Walking time	15, 20, 30 min
	Good pavement	Yes, no
	With roof or shade	Yes, no
	Obstacle-free	Yes, no
Motorcycle taxi	Total travel time	4, 6, 8 min
	Total travel cost	15, 25, 30 THB (USD 0.46, USD 0.77, USD 0.93)
	Helmet enforcement	Yes, no
Songthaew	Total travel time	10, 12, 15 min
	Total travel cost	THB 8, 10 (USD 0.25, USD 0.31)
	Seat available	Yes, no

Note: 1 United States Dollar (USD) = 32.41 Thai Baht (THB) as of 1 October 2024.

. The survey instrument was pretested with a pilot group to ensure clarity, consistency, and plausibility of the scenarios. During the SP experiment, representative origin–destination pairs were also introduced to help respondents visualize each given distance. An example of a choice task presented to respondents is shown in Figure 1.

The main survey was conducted in January 2024 at various residential-access metro stations along Bangkok’s Purple Line. In addition to the six stated choice tasks, respondents also provided socio-demographic information, including gender, age, income, car ownership, and commuting patterns.

As of November 2023, the Bangkok metro system served approximately 764,147 daily passengers. Using a 95% confidence level and a 5% margin of error, the minimum required sample size for representativeness was 384. A total of 409 valid responses were collected, exceeding this threshold.

3.2. Sample Characteristics and Travel Behavior

The descriptive statistics of the sample are summarized in

Table 3. Socio-Demographic Profile and Rail Travel Behavior of Survey Respondents. Respondents represent typical urban rail commuters in Bangkok, characterized by moderate income levels, low private car ownership, and frequent use of the rail system. Most access stations within short distances, primarily by walking, motorcycle taxi, or car drop-off.

Characteristic	Category	Frequency (n)	Proportion (%)
Gender	Male	204	49.9
	Female	205	50.1
Age	Under 20 years	28	6.8%
	21–25 years	53	13.0%
	26–30 years	135	33.0%
	31–35 years	90	22.0%
	36–40 years	43	10.5%
	41–45 years	32	7.8%
	46–50 years	18	4.4%
	Over 50 years	10	2.4%
Income	Less than THB 20,000	100	24.4%
	THB 20,000–30,000	202	49.4%
	More than THB 30,000	107	26.2%
Private car ownership	Yes	111	27.1%
	No	298	72.9%
Motorcycle ownership	Yes	188	46%
	No	221	54%
Number of rail travel days/week	1–2 days	68	16.6%
	3 days	86	21.0%
	4 days	53	13.0%
	5 days	143	35.0%
	6–7 days	59	14.4%
Station access mode	Walking	157	38.4%
	Motorcycle taxi	118	28.9%
	Bus or Songthaew	9	2.2%
	Car (drop-off)	120	29.3%
	Taxi and others	5	1.2%
Station access distance	Less than 0.5 km	0	0.0%
	0.5–1 km	350	85.6%
	1–2 km	43	10.5%
	2–4 km	13	3.2%
	More than 4 km	3	0.7%
Station access time	Less than 5 min	209	51.1%
	6–10 min	189	46.2%
	11–15 min	8	2.0%
	More than 15 min	3	0.7%

. The final dataset comprised 409 respondents, with a near-equal gender distribution: 50.1% female and 49.9% male, closely aligning with national population gender ratios in Thailand. The sample skewed young, with 33.0% aged 26–30 years, followed by 22.0% aged 31–35 and 13.0% aged 21–25. Only 2.4% of respondents were over 50 years old.

In terms of income, nearly half of the participants (49.4%) reported monthly earnings in the THB 20,000–30,000 range. This was followed by 26.2% earning more than THB 30,000, and 24.4% earning less than THB 20,000. Rates of private vehicle ownership were relatively low, with 27.1% owning a private car and 46.0% owning a motorcycle.

Rail usage patterns indicated a high frequency of commuting: 35.0% of respondents reported using rail five days per week, and 14.4% used it six to seven days per week. Regarding access mode choice, walking (38.4%), motorcycle taxis (28.9%), and car drop-offs (29.3%) were the most commonly reported. Use of Songthaews was low (2.2%), and only 1.2% accessed the station via taxis or other modes.

Access distances were generally short. A majority (85.6%) lived within 0.5–1 km of a metro station, while 96.1% resided within 2 km. In terms of time, 51.1% of respondents reached the station in under 5 min, and 46.2% within 6–10 min. Only 2.7% reported access times exceeding 10 min.

Regarding access cost, most respondents (68.7%) did not incur any monetary expense. Among those who paid, 19.6% spent up to THB 15, and only 2.0% reported costs above THB 20.

While the overall sample (n = 409) meets the minimum threshold for statistical representativeness, the distribution of actual access modes is uneven. Only 2.2% of respondents reported using Songthaews as their typical station access mode. This low observed share reflects current travel patterns in the surveyed urban catchment area, where Songthaews are less prevalent and primarily serve peripheral or suburban zones. Despite this, Songthaew was retained in the stated preference (SP) experiment to examine user trade-offs related to service quality attributes—such as the availability of seating or scheduled operations—which are not applicable to walking or motorcycle taxis. Including this mode allows the analysis to assess how users value service improvements in shared access transport, even if current usage is limited.

3.3. Discrete Choice Modeling Approach in WTP Space

This study employs a WTP-space specification to directly estimate marginal willingness-to-pay (WTP) for non-market attributes such as helmet availability, pavement quality, and seating. Compared to the preference-space framework, this approach produces more interpretable and policy-relevant outputs in monetary terms. It also avoids the instability associated with post-estimation transformations—particularly when the cost coefficient is imprecisely estimated or correlated with other parameters. As discussed by Train and Weeks (2005) [20], specifying the model in WTP space also facilitates the imposition of economically meaningful constraints—such as strictly positive WTP—and avoids the difficulties associated with estimating random cost parameters, thereby improving behavioral realism and estimation robustness.

(1)

Multinomial Logit (MNL) Model Specification

As a baseline specification, a Multinomial Logit (MNL) model was estimated to examine mode choice behavior. All coefficients were treated as fixed across individuals, and the unobserved error terms were assumed to follow an independently and identically distributed (i.i.d.) Gumbel distribution. The utility of each alternative was specified in willingness to pay (WTP) space, where all non-cost attribute coefficients are expressed in monetary terms (THB: Thai Baht) by normalizing or scaling the marginal utility of cost ${\beta }_c$. This allows direct interpretation of estimated parameters such as users’ WTP for improvements in safety, comfort, or travel time.

While the MNL model offers a straightforward structure and computational efficiency, it assumes homogeneous preferences across individuals. As such, it does not capture unobserved taste heterogeneity in attribute valuation. These limitations motivate the use of a more flexible Mixed Logit model, which incorporates random parameters to better represent individual-level variation in preferences.

(2)

Mixed Logit (ML) Model Specification and Estimation

To account for heterogeneity in preferences, a Mixed Logit (ML) model was estimated in WTP space, where willingness-to-pay for each attribute is directly specified, rather than derived as a ratio of coefficients. This specification improves interpretability, ensures economically consistent signs (e.g., strictly positive WTP for travel time), and facilitates the direct valuation of non-market attributes such as comfort, safety, and walkability.

The utility $U_{nit}$ that individual n derives from choosing access mode i is given as

$$U_{ni}={\delta }_i+{\beta }_c\left({\displaystyle \sum _k{v}_k^{(n)}\cdot x_{kni}}\right)+{\epsilon }_{ni}$$

(1)

where ${\delta }_i$ is the alternative-specific constant (ASC) for mode i, ${\beta }_c$ is the marginal utility of cost (used for scaling), $v_k^{(n)}$ is the WTP for attribute k, and $x_{kni}$ is the level of that attribute. The error term ${\epsilon }_{ni}$ is independently and identically Gumbel distributed.

To capture heterogeneity in time valuation, random coefficients were specified for walk and motorcycle taxi (MC) travel time. These coefficients were modeled as lognormally distributed, to ensure strictly positive values.

For walk and MC travel time, the individual-specific coefficients are defined as

$$\begin{array}{c}{v}_{tt,walk,n}=-\exp \left({\mu }_{vtt,walk}+{\sigma }_{vtt,walk}\cdot {\eta }_{walk,n}\right)\\ v_{tt,mc,n}=-\exp \left({\mu }_{vtt,mc}+{\sigma }_{vtt,mc}\cdot {\eta }_{mc,n}\right)\end{array}$$

(2)

where ${\mu }_{vtt,walk}$ and ${\mu }_{vtt,mc}$ are the log-means; ${\sigma }_{vtt,walk}$ and ${\sigma }_{vtt,mc}$ are the log-standard deviations; and ${\eta }_{walk}^{(n)}$ and ${\eta }_{mc}^{(n)}$ are standard normal draws. The exponential form ensures positive magnitudes, while the negative sign imposes strictly negative coefficients, consistent with travel time disutility.

The utility functions for each mode are specified as

$$\begin{array}{cc}\hfill U_{walk,n}& ={\delta }_{walk}+{\beta }_c\left(v_{tt,walk,n}\cdot Tim{e}_{walk,n}+v_{pave}\cdot Pavemen{t}_n\right)\hfill \\ \hfill U_{mc,n}& ={\delta }_{mc}+{\gamma }_{female}\times Femal{e}_n+{\beta }_c\left(v_{tt,mc,n}\times Tim{e}_{mc,n}+Cos{t}_{mc,n}+v_{helmet}\times Helme{t}_n\right)\hfill \\ \hfill U_{st,n}& ={\delta }_{st}+{\gamma }_{female}\times Femal{e}_n+{\beta }_c\left(v_{tt,st,n}\cdot Tim{e}_{st,n}+Cos{t}_{st,n}+v_{seat}\cdot Seatin{g}_n\right)\hfill \end{array}$$

(3)

where ${\delta }_{walk}=0$ (normalization); ${\delta }_{mc}$ and ${\delta }_{st}$ are ASCs for motorcycle taxi and Songthaew; ${\gamma }_{female}$ captures preference for motorized modes among female users; $v_{tt,st}$ is a fixed WTP for Songthaew travel time; and $v_{pave}$, $v_{helmet}$ and $v_{seat}$ are the fixed WTP for pavement quality, helmet availability, and seating availability, respectively.

The MNL and ML models were estimated using the Apollo 0.3.5 package in R [35]. To address potential bias arising from repeated choice tasks—a common issue in stated preference (SP) surveys—a panel data structure was employed. Rather than treating all 409 × 6 responses as independent observations, the model calculates the joint probability of choices at the individual level, thereby capturing within-respondent correlation. As the ML model does not have a closed-form solution, simulated maximum likelihood estimation was adopted, with integrals over random parameters approximated using Halton sequences and 500 inter-individual draws. Model fit was evaluated using log-likelihood, Akaike Information Criterion (AIC), and Bayesian Information Criterion (BIC).

4. Results

4.1. Model Estimation Outcomes

To examine user preferences for metro station access modes, both a Multinomial Logit (MNL) model and a Mixed Logit (ML) model were estimated in WTP space. The estimation results are summarized in

Table 4. Estimated Parameters of the Multinomial Logit (MNL) and Mixed Logit (ML) Models in WTP Space for Station Access Mode Choice.

	MNL (WTP Space)		ML (WTP Space)
	Estimate	Robust t-Stat.	Estimate	Robust t-Stat.	WTP (THB)
Alternative-specific constants
Walk (δwalk)	Reference		Reference
MCT (δmc)	–0.855 *	–1.913	–2.983 ***	–5.040
ST (δst)	–2.617 ***	–5.282	–5.707 ***	–8.448
User characteristic	0.721 ***	2.9931	1.350 ***	3.967
Female
Travel cost coefficient	0.023 *	1.681	0.118 ***	6.511
Travel Time coefficients
Walk travel time (vtt,walk)	–6.683	–1.621	—	—
Walk travel time (log μtt,walk)	—	—	1.135 ***	6.769	Mean = 3.35/min
Walk travel time (σtt,walk)	—	—	0.382 ***	5.974	S.D. = 1.33/min
MCT travel time	–8.158 **	–2.418	—	—
MCT travel time (log μtt,mc)	—	—	1.761 ***	14.406	Mean = 7.43/min
MCT travel time (σtt,mc)	—	—	0.699 ***	12.741	S.D. = 5.89/min
ST travel time (vtt,st)	–0.318	–0.334	–1.229 ***	–3.555	1.229/min
Safety and Comfort coefficients
Pavement quality (vpave)	44.786	1.547	15.722 ***	4.24	15.72/trip
Helmet availability (vmc)	11.478 **	2.099	8.039 ***	5.091	8.04/trip
Seating availability (vhelmet)	69.286	1.581	23.647 ***	5.606	23.65/trip
Model fit
Log-likelihood	–1930.95		–1551.72
Adjusted ρ2	0.2801		0.4200
AIC	3881.90		3127.43
BIC	3939.96		3197.10

*, **, *** indicate significance at the 0.10, 0.05, and 0.01 levels, respectively, based on robust t-statistics.

.

(1)

Multinomial Logit (MNL) Model Results

The MNL model includes alternative-specific constants (ASCs) for motorcycle taxi (MCT) and Songthaew (ST), a generic female dummy, fixed travel time coefficients for walk, MCT, and ST, and fixed WTP parameters for pavement quality, helmet availability, and seating. Among these, the ASCs, female dummy, and MCT travel time are statistically significant at conventional levels. Helmet availability also yields a significant and positive WTP, suggesting that users place value on safety features. In contrast, pavement quality, seating availability, and ST travel time exhibit positive but statistically insignificant coefficients.

(2)

Mixed Logit (ML) Model Results

The ML model extends the MNL specification by introducing random coefficients for walk and MCT travel time, modeled using lognormal transformations to ensure strictly positive WTP. All estimated parameters—including log-means, standard deviations, and attribute coefficients—are statistically significant at the 0.01 level.

Compared to the MNL model, the ML model shows substantially improved fit: the adjusted ${\rho }^2$ increases from 0.2801 to 0.4200, and both AIC and BIC are considerably lower. These improvements reflect the ML model’s greater capacity to capture unobserved preference heterogeneity, particularly in the valuation of travel time. Accordingly, the following discussion focuses on the ML model results.

4.2. Mode Preferences and Gender Effects

The ASCs capture intrinsic preferences for each access mode relative to walking, which is normalized to zero. The ASCs for motorcycle taxi (MCT) and Songthaew (ST) are significantly negative, indicating that—holding other factors constant—users prefer walking over these motorized alternative modes. The disutility associated with MCT increases from –0.855 in the MNL model to –2.983 in the ML model, and for ST from –2.617 to –5.707. These stronger negative values in the ML model suggest that, once preference heterogeneity is accounted for, walking becomes an even more preferred baseline, particularly among users sensitive to cost or travel time.

However, it is important to note that these negative ASCs should not be interpreted as a general rejection of MCT and ST modes, which remain prevalent, in practice. The ASC values reflect relative baseline utility after controlling for observed service attributes—such as travel time, fare, helmet provision, or seating—and may capture unobserved factors such as safety concerns, regulatory uncertainty, or inconsistent service quality. The widespread use of MCTs and Songthaews in Bangkok is often a reflection of availability, flexibility, and affordability—especially in areas lacking formal feeder services. Therefore, while the model indicates lower average utility for these modes, this does not contradict their real-world prevalence, which may be driven more by necessity than preference.

The female dummy is positive and statistically significant, indicating that female users have consistently higher utility for motorized access modes compared to male users. This may reflect greater sensitivity to environmental conditions, personal safety, or physical exertion.

4.3. Heterogeneity in Willingness to Pay for Travel Time

The ML model provides individual-specific WTP estimates for travel time, capturing heterogeneity across users while ensuring strictly positive valuations through the use of lognormal distributions. The estimated mean WTP for walking time is approximately THB 3.35/min, calculated as exp (1.135 + 0.382²/2), and for motorcycle taxi, it is THB 7.43/min, calculated as exp (1.761 + 0.699²/2). Both values are statistically significant and consistent with expectations.

The significant standard deviations associated with both WTP estimates highlight substantial heterogeneity. While the mean WTP for MC time is higher, its standard deviation is also larger—approximately THB 5.89/min—reflecting considerable variation in how users value time savings. This variability likely stems from differences in risk perception, urgency, or income constraints. In contrast, the standard deviation of WTP for walking time WTP is smaller, at around THB 1.33/min, indicating more consistent preferences across individuals.

For Songthaew, travel time is specified as a fixed parameter in the ML model and estimated at THB–1.229/min. This reflects general disutility from in-vehicle time, but does not account for user-specific heterogeneity

4.4. Valuation for Safety and Comfort Attributes

The ML model also estimates WTP for key safety and comfort attributes across modes, providing insights into the relative importance of service features beyond travel time.

For walking, the coefficient for pavement quality is estimated at THB 15.72 per trip, reflecting a strong preference for improved pedestrian environments. This finding underscores the value of investments in walking infrastructure—such as continuous sidewalks, smooth surfaces, and unobstructed paths—to enhance accessibility and comfort near metro stations.

For motorcycle taxis, the availability of a passenger helmet yields a WTP of approximately THB 8.04 per trip, equivalent to 40–50% of the typical fare for short station-access trips. This result highlights the importance of visible safety provisions, which may influence users’ perceptions of reliability, regulatory compliance, and personal protection.

For Songthaew services, seating availability yields a WTP of THB 23.65 per trip. This reflects the discomfort and disutility associated with standing or crowding, especially on longer or less frequent feeder trips. This suggests that users place considerable value on avoiding standing or crowding—particularly for longer or less frequent feeder services. Policy interventions such as improved vehicle design, higher service frequency, or passenger load regulation may significantly enhance user satisfaction and service attractiveness.

Compared to the MNL model, which yields similar directional effects but less stable estimates, the ML model produces more robust and interpretable WTP estimates. By accounting for heterogeneity in cost sensitivity, the ML approach enables clearer behavioral insights and supports more targeted recommendations for infrastructure improvements and service design.

Note that during model development, alternative specifications were tested with random parameters for safety and comfort attributes, including helmet availability, pavement quality, and seat availability. However, the estimated standard deviations for these parameters were not statistically significant, indicating limited unobserved heterogeneity in preferences for these attributes within the sample. To avoid overparameterization and preserve model stability, these variables were retained as fixed effects in the final specification.

5. Discussions and Policy Implications

5.1. Key Findings

This study provides insights into how metro users value various attributes of access transport, including travel time, safety features, and comfort elements. Using a WTP-space mixed logit model, the analysis highlights both generalizable behavioral patterns and context-specific nuances that reflect Bangkok’s unique urban environment.

(1)

Bangkok-Specific Conditions and Contextual Preferences

Several findings in this study reflect Bangkok’s specific climatic, infrastructural, and cultural conditions. A key result is the substantially higher willingness to pay (WTP) for travel time in the station access segment. In particular, the estimated value of time (VOT) for motorcycle taxi travel is THB 7.43 per minute (THB 445.8 per hour)—nearly four times higher than the THB 112.8 per hour VOT used in Bangkok’s travel demand models for metro users [36]. These official estimates are based solely on the in-vehicle metro segment and do not distinguish the access portion of the trip. This marked difference underscores the heightened time sensitivity in first-/last-mile segments, especially when users choose faster, paid options, such as motorcycle taxis. The elevated WTP likely reflects a combination of perceived urgency, the greater disutility of delay, and the importance of schedule adherence during peak periods. These findings suggest that applying a uniform VOT across all trip segments may lead to a systematic underestimation of access-related benefits in policy and investment evaluations.

Meanwhile, the result indicates a relatively lower willingness to pay (WTP) for walking, compared to motorized modes. Combined with prior evidence that average walking distances to metro stations in Bangkok are only about 320 m [37], this suggests a limited tolerance for pedestrian access, likely due to environmental discomfort—including high temperatures, lack of shade, and frequent obstacles such as parked vehicles and street vendors [10,38].

Another key contribution of this study is the explicit estimation of WTP for helmet availability in MCT services. The estimated value of THB 8.04 per trip, or about 40% of the typical fare, indicates strong demand for visible safety measures. While prior studies have examined WTP for helmet ownership or risk reduction [25,26,27,28], this study uniquely quantifies user valuation of helmet availability as part of semi-formal station access services, shedding lights on how users respond to Bangkok’s partially enforced safety regulations.

The high WTP for seating availability highlights the discomfort associated with standing or crowding. While crowding aversion is well researched, this result likely reflects Bangkok-specific service deficiencies, such as less comfort vehicles and low service reliability [21,31].

Gender differences also emerge as a locally influenced factor. Women show a stronger preference for motorcycle taxis or Songthaews over walking in first-/last-mile segments. These preferences appear to reflect behavioral adaptations to environmental and social conditions, rather than being driven purely by cost or distance. Bangkok’s fragmented pedestrian environment—marked by blocked sidewalks, encroachment, limited lighting, and heat exposure—creates both physical and psychological barriers. Safety concerns, particularly when walking alone or at night, further deter female pedestrians. These tendencies are consistent with previous studies linking women’s mode choice to greater sensitivity to safety, comfort, and cultural norms [4,14,39]. In this context, semi-formal modes like MCTs offer more socially acceptable and controlled alternatives, which women are willing to pay for. This behavior represents a form of compensatory adaptation, and highlights the importance of inclusive pedestrian infrastructure.

(2)

Broadly Consistent Patterns and Generalizable Insights

Several findings in this study are consistent with the prior literature, suggesting that station access behavior follows broadly recognizable patterns across settings.

First, walking is generally preferred over motorized modes such as motorcycle taxis (MCTs) and Songthaews for short distances, aligning with earlier studies that emphasize the simplicity, flexibility, and cost-free nature of walking for short access trips [14,15]. However, this preference tends to shift under time pressure.

The model reveals a higher WTP for MCT time (THB 7.34/min) compared to walk time (THB 3.35/min), indicating that speed and reliability are particularly valued during time-sensitive conditions. This supports prior findings that time savings are a dominant consideration in short-distance access decisions [19,21].

Furthermore, the WTP results demonstrate a temporal asymmetry in mode preferences. Walking remains acceptable when time pressure is low, such as during return trips or off-peak periods—a pattern also observed by [22], who highlight urgency as a determinant of travel time acceptability.

The study also uncovers substantial heterogeneity in WTP for walking, as reflected in the large standard deviation of the travel time coefficient. This suggests that individual-level factors such as physical ability, perceived safety, and trip purpose lead to varied preferences. These support prior calls for user segmentation in station access modeling [18,21].

Gendered travel patterns in first-/last-mile access are consistently observed across low- and middle-income contexts. Prior studies show that women frequently adjust their travel routes, timing, and mode choices to reduce exposure to unsafe, uncomfortable, or socially inappropriate conditions [40,41,42,43]. Broader evidence further confirms that safety concerns, harassment risks, and cultural expectations around visibility and propriety significantly constrain female mobility [44,45]. These constraints tend to be more severe among low-income women, who often report feeling marginalized or invisible within the transport system [45]. In response, many women adopt behavioral adaptations such as accepting longer or more expensive trips to minimize vulnerability, resulting in a heightened perceived burden of access [42,43].

Finally, the study finds clear evidence that safety and comfort attributes hold significant value for users. Participants were willing to pay for features such as helmet availability, seating in shared modes, and improved pedestrian pathways. These findings confirm earlier research showing that users view safety and comfort as core elements of service quality, rather than mere amenities [15].

5.2. Policy Implications

(1)

Pricing Strategies Reflecting User Valuations and Equity

The higher WTP for motorcycle taxi travel time indicates that many users are willing to pay more for faster service when under time pressure. This supports the use of differentiated pricing strategies that reflect the value users place on speed and reliability.

Examples from platform-based transport services, including Gojek and Grab in Jakarta and GrabBike in Bangkok, demonstrate the operational feasibility of such approaches, where dynamic pricing is used to balance demand and supply during peak periods. However, implementing similar mechanisms in the Bangkok context requires careful consideration of the local regulatory and socio-institutional environment. Motorcycle taxis in Bangkok are primarily operated through loosely organized cooperatives with partial state oversight, and fare structures are often negotiated informally. Unlike app-based platforms, these services lack centralized mechanisms for real-time fare adjustments, making dynamic pricing highly dependent on local enforcement, driver discretion, and inter-agency coordination.

Moreover, the use of differentiated pricing raises concerns about affordability and equity, as highlighted in other transport-related studies (e.g., [16,17]. Without proper regulation, such systems may disproportionately burden low-income users, particularly in areas where walking or alternative modes are not viable due to safety, distance, or infrastructure constraints. Overreliance on WTP-based pricing frameworks can risk excluding price-sensitive populations from essential access, potentially exacerbating spatial and economic inequalities.

To mitigate these risks, any adoption of differentiated pricing should be accompanied by robust regulatory frameworks that ensure transparency and fairness. This may include fare caps, publicly subsidized access tiers (e.g., flat-rate fares for certain user groups or during peak periods), and clearly posted fare structures to reduce opportunistic pricing. In the Bangkok context, this would require coordinated efforts among the Department of Land Transport, Bangkok Metropolitan Administration, and motorcycle taxi associations to institutionalize equitable pricing mechanisms that reflect user valuations without marginalizing vulnerable populations.

(2)

Walk Enhancement for Inclusive Access

The WTP of THB 15.72 for improved walking conditions indicates that users place significant value on safe, shaded, and obstacle-free walking environments, aligned with previous studies [38,46]. This supports prior findings that quality pedestrian infrastructure strongly influences user behavior and satisfaction [24]. Such improvements are essential components of transit-oriented development and can reduce reliance on motorized access modes.

Given this high WTP, targeted investments in walkability—such as resurfacing footpaths, providing shade, and removing barriers—can be economically justified. These investments also deliver wider social and environmental benefits, including improved inclusivity for low-income users and reduced emissions from short motorized trips, as emphasized in studies on sustainable urban transport [40].

(3)

Strengthening Safety Through Helmet Provision Schemes

The estimated WTP of approximately THB 8 for helmet availability affirms that safety is a valued component of motorcycle taxi (MCT) services. This finding supports earlier studies such as [15], which highlight user concerns related to hygiene, risk perception, and helmet provision in informal or semi-formal transport systems. A structured helmet scheme could mandate that all MCT drivers carry a passenger helmet, with compliance monitored through inspections, licensing requirements, or operator certification programs.

Despite existing legal mandates requiring helmet use in Thailand, actual compliance remains incomplete. As reported by [47], passenger helmet use has not reached universal levels, revealing a persistent enforcement and behavioral gap. This underscores the need for targeted interventions that address both supply-side factors (helmet availability) and demand-side factors (user motivation and awareness).

Given the demonstrated willingness to pay, policymakers could consider implementing a modest safety surcharge or integrating helmet provision into the standard fare structure. In lower-income areas or among small-scale operators, public–private co-financing models or municipal support schemes [25] could help ensure helmet availability without increasing users’ financial burden. These strategies present a feasible and user-supported pathway to enhance passenger safety in informal urban transport.

(4)

Toward an Integrated Policy Framework for Informal Access Modes

Beyond individual interventions, a coordinated policy framework is needed to systematically improve station access while addressing behavioral heterogeneity and equity concerns. Achieving long-term improvements requires institutional alignment and robust enforcement mechanisms. For informal and semi-regulated modes such as motorcycle taxis and local minibuses, it is essential to establish enforceable quality standards, including mandatory safety equipment, basic comfort provisions, and timetable adherence. These standards should be reinforced through licensing requirements, routine inspections, and rider feedback mechanisms.

Equity considerations must be embedded throughout policy design, particularly for low-income and mobility-constrained users. Potential measures include subsidized fares, co-financing of infrastructure or equipment in underserved areas, and gender-sensitive initiatives, such as women-prioritized boarding zones and targeted safety programs.

5.3. Limitations and Future Recommendations

(1)

Model Structure, Experimental Design, and Estimation

Several aspects of the model and experimental design warrant further consideration. The mixed logit model was estimated using a panel data structure, which appropriately incorporates intra-respondent correlation arising from repeated choice tasks. However, this structure assumes stable preferences across tasks, and does not account for potential learning, fatigue, or simplification strategies adopted by respondents over the course of the survey. More broadly, stated preference (SP) data are inherently hypothetical and may not fully reflect real-world behavior, particularly in contexts involving safety perceptions or informal services. Although the choice scenarios were carefully designed and pretested to enhance realism, hypothetical bias remains a potential limitation.

Additionally, the design was constrained to three access modes and a limited set of attributes to reduce cognitive burden. While this helped ensure clarity and response reliability, it may have excluded factors—such as waiting time, insurance coverage, or environmental discomforts—that could influence decision-making in real settings. These simplifications, while necessary for practical survey administration, may affect the external validity of the findings. Furthermore, the specification of lognormal distributions for travel time parameters—used to ensure strictly positive WTP—may introduce skewness that inflates upper-tail heterogeneity. As with all discrete choice models, the utility specification and distributional assumptions involve trade-offs and potential misspecification risks. Future research could explore alternative distributions, extend the design to cover a broader range of attributes, or apply non-parametric approaches to test the robustness of preference estimates and improve behavioral realism.

(2)

Capturing Behavioral Heterogeneity

The mixed logit model employed in this study captures random taste variation, reflecting diverse behavioral preferences for station access modes. The results reveal substantial heterogeneity in willingness to pay (WTP) for travel time, particularly for walking. This suggests that individual preferences are influenced by factors such as physical ability, trip purpose, and trip-related burden. For example, walking may be more demanding for individuals traveling with children or carrying heavy items.

It is also important to note that the sample primarily comprises regular metro users, many of whom report stable income and formal employment. As a result, the findings may under-represent the access preferences and WTP patterns of lower-income or informal workers, who are not currently using the metro, but often rely more heavily on low-cost or informal access modes.

While the mixed logit framework effectively captures continuous heterogeneity, it does not segment the population into clearly interpretable groups. Consequently, systematic differences among specific traveler types may be masked. Future research could apply latent class choice models (LCCMs) to identify distinct user segments based on shared characteristics and access behaviors [48,49]. These models can uncover meaningful groupings—such as low-income users, frequent commuters, or older adults—providing deeper insight into how each segment values safety, comfort, and travel time. Such segmentation would support the design of more targeted and equitable station access policies.

(3)

Integrating Latent Perceptions Using Hybrid Choice Modeling

The current model incorporates alternative-specific constants (ASCs) and gender term to reflect mode preferences. However, these variables serve only as statistical adjustments, and do not capture the underlying attitudinal or perceptual factors that influence access mode choices. For example, the observed preference among female respondents for motorcycle taxis and Songthaews may be influenced by safety perceptions, cultural norms, or previous experiences—factors that are not explicitly modeled here.

Attempts to include multiple perception indicators directly in the utility function can result in multicollinearity, especially when those indicators are highly correlated. This can bias parameter estimates and reduce model reliability. To overcome this, future research should consider structuring perceptions as latent variables. A combined use of Exploratory Factor Analysis (EFA) and Confirmatory Factor Analysis (CFA) can identify stable latent constructs, while a Multiple Indicators Multiple Causes (MIMIC) model can link these constructs to socio-demographic characteristics. These latent variables can then be integrated into a hybrid choice model (HCM) framework [50,51], allowing researchers to assess how perceptions influence access mode utility [46,52].

For future research, the application of hybrid choice modeling (HCM) approaches is recommended to jointly estimate choice behavior and the latent psychological constructs that influence it. This framework addresses the issue of multicollinearity that arises when multiple perception indicators are included directly in the utility function. By incorporating latent variables, HCM can provide more nuanced insights into the factors shaping access mode preferences and support the development of more targeted, perception-informed interventions—particularly for user groups whose choices are influenced by unobserved attitudes and social norms.

6. Conclusions

This study addresses key research gaps in the valuation of informal access transport by applying a WTP-space mixed logit model to examine how Bangkok metro users value travel time, safety, and comfort attributes in first-/last-mile travel. While prior studies in Bangkok and Southeast Asia have largely focused on main-line metro segments or employed preference-space models with limited policy interpretability, this research contributes new insights by applying a WTP-space framework that produces direct monetary valuations and accounts for unobserved heterogeneity in user preferences. It represents one of the first applications of this modeling approach to station access segments involving semi-formal and informal transport modes in the Southeast Asian context, highlighting the behavioral and policy relevance of non-market attributes.

The analysis reveals several key findings: the value of time (VOT) for motorcycle taxi access is nearly four times higher than the official metro VOT, reflecting high urgency and sensitivity in short-distance access segments. Walking, in contrast, shows lower WTP with substantial variation, suggesting limited tolerance due to environmental discomfort and perceived inconvenience. Users place strong value on safety and comfort, with helmet availability valued at approximately 40% of a typical motorcycle taxi fare and seating availability on shared minibuses also highly valued, despite the short travel distances. Gender differences are evident, with female users showing stronger preferences for motorized options—driven by safety concerns and vulnerability in pedestrian environments. A temporal asymmetry in preferences is also observed: walking is more acceptable when time pressure is low, while motorized modes are favored under urgency.

Beyond the Bangkok context, this study offers regionally relevant insights that extend the previous Southeast Asian literature. It provides the first WTP-based valuation of informal access modes in the region, contributing empirical evidence that better reflects the realities of user behavior in cities with widespread informal transport. Compared to earlier studies in low- and middle-income countries, it delivers more disaggregated, access-segment-specific estimates that are sensitive to local infrastructure, service characteristics, and social norms. These insights help bridge the gap between formal transport appraisal frameworks and the informal access conditions prevalent across Global South cities.

The findings carry several policy implications. Relying on uniform value-of-time assumptions may underestimate the benefits of improving access; segment-specific valuations are needed for more accurate investment appraisal. The results support the consideration of differentiated pricing strategies, though care must be taken to address equity risks—particularly for low-income and vulnerable users. Investments in walkability enhancements and helmet provision schemes are behaviorally justified and supported by users’ willingness to pay. Establishing service standards for semi-formal modes—such as basic safety and comfort regulations—alongside equity-promoting measures (e.g., subsidies, gender-sensitive access facilities) can substantially improve overall access quality.

To extend this work, the application of hybrid choice models is recommended to incorporate latent perceptions—such as safety and comfort—into utility estimation. Such approaches would allow for a deeper exploration of the psychological and social factors influencing access decisions. Collectively, these contributions represent a meaningful application of stated preference modeling to informal metro access, offering practical insights for designing more inclusive, equitable, and sustainable urban transit systems aligned with Sustainable Development Goal (SDG) 11.2.