Investigating Travel Mode Choices Under Environmental Stress: Evidence from Air Pollution Events in Chiang Rai, Thailand

Abstract

Air pollution poses growing challenges to public health and urban mobility in Southeast Asia. This study investigates how air quality crises affect travel mode choices in Chiang Rai, Thailand, a secondary city experiencing seasonal PM2.5 smog episodes. A structured online survey was conducted with 406 respondents, collecting paired data on travel behaviors during non-air quality crisis (N-AQC) and air quality crisis (AQC) periods. Using a multinomial logit model (MNL), key socioeconomic and trip-related variables were analyzed to estimate mode choice probabilities. The results reveal significant behavioral shifts during an air quality crisis, with private car usage increasing from 30.30% to 34.70% and motorcycle usage decreasing from 50.20% to 42.90%. Multinomial logit models attained correct classification rates of 67.5% and 63.8%, with pseudo R² values exceeding 0.50 for both periods. These findings highlight how environmental stress alters travel behavior, especially among younger and low-income populations. The study contributes new insights from a Southeast Asian urban context, emphasizing the need for adaptive transport policies, protective infrastructure, and equity-focused interventions to promote sustainable mobility during an environmental crisis.

1. Introduction

Air pollution has emerged as one of the most pressing global environmental challenges of the 21st century, with significant implications for public health and urban mobility patterns. This study examines how air quality crises affect travel mode choices in Chiang Rai, Thailand, with particular focus on the shifts between private vehicles, motorcycles, public transport, and alternative modes, including walking, bicycle, e-hailing service, and taxi. The primary objectives are to (1) quantify changes in travel mode distributions between non-air quality crisis and air quality crisis, (2) identify key socioeconomic and environmental factors influencing mode choice during these periods, and (3) develop evidence-based recommendations for transportation policy adaptation during an air quality crisis. This study makes novel contributions by providing empirical evidence of travel behavior adaptation in medium-sized Southeast Asian cities, introducing a temporal comparative framework for examining mode choice during non-air quality crisis (N-AQC) versus air quality crisis (AQC), and establishing a methodological approach for analyzing environmental impacts on transportation choices in developing urban contexts.

Transportation mode choice behaviors have been significantly influenced by environmental conditions across various Asian contexts [1]. Studies have shown that air quality concerns can substantially alter travel patterns and modal preferences, particularly in developing countries where rapid economic growth often outpaces environmental regulations [2]. Among these pollutants, particulate matter (PM) has garnered significant attention, with PM2.5 and PM10 being of particular concern [3]. According to the World Health Organization, air pollution exposure significantly impacts public health globally, highlighting the critical need to understand how air quality affects transportation choices [4].

In the context of Southeast Asia, rapid urbanization and economic development have led to significant air quality challenges. The region’s unique geographical and meteorological conditions, combined with anthropogenic activities such as biomass burning and increasing vehicle emissions, contribute to recurring episodes of severe air pollution, often referred to as “haze” or “smog” crisis [5]. These events not only pose immediate health risks but also disrupt daily activities and potentially influence long-term behavioral patterns, including travel mode choices. Thailand, as one of the more economically developed countries in Southeast Asia, provides an interesting case study for examining the interplay between air pollution and travel behavior. While much attention has been focused on the capital city of Bangkok, smaller cities and provinces in Thailand also face significant air quality issues, often with fewer resources to address them [6].

Chiang Rai province, located in northern Thailand, exemplifies this challenge, experiencing recurring air quality crises primarily caused by a combination of agricultural burning, industrial emissions, and unfavorable weather patterns [7]. According to the Pollution Control Department of Thailand, air pollution in Chiang Rai is characterized by elevated levels of PM10 and PM2.5, with data from 2012 to 2022 indicating that March is typically the month most affected by air pollution [8]. During these periods, both PM10 and PM2.5 levels frequently exceed the recommended standards set by Thai national guidelines. For the purposes of this study, we define two distinct periods: (1) non-air quality crisis, when air quality is within acceptable limits (PM2.5 below 50 μg/m³ and PM10 below 120 μg/m³), and (2) air quality crisis, when air quality significantly deteriorates (PM2.5 exceeding 50 μg/m³ and/or PM10 exceeding 120 μg/m³) for prolonged periods. There are a few studies about air pollution impacting travel commuting in Chiang Rai city. For example, studies have investigated factors affecting travel decisions during an air pollution crisis in Chiang Rai [9]. The transportation landscape in Chiang Rai offers a diverse range of options, reflecting both traditional and evolving mobility patterns in medium-sized Southeast Asian cities. Figure 1 shows that available transportation modes in Chiang Rai can be classified into four main categories: (1) private car, (2) motorcycle, (3) public transport (including bus, mini-bus (Song-Teaw), and van with services operating both intra-city and inter-city), and (4) alternatives (including walking, bicycle, e-hailing service, and taxi).

The aggregation of these alternatives into a single category was necessitated by their relatively low individual modal shares in the study area. The quantitative threshold for aggregation was established based on the minimum sample size requirements for multinomial logit modeling, which typically requires at least 30–50 observations per category to ensure parameter stability [10,11]. With only 8–26 individual responses per alternative mode, separate analysis would have resulted in unreliable coefficient estimates and wide confidence intervals, potentially leading to misleading conclusions about environmental crisis impacts. This consolidation approach aligns with established practices in transportation studies of medium-sized cities where emerging mobility options represent smaller modal splits [12,13]. Likewise, previous studies have examined the influence of air pollution on travel behavior in various global contexts [14]; however, there remains a notable gap in understanding these dynamics within Southeast Asian settings, particularly in smaller cities like Chiang Rai.

This study employs a multinomial logit model to analyze mode choice patterns across non-air quality crisis and air quality crisis periods [15]. Through this approach, the probability of specific travel mode selection is estimated based on various predictor variables, including socioeconomic factors, trip characteristics, and air quality conditions. The investigation is structured across two distinct temporal contexts, non-air quality crisis (N-AQC) and air quality crisis (AQC) periods, through which behavioral adaptations in response to air quality deterioration can be analyzed comparatively. The findings from this research are expected to provide valuable insights for developing targeted transportation and environmental policies, particularly for medium-sized cities facing recurring air quality challenges. By bridging the gap between global air quality concerns and local travel behaviors, this study offers important implications for policymakers, urban planners, and researchers working towards more sustainable and health-conscious urban environments in similar contexts across Southeast Asia and beyond.

This paper is organized as follows: Section 2 reviews the literature on the impact of air pollution on travel behavior and the methodological approaches employed in mode choice analysis. Section 3 presents the research methodology, including the study area, data collection, and analytical methods. Section 4 reports the results, focusing on travel behavior patterns and mode choice models during both non-air quality crisis and air quality crisis periods. Section 5 discusses the key findings, policy implications, and limitations of the study. Finally, Section 6 concludes the paper.

2. Literature Review

2.1. Air Pollution and PM2.5 Smog Impact on Travel Behavior

Fine particulate matter (PM), which consists of airborne particles with aerodynamic diameters smaller than 2.5 μm (PM2.5), has been identified as a major air pollutant affecting human health and travel behavior. PM2.5 is particularly concerning due to its ability to penetrate deep into the respiratory system and enter the bloodstream, causing various health issues [4]. Studies consistently show that deteriorating air quality, especially elevated PM2.5 levels, influences travelers to shift between different transportation modes. In Delhi, India, when PM2.5 concentrations exceed 150 μg/m³ (six times higher than the WHO’s recommended 24 h guideline of 25 μg/m³), commuters increasingly prefer closed modes of transportation over open modes. This was demonstrated through machine learning models and logit analysis comparing enclosed vehicles (cars, air-conditioned buses) versus open vehicles (auto rickshaws, non-air-conditioned buses) [16]. The impact of air pollution on mode choice varies across different urban contexts. In Taiyuan, China, high PM2.5 concentrations negatively impacted the selection of non-motorized transport modes (walking, cycling) in favor of motorized vehicles (cars, buses) [17]. Similarly, in Zhengzhou, China, a study examining car use, public transit (buses and subway), and active modes (walking and cycling) found that car commuters tend to maintain their preference for private vehicles even after receiving health information about PM2.5 exposure [18].

Income levels play a significant role in mode choice during pollution events. In Seoul, South Korea, public transit usage increases among lower-income groups during high pollution periods, particularly for bus and subway services, as revealed through an Integrated Choice and Latent Variable (ICLV) model [19]. This finding was further supported by multilevel logistic regression modeling comparing non-motorized modes (walking or biking), public transit (bus or subway), and cars [20]. In Karaj, Iran, when PM2.5 concentrations exceed 75 μg/m³, poor air quality increases private car usage compared to walking and public transit options, as shown through exploratory factor analysis and hybrid choice modeling [21]. In the United States, a comprehensive study across 929 urban areas examined how emissions affect choices between driving alone, carpooling, public transportation, walking, and other modes, finding that vehicle ownership significantly influences transportation mode choices [22]. Recent research in Delhi has employed sophisticated analytical methods to study a wide range of transportation modes, including auto rickshaws, buses, cars, two-wheelers, two-wheeler sharing, walking, bicycles, car sharing, and metro services. These studies consistently show that as PM2.5 levels rise above national and WHO standards, travelers modify their behavior to minimize exposure, with a clear preference for enclosed modes of transportation [23]. In Beijing, daily average PM2.5 concentrations significantly impact choices between cycling, cars, taxis, buses, metros, and walking, with notable shifts observed when levels exceed 150 μg/m³ [24]. The impact of PM2.5 pollution extends beyond mode choice to affect spatial travel patterns, as revealed by a big data field study in Xi’an, China [25]; these studies are shown in

Table 1. Summary of previous studies on mode choice during air pollution events.

Reference	Pollution Type	Location	Method Used	Mode of Transport	Finding
Li and Kamargianni [17]	Air pollution	Taiyuan, China	Mode choice models	Motorized and Non-motorized vehicles	Air pollution negatively impacts non-motorized transport mode choice.
Zhao et al. [24]	Air pollution, PM2.5	Beijing, China	Binary logit model	Cycle, cars, taxis, buses, metros, and walking	Air quality significantly influences travel mode choices.
Luo et al. [18]	Air pollution, PM2.5	Zhengzhou, China	Multinomial logit model (MNL) and difference-in-difference (DID) regression methods	Car, public transit, and active modes	Car commuters rebound towards car travel after health information.
Xu et al. [25]	Ambient air pollution	Xi’an, China	Regression model	N/A	People reduce travel area more than travel distance on polluted days.
Ercan et al. [22]	Emissions (CO, CO2, NOX, SOX, PM10, PM2.5, and VOCs)	929 urban areas in the U.S.	Multinomial logit model (MNL) and system dynamics (SD) modeling	Drive alone, carpool, public transportation, walk, and other	Vehicle ownership significantly impacts transportation mode choices.
Kim et al. [20]	Air pollution	Seoul, South Korea	Multilevel logistic regression modeling	Non-motorized modes (walking or biking), public transit (bus or subway), and cars	Lower-income groups shift to public transit during poor air quality.
Meena et al. [16]	Air pollution	Delhi, India	Machine learning models and logit model	Open and closed travel modes	Commuters prefer closed modes as air quality worsens.
Dabirinejad et al. [21]	Air pollution	Karaj, Iran	Exploratory factor analysis (EFA) and hybrid choice modeling (HCM)	Walking, car, and public transit	Poor air quality increases private car usage.
Kim et al. [19]	Particulate matter (PM)	Seoul, South Korea	Integrated Choice and Latent Variable (ICLV) model	Public transit	Public transit usage increases among lower-income groups during pollution events.
Meena et al. [23]	Air pollution	Delhi, India	Random Forest, XGBoost, Naive Bayes (NB), K-Nearest Neighbor, Support Vector Machine (SVM), and Multinomial logit model (MNL)	Auto rickshaw, bus, car, two-wheeler, two-wheeler sharing, walk, bicycle, car sharing, and metro	Commuters shift to closed modes during poor air quality.
Present study	Air pollution, PM2.5	Chiang Rai, Thailand	Descriptive statistic and Multinomial logit model (MNL),	Private car, motorcycle, public transport, and alternatives	Travel mode choice during non-air quality crisis and air quality crisis in urban area

.

While these studies have provided valuable insights into travel behavior during air pollution events, there remains a gap in understanding how the PM2.5 smog crisis specifically affects travel mode choices in Southeast Asian secondary cities, particularly in areas prone to seasonal smog events. The present study addresses this gap by examining travel mode choice behavior in Chiang Rai, Thailand, where severe seasonal air quality crises frequently occur. Using a combination of multinomial logit modeling (MNL) and descriptive statistics, this research investigates how commuters adapt their travel mode choices during non-air quality crisis and air quality crisis in an urban area where PM2.5 levels often exceed 150 μg/m³ during the dry season. Unlike previous studies that primarily focused on major metropolitan areas, this research provides insights into travel behavior adaptations in a secondary city context, where transportation options and infrastructure may differ significantly from larger urban centers. The study’s findings are particularly relevant for urban areas in the Greater Mekong Subregion that face similar seasonal air quality challenges, contributing to a more comprehensive understanding of how environmental crises affect transportation choices in developing regions.

2.2. Commuting Preference Analysis Methods

The analysis of commuting preferences and travel mode choices has employed various methodological approaches in transportation research. These methods aim to understand the factors influencing individuals’ travel decisions and predict future travel behaviors. Transportation research has employed various methodological approaches to analyze commuting preferences and travel mode choices. Discrete choice models, particularly multinomial logit (MNL) and nested logit models, have been widely used to understand travel decisions [26]. The MNL model assumes that travelers choose the option that maximizes their utility, with the probability of choosing a particular mode expressed as a function of its attributes and the individual’s characteristics. These models are often supported by data collected through stated preference (SP) and revealed preference (RP) surveys. SP surveys present hypothetical scenarios to understand potential behavioral responses, while RP surveys collect data on actual travel behaviors, emphasizing the value of combining both approaches for robust analysis [27]. Advanced econometric techniques have also been applied, such as the multiple discrete-continuous extreme value (MDCEV) model for analyzing activity–travel behavior [28]. This model extends traditional discrete choice frameworks by simultaneously considering multiple alternatives and their usage intensities. Machine learning techniques have gained prominence in recent years, with methods such as random forests and support vector machines often achieving higher predictive accuracy than traditional logit models [29]. These advanced techniques can capture complex non-linear relationships and interactions between variables that might be missed by conventional approaches. Structural equation modeling (SEM) has also been employed to examine the intricate relationships between various factors influencing travel behavior, with SEM being used to analyze how residential location choice, travel attitudes, and actual travel behavior interact [30]. Logistic regression models have been used to examine the impact of air pollution on cycling behavior in Beijing [24]. Time series analysis has been applied to investigate the relationship between extreme haze events and public transit ridership [31].

While substantial research has been conducted on the relationship between air pollution and travel behavior, there remains a gap in understanding these dynamics in the specific context of Chiang Rai, Thailand. The unique characteristics of the region, including its seasonal air quality crisis, socioeconomic factors, and existing transportation infrastructure, warrant a focused investigation. Additionally, most studies have examined general air pollution levels rather than acute air quality crisis events, which may elicit different behavioral responses. This study aims to address these gaps by providing insights into travel mode choice preferences during an air quality crisis in Chiang Rai, contributing to the broader understanding of how environmental factors influence urban mobility in Southeast Asian contexts.

3. Methodology

3.1. Study Area

This study was conducted in Chiang Rai province, the northernmost province of Thailand. Covering an area of 11,678 km², Chiang Rai is characterized by its diverse topography, including mountains, hills, and lowland plains. As of 2024, Chiang Rai had a population of 1,298,977 [32]. The climate of Chiang Rai is tropical savanna, characterized by distinct wet and dry seasons. The dry season, from November to April, coincides with the period of severe air pollution events. The practice of crop residue burning, particularly prevalent during the dry season, significantly contributes to the region’s air pollution problems [7]. During these air quality crises, PM2.5 levels often exceed 150 μg/m³, significantly higher than the World Health Organization’s guideline of 25 μg/m³ for 24 h mean [4]. In 2023, Chiang Rai experienced 76 days where PM2.5 levels exceeded the Thai national standard of 50 μg/m³ [33].

3.2. Data Collection

This study employed a probability sampling approach to ensure representative data collection from Chiang Rai province’s population. The sample size was calculated using Taro Yamane’s Formula as Equation (1) [34], considering Chiang Rai’s population of 1,298,977, a 95% confidence level, and a 5% margin of error, yielding a minimum required sample size of 400 respondents.

$$n={\displaystyle \frac{N}{1+N{e}^2}}$$

(1)

Initially, stratified random sampling was planned to ensure proportional demographic representation. However, practical limitations during implementation, including restricted access to comprehensive population lists, necessitated adopting simple random sampling while maintaining the probability framework. The survey was administered through an online questionnaire using Google Forms, optimized for mobile devices and available in Thai with English translation. Average completion time was 15–20 min. Prior to deployment, the questionnaire underwent pilot testing with 20 respondents and cognitive interviews using think-aloud protocols to ensure clarity and effectiveness. Minor adjustments were made based on feedback before online deployment from February to March 2024. Multiple distribution channels maintained random sampling integrity, including social network service (SNS), community networks through district offices, university networks, and community leaders in rural areas. QR codes were strategically posted in public spaces for easy access. To address online survey coverage bias, supplementary data collection methods were employed for respondents lacking internet access, including telephone interviews and paper-based questionnaires administered by trained enumerators. A total of 428 responses were received, with 406 deemed valid after data cleaning (response rate: 95%). Invalid responses were removed due to incomplete information, duplicate submissions, out-of-province residents, or inconsistent response patterns. This implementation enhances sample representativeness while acknowledging demographic limitations inherent in online survey methodology.

3.3. Survey Instruments

The questionnaire was designed with four sections: (1) demographic information, (2) impact of the air quality crisis, (3) travel behavior under non-air quality crisis, and (4) travel behavior during the air quality crisis. Demographic information included gender, age, occupation, income, marital status, vehicle ownership, and driving license status. Questions about experiences during air quality crisis and perceived main causes were included. Travel behavior sections covered mode choice, travel time, cost, frequency, and distance for both non-air quality crisis and air quality crisis conditions. This approach of comparing travel behavior across different environmental conditions aligns with previous studies examining the impact of air pollution on mode choice [24]. The questionnaire utilized a multiple-choice approach to capture both quantitative data. This approach allows for a nuanced understanding of travel behavior changes, similar to methods used in recent studies examining environmental impacts on transportation choices [35].

3.4. Analysis Method

The multinomial logit model (MNL) approach has been successfully applied in various Southeast Asian contexts to identify statistically significant parameters that determine choice probabilities among discrete alternatives [36,37]. A comprehensive analytical approach was employed to examine travel mode choices during non-air quality crisis (N-AQC) and air quality crisis (AQC) periods in Chiang Rai, based on random utility theory. The dependent variable was travel mode choice, with data collected from identical respondents across both periods, creating paired samples for direct comparison of individual behavioral changes. Prior to analysis, the dataset underwent preparation, including categorical variable coding using effect coding, continuous variable standardization, and missing data handling using multiple imputation with chained equations [38]. The utility function for individual mode choice was specified as Equation (2) [39].

$$U_i=V_i+{\epsilon }_i$$

(2)

Here, $U_i$ represents the utility of mode $i$, $V_i$ is the deterministic component of utility for mode $i$ often modeled as Equation (3), and ${\epsilon }_i$ is the random error term assumed to be independently capturing unobservable factors influencing the choice.

$$V_i={\beta }_0+{\beta }_1{x}_{1i}+{\beta }_2{x}_{2i}+\cdots +{\beta }_k{x}_{ki}$$

(3)

This modeling method was employed to model travel mode choices, a method widely used in transportation research for analyzing discrete choice scenarios [26]. This technique allows for estimating the probability of selecting a specific travel mode from multiple alternatives, considering various predictor variables. Information on travel costs, time, and other characteristics was collected for the respondent’s actual chosen mode rather than for all potential alternatives. This approach provides revealed preference data about the selected mode but does not include attributes of non-chosen alternatives. Consequently, the multinomial logit model estimated in this study captured the influence of sociodemographic factors and trip characteristics on mode choice.

The probability estimation model, as expressed in Equation (4), where $P\left(Y=i\right)$ represents the probability of choosing travel mode $i$ and $V_i$ is the systematic utility of that mode. The numerator, $\mathrm{e}\mathrm{x}\mathrm{p}\left(V_i\right)$, converts the utility into a positive number, making it comparable across options. The denominator, ${\sum }_{j=1}^J\mathrm{e}\mathrm{x}\mathrm{p}\left(V_j\right)$, sums the exponentiated utilities of all available travel modes to ensure that the total probability across all choices adds up to 1. This model specification allows for the examination of how factors such as travel time, cost, frequency, and distance, as well as demographic characteristics, influence mode choice during non-air quality crisis and air quality crisis. Similar approaches have been used in studies examining the impact of environmental factors on travel behavior [40].

$$P\left(Y=i\right)={\displaystyle \frac{\mathrm{e}\mathrm{x}\mathrm{p}\left(V_i\right)}{{\sum }_{j=1}^J\mathrm{e}\mathrm{x}\mathrm{p}\left(V_j\right)}}$$

(4)

The model’s performance was assessed using Akaike’s Information Criteria (AIC) and Bayesian Information Criteria (BIC). These information criteria are valuable tools for comparing different model specifications and selecting the most appropriate model. In addition to AIC and BIC, McFadden, Cox and Snell, and Nagelkerke pseudo R-squared (R²) were calculated to assess model fit [41].

4. Results

4.1. Demographic Characteristics of Respondents

The respondent profile reveals critical socioeconomic and demographic patterns relevant to travel behavior under environmental stress. The sample was predominantly female (63.8%), with males comprising 31.3% and individuals identifying as other genders making up 4.9%. A significant proportion of respondents were young adults, with 85.9% under the age of 30, indicating a digitally engaged and mobile demographic often reached through online survey platforms. The largest age group, 21–30 years, accounted for 59.1% of the sample, likely influenced by outreach through universities and social media channels. Income levels were skewed toward the lower end of the spectrum, with over half (53.9%) earning less than 10,000 THB per month and only a small minority (2.7%) earning more than 50,000 THB. This economic distribution highlights potential vulnerabilities to rising transport costs, particularly during an air quality crisis. Regarding marital status, 83.5% of respondents were unmarried, again underscoring a young, possibly student-heavy population. Vehicle ownership was relatively high at 82.3%, although the majority held motorcycle licenses (56.4%) rather than private car licenses (40.6%). This suggests a mobility pattern dominated by two-wheeled transport, a common feature in many Southeast Asian secondary cities where affordability and accessibility shape transport choices. These demographic trends provide essential context for interpreting the model estimates and behavioral responses reported later in the study. The socioeconomic composition of the sample, predominantly young, low-income, and motorcycle-reliant, directly informs how individuals respond to environmental hazards such as air pollution. A detailed breakdown of respondent characteristics is provided in

Table 2. Respondent characteristics.

Item	Value: Description	Count	Percent
Total of respondents		406	100
Gender	1: Male	127	31.3
	2: Female	259	63.8
	3: Others	20	4.9
Age (years)	1: <21	109	26.8
	2: 21–30	240	59.1
	3: 31–40	22	5.4
	4: 41–50	25	6.2
	5: 51–60	10	2.5
Monthly income (THB)	1: <10,000	219	53.9
	2: 10,001–15,000	38	9.4
	3: 15,000–20,000	82	20.2
	4: 20,001–30,000	37	9.1
	5: 30,001–40,000	11	2.7
	6: 40,001–50,000	8	2.0
	7: >50,001	11	2.7
Marital status	1: Unmarried	339	83.5
	2: Married	37	9.1
	3: Not mentioned	30	7.4
Vehicle ownership	0: No	72	17.7
	1: Yes	334	82.3
Holding motorcycle driving license	0: No	177	43.6
	1: Yes	229	56.4
Holding private car driving license	0: No	241	59.4
	1: Yes	165	40.6

Note: 32.435 THB = 1 USD (Bank of Thailand, 27 September 2024).

.

4.2. Analysis of Travel Distributions

Travel behavior patterns were analyzed across five key dimensions during non-air quality crisis (N-AQC) and air quality crisis (AQC) periods: travel time, travel cost, travel frequency, travel distance, and travel mode distributions. The analysis revealed significant shifts in travel patterns between these periods, reflecting adaptations in response to air quality conditions. Each distribution was examined to identify changes in modal preferences and travel characteristics that emerged during an air quality crisis.

4.2.1. Travel Time Distributions

Based on the travel time distribution data in Figure 2, the most significant change occurs in motorcycle usage, with an 8% increase for trips lasting 11–20 min during an air quality crisis. This suggests a shift towards shorter motorcycle trips when air quality deteriorates. Alternatives also see a marked increase, with a 5% rise in the same 11–20 min range, indicating a greater preference for using alternatives for short trips during an air quality crisis. Private car use shows a modest 2% decrease for trips in the 21–30 min range, while public transport experiences a slight 1 percentage point increase for 11–20 min journeys. These patterns collectively suggest that during an air quality crisis, residents of Chiang Rai tend to opt for shorter trips, with a particular preference for motorcycles and alternatives for shorter distances.

4.2.2. Travel Cost Distribution

The travel cost distributions reveal several notable shifts in commuter behavior. As demonstrated in Figure 3, the most significant change is observed in motorcycle usage, where trips costing less than 50 THB decreased 5%, indicating a drop in all motorcycle trips. This substantial reduction in low-cost motorcycle journeys suggests a shift away from this mode for shorter trips during an air quality crisis. An increase of 2–3% in private car use in the 101–150 THB range during an air quality crisis indicates a potential shift towards longer or different car trips. Public transport usage decreased marginally (1–2%) across all cost categories, reflecting a general reduction in public transit use. Interestingly, alternative transport experienced a small increase of 2–3% in the 101–150 THB range, suggesting some commuters may be opting for alternatives during an air quality crisis. Overall, these changes point to a general trend of slightly higher travel costs across most modes during an air quality crisis, with the most pronounced shift being away from motorcycle trips.

4.2.3. Travel Frequency Changes

The shifts in travel frequency behavior are shown in Figure 4, particularly in lower-frequency travel categories. Private car use in the 2–3 trips per week category increases 5%, while motorcycle use in the same category shows an even greater increase, by 8 percent. Public transport experiences a modest rise of 2% for 2–3 trips per week. However, the most striking change is observed in alternative transportation. The percentage of people using alternatives for 1 trip per week jumps from near 0% to 5%, and for 2–3 trips per week it increases from 1% to 4%. These trends indicate a general reduction in travel frequency during an air quality crisis, especially for private vehicles, coupled with a substantial increase in the use of alternatives. This shift suggests that residents are adapting their travel behaviors in response to air quality crisis, likely prioritizing essential trips and perceived safer travel options during periods of high air pollution.

4.2.4. Travel Distance Changes

The statistic of travel distance reveals several notable shifts in behavior. As shown in Figure 5, the most significant change is observed in motorcycle usage for longer trips (>5 km), which decreases by approximately 11% during air quality crisis. Conversely, alternatives for shorter trips (<5 km) see a substantial increase of around 6%. Motorcycle use for shorter trips shows a modest increase of about 3% points, while private car use for short distances also rises by roughly 4%. These shifts indicate a clear tendency towards reduced exposure during longer trips and increased use of alternatives for shorter distances when air quality is poor. The data suggest that residents adapt their travel behaviors in response to air quality crisis, prioritizing shorter trips and exploring alternatives to minimize their exposure to pollutants.

4.2.5. Travel Mode Distributions

Analysis of travel mode selection in Chiang Rai, Thailand, as shown in

Table 3. Travel mode distributions.

Travel Mode	Percentage
	N-AQC	AQC
Private car	30.30	34.70
Motorcycle	50.20	42.90
Public transport	13.10	8.60
Alternatives	6.40	13.80

Note: N-AQC is non-air quality crisis; AQC is air quality crisis.

, reveals significant shifts during an air quality crisis compared to a non-air quality crisis. Private car usage increased by 4.40% points to 34.70%, indicating a preference for enclosed transportation during an air quality crisis. Motorcycle usage saw the largest decrease, dropping 7.30% points to 42.90%, likely due to increased pollution exposure. Public transport usage declined by 4.50% points to 8.60%, possibly reflecting concerns about shared spaces. Notably, alternatives experienced the most substantial increase, rising 7.40% points to 13.80%, suggesting adaptive behavior among residents seeking flexible or less exposed travel options. The “Alternatives” category combines transportation modes with varying air pollution exposure profiles. In Chiang Rai, walking and biking constitute small portions of overall mode share, necessitating their combination with other alternative modes for adequate statistical analysis. Preliminary examination indicated taxi-based services showed proportionally larger increases during air quality crisis compared to active transportation, though sample constraints prevented robust conclusions. This suggests a preference for enclosed alternatives that minimize pollutant exposure, demonstrating that air quality significantly influences mode choices toward enclosed transportation during an air quality crisis.

4.3. Mode Choice for Non-Air Quality Crisis (N-AQC)

4.3.1. Likelihood Ratio Tests and Collinearity Matrix (N-AQC)

The likelihood ratio tests reveal that all examined factors significantly influence travel mode choice under a non-air quality crisis in Chiang Rai, Thailand (p < 0.05), as shown in

Table 4. Likelihood ratio tests of non-air quality crisis (N-AQC).

Effect	Chi-Square	df	Sig.
Intercept	15.011	3	0.002 **
Gender	13.772	3	0.003 **
Age	11.660	3	0.009 **
Monthly income	17.954	3	<0.000 ***
Marital status	8.407	3	0.038 *
Vehicle ownership	24.308	3	<0.000 ***
Holding a motorcycle driving license	31.509	3	<0.000 ***
Holding a private car driving license	36.885	3	<0.000 ***
Effect on healthcare	16.341	3	0.001 **
Effect on finance	9.509	3	0.023 *
Travel time	16.703	3	0.001 **
Travel cost	20.630	3	<0.000 ***

Note: * p < 0.05, ** p < 0.01, *** p < 0.001.

. The test results demonstrate high statistical significance (p < 0.001). The most influential factors include holding a private car driving license, holding a motorcycle driving license, vehicle ownership, travel cost, and income. Other significant factors encompass travel time, healthcare effects, gender, age, financial effects, and marital status. The high chi-square values and low p-values (all p ≤ 0.038) suggest these variables are Thistrong predictors of mode choice in a non-air quality crisis. The collinearity matrix presented in

Table 5. Collinearity matrix of independent variable for non-air quality crisis (N-AQC).

	${\mathit{X}}_{\mathit{G}}$	${\mathit{X}}_{\mathit{A}}$	${\mathit{X}}_{\mathit{M}\mathit{I}}$	${\mathit{X}}_{\mathit{M}\mathit{S}}$	${\mathit{X}}_{\mathit{V}\mathit{O}}$	${\mathit{X}}_{\mathit{M}\mathit{C}\mathit{D}\mathit{L}}$	${\mathit{X}}_{\mathit{P}\mathit{C}\mathit{D}\mathit{L}}$	${\mathit{X}}_{\mathit{E}\mathit{H}}$	${\mathit{X}}_{\mathit{E}\mathit{F}}$	${\mathit{X}}_{\mathit{T}\mathit{T}}$	${\mathit{X}}_{\mathit{T}\mathit{C}}$
$X_G$	-
$X_A$	−0.112	-
$X_{MI}$	−0.149	0.562	-
$X_{MS}$	−0.075	0.177	0.074	-
$X_{VO}$	−0.107	0.180	0.119	−0.076	-
$X_{MCDL}$	−0.043	0.072	0.048	−0.023	0.294	-
$X_{PCDL}$	−0.153	0.360	0.411	0.005	0.187	0.111	-
$X_{EH}$	0.109	0.066	0.040	−0.081	−0.046	0.028	0.053	-
$X_{EF}$	0.034	0.000	−0.027	−0.004	−0.084	0.031	−0.048	0.185	-
$X_{TT}$	−0.068	0.189	0.126	0.023	0.051	0.015	0.196	−0.009	−0.066	-
$X_{TC}$	−0.072	0.202	0.214	0.047	0.076	−0.058	0.185	−0.002	−0.073	0.462	-

Note: $X_G$: gender, $X_A$: Age, $X_{MI}$: monthly income, $X_{MS}$: marital status, $X_{VO}$: vehicle ownership, $X_{MCDL}$: holding a motorcycle driving license, $X_{PCDL}$: holding a private car driving license, $X_{EH}$: effect on healthcare, $X_{EF}$: effect on finance, $X_{TT}$: travel time, $X_{TC}$: travel cost.

confirms the statistical robustness of the model. Pearson correlation analysis demonstrates that all pairs of independent variables maintain correlation coefficients below 0.80, effectively ruling out significant multicollinearity concerns [42]. Among the observed relationships, three moderate correlations emerge: age and income (r = 0.562), travel time and cost (r = 0.462), and income and holding a private car driving license (r = 0.411). All other variable pairs exhibit weak correlations (r < 0.5), indicating their relative independence. This statistical independence of predictor variables establishes a sound methodological foundation, enabling reliable analysis of travel behavior patterns both under non-air quality crisis and during air quality crisis.

4.3.2. Multinomial Logit Model Parameter Estimates and Utility Function (N-AQC)

A multinomial logit model analysis was conducted to examine travel mode choice behavior under a non-air quality crisis, with private car designated as the reference category, as presented in

Table 6. MNL parameter estimates of non-air quality crisis model (N-AQC).

Mode	Variable	Coef.	Sig.	Odds Ratio
Motorcycle	Intercept	2.832	0.002 **
	Gender $(X_G$)	0.019	0.944	1.019
	Age $(X_A$)	0.058	0.780	1.059
	Monthly income $(X_{MI}$)	−0.454	***	0.635
	Marital status $(X_{MS}$)	−0.475	0.046 *	0.622
	Vehicle ownership $(X_{VO}$)	−0.331	0.474	0.718
	Holding motorcycle driving license $(X_{MCDL}$)	1.252	***	3.496
	Holding private car driving license $(X_{PCDL}$)	−1.776	***	0.169
	Effect on healthcare $(X_{EH}$)	0.867	0.047 *	2.379
	Effect on finance $(X_{EF}$)	0.180	0.588	1.197
	Travel time $(X_{TT}$)	−0.097	0.344	0.907
	Travel cost $(X_{TC}$)	−0.477	***	0.621
Public transport	Intercept	−0.375	0.746
	$\mathrm{Gender} (X_G$)	1.248	0.001 **	3.485
	$\mathrm{Age} (X_A$)	0.857	0.002 **	2.356
	$\mathrm{Monthly} \mathrm{income} (X_{MI}$)	−0.475	0.005 **	0.622
	$\mathrm{Marital} \mathrm{status} (X_{MS}$)	−0.681	0.048 *	0.506
	$\mathrm{Vehicle} \mathrm{ownership} (X_{VO}$)	−2.045	***	0.129
	$\mathrm{Holding} \mathrm{motorcycle} \mathrm{driving} \mathrm{license} (X_{MCDL}$)	−0.392	0.355	0.676
	$\mathrm{Holding} \mathrm{private} \mathrm{car} \mathrm{driving} \mathrm{license} (X_{PCDL}$)	−1.346	0.003 **	0.260
	$\mathrm{Effect} \mathrm{on} \mathrm{healthcare} (X_{EH}$)	−1.030	0.067	0.357
	$\mathrm{Effect} \mathrm{on} \mathrm{finance} (X_{EF}$)	1.200	0.006 **	3.321
	$\mathrm{Travel} \mathrm{time} (X_{TT}$)	0.388	0.003 **	1.474
	$\mathrm{Travel} \mathrm{cost} (X_{TC}$)	−0.427	0.025 *	0.652
Alternatives	Intercept	1.396	0.335
	$\mathrm{Gender} (X_G$)	0.431	0.381	1.539
	$\mathrm{Age} (X_A$)	0.452	0.206	1.572
	$\mathrm{Monthly} \mathrm{income} (X_{MI}$)	−0.236	0.210	0.790
	$\mathrm{Marital} \mathrm{status} (X_{MS}$)	−1.111	0.039 *	0.329
	$\mathrm{Vehicle} \mathrm{ownership} (X_{VO}$)	−1.797	0.003 **	0.166
	$\mathrm{Holding} \mathrm{motorcycle} \mathrm{driving} \mathrm{license} (X_{MCDL}$)	−0.457	0.395	0.633
	$\mathrm{Holding} \mathrm{private} \mathrm{car} \mathrm{driving} \mathrm{license} (X_{PCDL}$)	−2.045	0.001 **	0.129
	$\mathrm{Effect} \mathrm{on} \mathrm{healthcare} (X_{EH}$)	−1.056	0.093	0.348
	$\mathrm{Effect} \mathrm{on} \mathrm{finance} (X_{EF}$)	0.869	0.108	2.386
	$\mathrm{Travel} \mathrm{time} (X_{TT}$)	0.027	0.868	1.027
	$\mathrm{Travel} \mathrm{cost} (X_{TC}$)	0.141	0.481	1.152

Note: * p < 0.05, ** p < 0.01, *** p < 0.001; reference mode: private car.

. The model revealed several significant socioeconomic and travel-related factors influencing mode choice across different transportation options. The utility function for mode choice during non-air quality crisis (N-AQC) was specified as Equation (5).

$$\begin{gathered} U_{N-AQC}=\alpha X_G+\beta X_A+\gamma X_{MI}+\delta X_{MS}+\epsilon X_{VO}+\zeta X_{MCDL}+\eta X_{PCDL}+\theta X_{EH} \\ +\iota X_{EF}+\kappa X_{TT}+\lambda X_{TC}+c \end{gathered}$$

(5)

Here, the variables represent $X_G$: gender, $X_A$: age, $X_{MI}$: monthly income, $X_{MS}$: marital status, $X_{VO}$: vehicle ownership, $X_{MCDL}$: holding a motorcycle driving license, $X_{PCDL}$: holding a private car driving license, $X_{EH}$: effect on healthcare, $X_{EF}$: effect on finance, $X_{TT}$: travel time, and $X_{TC}$: travel cost.

The analysis of motorcycle usage revealed that income has a significant negative effect ($\alpha$ = −0.454), indicating that higher-income individuals are less likely to choose motorcycles over private cars. License possession showed contrasting effects—holding a motorcycle license significantly increased the likelihood of motorcycle use ($\beta$ = 1.252), while possession of a car license decreased it ($\gamma$ = −1.776). Travel cost demonstrated a negative relationship with motorcycle choice ($\delta$ = −0.477), suggesting cost sensitivity among motorcycle users. For public transport, gender and age emerged as significant positive factors, with females ($\epsilon$ = 1.248) and older individuals ($\zeta$ = 0.857) showing higher probabilities of choosing this mode. Income exhibited a negative relationship with public transport use ($\alpha$ = −0.475), while travel time showed a positive effect ($\theta$ = 0.388), possibly reflecting the reliability of scheduled services for longer journeys. Vehicle ownership was found to significantly decrease public transport use ($\eta$ = −2.045). Regarding alternative modes, marital status showed a significant negative effect ($\iota$ = −1.111). Both vehicle ownership ($\eta$ = −1.797) and possession of a car license ($\gamma$ = −2.045) were found to substantially decrease the likelihood of choosing alternative transportation modes. The findings align with previous studies on mode choice behavior in Southeast Asian contexts. This comprehensive analysis provides valuable insights into the factors influencing travel mode choices under a non-air quality crisis, establishing a baseline for comparison with behavior during an air quality crisis. The model’s results highlight the complex interplay between socioeconomic characteristics, travel attributes, and mode choice decisions.

4.3.3. Model Fitting and Predictive Accuracy of Non-Air Quality Crisis Model (N-AQC)

The model fitting results for the multinomial logistic regression (MNL) under non-air quality crisis (N-AQC) conditions are summarized in

Table 7. Model summary of multinomial logistic regression for non-air quality crisis (N-AQC).

Model Info	Model Fitting Criteria	Likelihood Ratio Tests	Goodness-of-Fit	Pseudo R-Square
Model (N-AQC)	AIC: 939.296 (Intercept) 746.099 (Final)	Chi-Square: 259.197 df: 33 Sig.: 0.000 ***	Pearson Chi-Square: 1306.169 df: 1182 p = 0.007	Cox and Snell: 0.472 Nagelkerke: 0.525 McFadden: 0.278
	BIC: 951.316 (Intercept) 890.328 (Final)		Deviance: 674.099 df: 1182 p = 1.000
	−2 Log Likelihood: 933.296 (Intercept) 674.099 (Final)

Note: *** p < 0.001.

. The model fitting criteria indicated that the final model (AIC = 746.099; BIC = 890.328) performed substantially better than the intercept model (AIC = 939.296; BIC = 951.316), suggesting an improved fit with the inclusion of explanatory variables. The decrease in −2 Log Likelihood from 933.296 to 674.099 further supported this improvement. The likelihood ratio test revealed a statistically significant chi-square value of 259.197 with 33 degrees of freedom (p < 0.001), confirming that the predictors significantly contributed to explaining mode choice behavior during non-crisis periods [10,43].

In terms of goodness-of-fit, the deviance statistic was 674.099 with a p-value of 1.000, indicating an excellent model fit with no significant deviation from the saturated model. However, the Pearson chi-square value of 1306.169 (df = 1182, p = 0.007) suggested a marginal lack of fit, which could be attributed to the large sample size or data sparsity in certain categories [44]. The model’s explanatory power, assessed through pseudo R-square statistics, showed acceptable levels: Cox and Snell R² = 0.472, Nagelkerke R² = 0.525, and McFadden R² = 0.278. These values indicate that the model accounted for approximately 47% to 53% of the variance in mode choice [41].

The classification results in

Table 8. Percentage correct for non-air quality crisis (N-AQC).

Classification
Observed	Private Car	Motorcycle	Public Transport	Alternatives	Percent Correct
Private car	81	38	2	2	65.90%
Motorcycle	24	170	10	0	83.30%
Public transport	7	26	19	1	35.80%
Alternatives	6	12	4	4	15.40%
Overall Percentage	29.10%	60.60%	8.60%	1.70%	67.50%

evaluated the predictive accuracy of the model. The overall correct classification rate was 67.5%, indicating that the model correctly predicted two-thirds of the respondents’ travel mode choices under non-crisis conditions. The highest predictive accuracy was observed for motorcycle users at 83.3%, followed by private car users at 65.9%. In contrast, the model performed less effectively in predicting public transport users (35.8%) and those choosing alternative modes (15.4%). These findings suggest that while the model performs well for dominant travel modes, further refinement may be necessary to improve predictive performance for less frequently used or more heterogeneous modes [45,46].

4.4. Mode Choice for Air Quality Crisis (AQC)

4.4.1. Likelihood Ratio Tests and Collinearity Matrix (AQC)

The Likelihood ratio tests, as presented in

Table 9. Likelihood ratio tests of air quality crisis (AQC).

Effect	Chi-Square	df	Sig.
Intercept	22.067	3	<0.000 ***
Monthly income	16.396	3	<0.000 ***
Vehicle ownership	19.720	3	<0.000 ***
Holding motorcycle driving license	18.502	3	<0.000 ***
Holding private car driving license	42.441	3	<0.000 ***
Effect on finance	11.451	3	0.010 *
Travel time	28.396	3	<0.000 ***
Travel cost	17.775	3	<0.000 ***
Travel frequency	41.544	3	<0.000 ***

Note: * p < 0.05, *** p < 0.001.

, indicate that holding a private car driving license, travel frequency, travel time, and vehicle ownership are the most influential predictors. The test results demonstrate high statistical significance (p < 0.001). Monthly income, travel cost, and financial impacts of the air quality crisis also significantly affect mode choice. Notably, possession of a motorcycle license was not statistically significant. The collinearity matrix presented in

Table 10. Collinearity matrix of independent variables for air quality crisis (AQC).

	${\mathit{X}}_{\mathit{M}\mathit{I}}$	${\mathit{X}}_{\mathit{V}\mathit{O}}$	${\mathit{X}}_{\mathit{M}\mathit{C}\mathit{D}\mathit{L}}$	${\mathit{X}}_{\mathit{P}\mathit{C}\mathit{D}\mathit{L}}$	${\mathit{X}}_{\mathit{E}\mathit{F}}$	${\mathit{X}}_{\mathit{T}\mathit{T}}$	${\mathit{X}}_{\mathit{T}\mathit{C}}$	${\mathit{X}}_{\mathit{T}\mathit{F}}$
$X_{MI}$	-
$X_{VO}$	0.119	-
$X_{MCDL}$	0.048	0.294	-
$X_{PCDL}$	0.411	0.187	0.111	-
$X_{EF}$	−0.027	−0.084	0.031	−0.048	-
$X_{TT}$	0.161	0.114	0.074	0.215	−0.102	-
$X_{TC}$	0.215	0.079	−0.041	0.177	−0.068	0.559	-
$X_{TF}$	0.161	0.224	0.112	0.154	−0.018	0.264	0.140	-

Note: $X_{MI}:$ monthly income, $X_{VO}:$ vehicle ownership, $X_{MCDL}:$ holding a motorcycle driving license, $X_{PCDL}:$ holding a private car driving license, $X_{EF}:$ effect on finance, $X_{TT}:$ travel time, $X_{TC}:$ travel cost, $X_{TF}:$ travel frequency.

confirms the statistical soundness of the model through Pearson correlation analysis. The analysis reveals that no pairs of independent variables exceed the correlation coefficient threshold of 0.80, effectively addressing multicollinearity concerns [42]. The strongest observed correlation exists between travel time and cost (r = 0.559), while moderate correlations are identified between monthly income and private car license ownership (r = 0.411) and between vehicle ownership and motorcycle license possession (r = 0.294). All other variable pairs demonstrate weak correlations (r < 0.5), indicating their relative independence. These statistical relationships, particularly those involving vehicle access and travel characteristics, illuminate the significant role of socioeconomic factors in shaping travel behavior during periods of air quality crisis. The robust model structure, characterized by significant predictors and minimal variable interdependence, provides a reliable framework for understanding mode choice dynamics during an air quality crisis in this urban context.

4.4.2. Multinomial Logit Model Parameter Estimates and Utility Function (AQC)

An MNL model analysis was conducted to examine travel mode choice behavior during an air quality crisis, with private car serving as the reference category, as presented in

Table 11. MNL parameter estimates of air quality crisis model (AQC).

Mode	Variable	Coef.	Sig.	Odds Ratio
Motorcycle	Intercept	1.814	0.002 **
	$\mathrm{Monthly} \mathrm{income} (X_{MI}$)	−0.386	***	0.680
	$\mathrm{Vehicle} \mathrm{ownership} (X_{VO}$)	0.352	0.422	1.423
	$\mathrm{Holding} \mathrm{motorcycle} \mathrm{driving} \mathrm{license} (X_{MCDL}$)	1.087	***	2.967
	$\mathrm{Holding} \mathrm{private} \mathrm{car} \mathrm{driving} \mathrm{license} (X_{PCDL}$)	−1.886	***	0.152
	$\mathrm{Effect} \mathrm{on} \mathrm{finance} (X_{EF}$)	0.540	0.097	1.715
	$\mathrm{Travel} \mathrm{time} (X_{TT}$)	−0.326	0.002 **	0.722
	$\mathrm{Travel} \mathrm{cost} (X_{TC}$)	−0.282	0.041 *	0.755
	$\mathrm{Travel} \mathrm{frequency} (X_{TF}$)	0.147	0.252	1.158
Public transport	Intercept	1.539	0.039 *
	$\mathrm{Monthly} \mathrm{income} (X_{MI}$)	−0.400	0.040 *	0.670
	$\mathrm{Vehicle} \mathrm{ownership} (X_{VO}$)	−1.318	0.010 *	0.268
	$\mathrm{Holding} \mathrm{motorcycle} \mathrm{driving} \mathrm{license} (X_{MCDL}$)	−0.315	0.511	0.730
	$\mathrm{Holding} \mathrm{private} \mathrm{car} \mathrm{driving} \mathrm{license} (X_{PCDL}$)	−1.398	0.006 **	0.247
	$\mathrm{Effect} \mathrm{on} \mathrm{finance} (X_{EF}$)	1.304	0.005 **	3.685
	$\mathrm{Travel} \mathrm{time} (X_{TT}$)	0.337	0.025 *	1.400
	$\mathrm{Travel} \mathrm{cost} (X_{TC}$)	−0.226	0.283	0.797
	$\mathrm{Travel} \mathrm{frequency} (X_{TF}$)	−0.413	0.036 *	0.662
Alternatives	Intercept	2.798	***
	$\mathrm{Monthly} \mathrm{income} (X_{MI}$)	−0.058	0.652	0.944
	$\mathrm{Vehicle} \mathrm{ownership} (X_{VO}$)	−1.159	0.011 *	0.314
	$\mathrm{Holding} \mathrm{motorcycle} \mathrm{driving} \mathrm{license} (X_{MCDL}$)	0.086	0.828	1.090
	$\mathrm{Holding} \mathrm{private} \mathrm{car} \mathrm{driving} \mathrm{license} (X_{PCDL}$)	−1.101	0.009 **	0.332
	$\mathrm{Effect} \mathrm{on} \mathrm{finance} (X_{EF}$)	1.109	0.006 **	3.030
	$\mathrm{Travel} \mathrm{time} (X_{TT}$)	−0.384	0.012 *	0.681
	$\mathrm{Travel} \mathrm{cost} (X_{TC}$)	0.418	0.015 *	1.520
	$\mathrm{Travel} \mathrm{frequency} (X_{TF}$)	−0.849	***	0.428

Note: * p < 0.05, ** p < 0.01, *** p < 0.001; reference mode: private car.

. The model identified several significant factors that influence mode choice during an air quality crisis. The utility function for mode choice during an air quality crisis (AQC) was specified as Equation (6).

$$U_{AQC}=\alpha X_{MI}+\beta X_{VO}+\gamma X_{MCDL}+\delta X_{PCDL}+\epsilon X_{EF}+\zeta X_{TT}+\eta X_{TC}+\theta X_{TF}+c$$

(6)

Here, the variables represent $X_{MI}:$ monthly income, $X_{VO}:$ vehicle ownership, $X_{MCDL}:$ holding a motorcycle driving license, $X_{PCDL}:$ holding a private car driving license, $X_{EF}:$ effect on finance, $X_{TT}:$ travel time, $X_{TC}:$ travel cost, and $X_{TF}:$ travel frequency.

The analysis of motorcycle usage during air quality crisis revealed that income maintains a significant negative effect ($\alpha$ = −0.386), indicating that higher-income individuals are less likely to choose motorcycles over private cars. License possession continued to show contrasting effects; holding a motorcycle license remained positively significant ($\gamma$ = 1.087), while possession of a car license decreased motorcycle utility ($\delta$ = −1.886). Both travel time ($\zeta$ = −0.326) and costs ($\eta$ = −0.282) emerged as significant deterrents during air quality crisis conditions, suggesting increased sensitivity to these factors during an air quality crisis.

For public transport, income demonstrated a persistent negative influence ($\alpha$ = −0.400), while vehicle ownership substantially reduced its utility ($\beta$ = −1.318). Notably, financial considerations showed heightened importance during the air quality crisis ($\epsilon$ = 1.304), suggesting that economic factors become more crucial in mode choice decisions during these periods. Travel frequency exhibited a negative association ($\theta$ = −0.413), indicating that frequent travelers were less likely to opt for public transport during air quality crisis conditions. Alternative modes revealed distinct patterns during the air quality crisis, with financial effects showing a significant positive influence ($\epsilon$ = 1.109), while travel frequency demonstrated a strong negative relationship ($\theta$ = −0.849). Vehicle ownership maintained its negative influence ($\beta$ = −1.159) on the choice of alternatives. Interestingly, travel time ($\zeta$ = −0.384) and travel costs ($\eta$ = 0.418) showed opposing effects during air quality crisis conditions, suggesting a complex trade-off between these factors in mode choice decisions.

These findings highlight the significant shifts in travel mode choice behavior during an air quality crisis, particularly in how individuals weigh various factors such as financial impacts, travel characteristics, and personal resources in their transportation decisions. The differential coefficient patterns between N-AQC and AQC periods suggest the presence of implicit interaction effects between environmental conditions and behavioral determinants. For instance, the emergence of travel frequency as significant only during AQC periods (coefficients ranging from −0.849 to 0.147 across modes), as shown in Table 11, while demographic variables lose significance, indicates that environmental stressors fundamentally alter decision-making processes. The results provide valuable insights for transportation planning and policy development during an air quality crisis.

4.4.3. Model Fitting and Predictive Accuracy of Air Quality Crisis Model (AQC)

The model fitting results for the multinomial logistic regression (MNL) under air quality crisis (AQC) conditions are presented in

Table 12. Model summary of multinomial logistic regression for air quality crisis (AQC).

Model Info	Model Fitting Criteria	Likelihood Ratio Tests	Goodness-of-Fit	Pseudo R-Square
Model (AQC)	AIC: 967.119 (Intercept) 745.348 (Final)	Chi-Square: 269.771 df: 24 Sig.: 0.000 ***	Pearson Chi-Square: 1054.790 df: 1047 p = 0.427	Cox and Snell: 0.485 Nagelkerke: 0.532 McFadden: 0.273
	BIC: 979.138 (Intercept) 853.520 (Final)		Deviance: 669.272 df: 1047 p = 1.000
	−2 Log Likelihood: 961.119 (Intercept) 691.348 (Final)

Note: *** p < 0.001.

. The reduction in AIC from 967.119 (intercept) to 745.348 (final model), and in BIC from 979.138 to 853.520, indicates that the model with predictors performs substantially better. Similarly, the drop in −2 Log Likelihood from 961.119 to 691.348 supports an improved model fit. The likelihood ratio test reported a significant chi-square value of 269.771 with 24 degrees of freedom (p < 0.001), suggesting that the final model provides a significantly better fit than the null model [35,47]. Regarding model adequacy, the deviance statistic was 669.272 with a non-significant p-value (p = 1.000), indicating no significant deviation from the saturated model. Additionally, the Pearson chi-square (1054.790, df = 1047, p = 0.427) further supports a good overall fit (Louviere, Hensher, & Swait, 2000) [48]. The pseudo R-square values Cox and Snell R² = 0.485, Nagelkerke R² = 0.532, and McFadden R² = 0.273 indicate56moderate explanatory power of the model, with Nagelkerke R² exceeding the commonly accepted threshold of 0.5 for behavioral models [49,50].

Table 13. Percentage correct for air quality crisis (AQC).

Classification
Observed	Private Car	Motorcycle	Public Transport	Alternatives	Percent Correct
Private car	91	37	4	9	64.50%
Motorcycle	24	141	2	7	81.00%
Public transport	8	13	6	8	17.10%
Alternatives	10	23	2	21	37.50%
Overall Percentage	32.80%	52.70%	3.40%	11.10%	63.80%

presents the model’s classification accuracy. An overall correct prediction rate of 63.8% was achieved, indicating that nearly two-thirds of actual mode choices were correctly classified. The highest prediction accuracy occurred for motorcycle users (81.0%), followed by private car users (64.5%). In contrast, the model showed weaker predictive accuracy for public transport (17.1%) and alternative modes (37.5%). This distribution of accuracy suggests that while the model is effective in capturing dominant travel behaviors, it may require further refinement or additional variables to improve prediction for less frequently chosen modes [51,52].

The analysis revealed significant differences in mode choice factors between non-air quality crisis (N-AQC) and air quality crisis (AQC) periods in Chiang Rai, Thailand, as shown in

Table 14. Significant variables in non-air quality crisis and air quality crisis models.

Variable	Significance in N-AQC Model	Significance in AQC Model	Interpretation
Gender	¡		Significant only during non-air quality crisis
Age	¡		Relevant for mode choice only in non-air quality crisis
Monthly income	¡	¡	Key factor influencing mode choice in both periods
Marital status	¡		Only significant under non-air quality crisis
Vehicle ownership	¡	¡	Consistently significant; access to a vehicle strongly affects mode selection
Holding motorcycle driving license	¡	¡	Strong predictor for motorcycle use in both periods
Holding private car driving license	¡	¡	Influences private car and alternative mode decisions in both periods
Effect on healthcare	¡		Perceived health impact mattered only during non-air quality crisis
Effect on finance	¡	¡	Financial concern significantly affects decisions in both scenarios
Travel time	¡	¡	Travel duration influences choice across both periods
Travel cost	¡	¡	Cost remains a significant determinant under all air quality conditions
Travel frequency		¡	Becomes significant only during air quality crisis, indicating crisis-driven behavior change

Note: N-AQC is non-air quality crisis; AQC is air quality crisis; ¡ is statistically significant (p < 0.05).

. Core economic and operational variables, including monthly income, vehicle ownership, driving licenses, travel time, and travel cost, remained significant across both conditions. However, demographic factors (gender, age, and marital status) were only significant during non-air quality crisis, while travel frequency emerged as significant exclusively during air quality crisis. This shift in significant variables suggests that environmental conditions substantially alter commuters’ mode choice decision-making processes, with economic considerations maintaining importance regardless of air quality conditions. These findings provide valuable insights for developing targeted transportation policies that can adapt to varying air quality conditions.

4.5. The Predicted Probabilities of Transportation Mode

Table 15. The predicted probabilities of transportation mode during non-air quality crisis (%).

	Private Car	Motorcycle	Public Transport	Alternatives
Gender
Male	32	60	4	4
Female	28	54	12	6
Others	21	41	32	7
Age (years)
<21	33	59	4	4
21–30	29	56	9	5
31–40	24	50	19	7
41–50	18	40	33	8
51–60	12	28	51	9
Monthly income (THB)
<10,000	20	64	11	5
10,001–15,000	28	57	10	5
15,000–20,000	38	49	8	6
20,001–30,000	48	40	6	6
30,001–40,000	59	31	5	5
40,001–50,000	69	23	3	5
>50,001	77	16	2	4
Marital status
Unmarried	27	57	10	6
Married	38	51	7	3
Not mentioned	51	43	5	1
Vehicle ownership
No	17	42	28	13
Yes	32	57	7	4
Holding motorcycle driving license
No	39	37	15	9
Yes	21	70	6	3
Holding private car driving license
No	17	67	9	7
Yes	53	36	8	3
Effect on healthcare
No	32	29	25	15
Yes	28	60	8	4
Effect on finance
No	31	57	7	4
Yes	23	51	18	8
Travel time (Minutes)
<10	28	63	4	5
10–20	29	60	7	5
21–30	29	56	10	5
31–40	29	51	15	6
41–50	29	45	21	6
51–60	27	38	29	5
>60	25	32	39	5
Travel cost (THB)
<50	20	66	10	3
50–100	29	57	9	5
101–150	38	47	8	8
151–200	47	36	7	11
>200	54	26	5	15

Note: The predicted probabilities were calculated using the MNL model presented in Table 6. When generating predictions for each variable of interest, all other quantitative variables were held constant at their respective mean values, while categorical variables were fixed at their modal values.

and

Table 16. The predicted probabilities of transportation mode during air quality crisis (%).

	Private Car	Motorcycle	Public Transport	Alternatives
Monthly income (THB)
<10,000	29	54	8	8
10,001–15,000	36	47	7	10
15,000–20,000	44	39	6	12
20,001–30,000	52	31	5	13
30,001–40,000	59	24	4	14
40,001–50,000	65	18	3	14
>50,001	70	13	2	15
Vehicle ownership
No	32	29	17	22
Yes	37	49	6	8
Holding motorcycle driving license
No	47	31	10	12
Yes	29	58	5	8
Holding private car driving license
No	23	60	8	10
Yes	62	25	5	9
Effect on finance
No	42	44	5	8
Yes	26	46	12	16
Travel time (Minutes)
<10	26	57	3	14
10–20	32	51	5	12
21–30	38	44	8	10
31–40	44	36	12	8
Travel time (Minutes)
41–50	47	28	18	6
51–60	49	21	26	4
>60	47	15	36	3
Travel cost (THB)
<50	32	55	8	5
50–100	36	47	7	9
101–150	40	39	6	15
151–200	41	30	5	23
>200	40	22	4	34
Travel frequency (per weeks)
No travel	24	20	11	45
1	32	32	10	26
2–3	37	43	8	13
4–5	38	51	5	6
6–7	51	20	12	16

Note: The predicted probabilities were calculated using the MNL model presented in Table 11. When generating predictions for each variable of interest, all other quantitative variables were held constant at their respective mean values, while categorical variables were fixed at their modal values.

present the predicted probabilities of transportation mode choices during non-air quality crisis and air quality crisis in Chiang Rai, Thailand. Transport mode preferences are strongly influenced by socioeconomic and demographic factors. Motorcycles emerge as the dominant mode of transport, particularly favored by younger individuals and lower-income groups. Private car usage increases significantly with income levels, while public transport becomes more prevalent among older age groups. Travel characteristics also play a key role, with longer journeys associated with higher public transport use and higher travel costs linked to increased private car usage. Vehicle ownership and possession of driving licenses strongly correlate with respective mode choices, with non-vehicle owners showing greater reliance on public transport. These findings reflect typical transportation patterns observed in Southeast Asian urban contexts, characterized by high motorcycle dependency among certain demographic groups.

For predicted probabilities of transportation mode choices during the air quality crisis in Chiang Rai, socioeconomic factors, particularly monthly income levels, significantly influence transportation preferences, with higher-income groups demonstrating a strong preference for private cars while lower-income groups predominantly rely on motorcycles. Vehicle ownership and licensing status also play crucial roles in mode selection, with license holders typically choosing their respective vehicle types and non-vehicle owners showing higher public transport usage. Travel characteristics emerge as important determinants—longer journey durations correlate with increased public transport use, while shorter journeys associate with higher motorcycle usage, and higher travel costs lead to greater utilization of alternative modes. The frequency of travel impacts mode selection, with regular commuters showing stronger preferences for private cars while occasional travelers demonstrate more varied choices. Additionally, the financial impact of the air quality crisis notably influences transportation decisions, where individuals reporting financial effects show distinct changes in their choices, including reduced private car usage and increased utilization of both public transport and alternative modes, demonstrating how economic constraints during an environmental crisis can significantly reshape transportation behavior.

The analysis of travel mode choices in Chiang Rai reveals distinct shifts between non-air quality crisis and air quality crisis. During an air quality crisis, private car usage decreased among higher-income groups (from 77 to 70%), while motorcycle dependency reduced among lower-income groups (from 64 to 54%). Public transport usage also declined, particularly among non-vehicle owners (from 28 to 17%). These changes indicate four key impacts of the air quality crisis: reduced overall mobility, preference for enclosed transport modes, increased health considerations, and stronger economic influences on travel decisions. These findings contribute to understanding environmental crisis impacts on travel behavior in Southeast Asian contexts, particularly in areas with high motorcycle dependency.

5. Discussion

5.1. Key Finding

The analysis revealed significant modal shifts in travel behavior during air quality crisis in Chiang Rai, Thailand, with private car usage increasing from 30.30% to 34.70% and motorcycle usage decreasing from 50.20% to 42.90%, while public transport declined from 13.10% to 8.60% and alternative modes increased substantially from 6.40% to 13.80%, indicating a clear preference for enclosed transportation during air quality crisis. Factors influencing mode choice differed between crisis and non-crisis periods, with socioeconomic variables including monthly income, vehicle ownership, and driving licenses remaining significant across both periods, while demographic factors such as gender, age, and marital status were only significant during non-crisis periods, and travel frequency emerged as a crucial determinant specifically during the air quality crisis. Economic factors became more influential during crisis periods, with financial impacts particularly pronounced among lower-income groups who shifted from private vehicles to public transport or alternative modes, while travel characteristics showed varying impacts across periods, with shorter trips becoming more prevalent during crisis and increased sensitivity to travel time and cost, particularly among motorcycle users, reflecting attempts to minimize pollution exposure.

5.2. Policy Implication

The analysis revealed significant modal shifts in travel behavior during an air quality crisis in Chiang Rai, Thailand. Specifically, private car usage increased from 30.30% to 34.70%, while motorcycle usage decreased from 50.20% to 42.90%. Public transport usage declined from 13.10% to 8.60%, and alternative modes (e.g., walking, biking, e-hailing) rose notably from 6.40% to 13.80%. These shifts highlight a clear preference for enclosed transportation during an air quality crisis, driven by concerns over pollutant exposure. Socioeconomic factors such as monthly income, vehicle ownership, and driving license status remained consistently influential across both the non-air quality crisis and air quality crisis. In contrast, demographic variables (gender, age, marital status) were significant only in non-crisis periods, while travel frequency emerged as a critical determinant specifically during air quality crises. These findings underscore the role of economic and behavioral adaptation in response to environmental stress and support the need for responsive transportation policies that can ensure mobility, safety, and equity during environmental crises.

Importantly, this study aligns with the United Nations Sustainable Development Goals (SDGs) [53], particularly SDG 11: Sustainable Cities and Communities, by providing actionable insights into how urban transport systems can become more resilient and inclusive in the face of environmental challenges. The findings also contribute to SDG 3: Good Health and Well-being by addressing how transport behaviors interact with public health risks from air pollution.

5.2.1. Low-Emission Public Transport Fleet Transition

The observed shift toward enclosed modes during air quality crisis conditions necessitates a phased transition to low-emission public transport, particularly targeting the Song-Teaw (mini-bus) fleet. Implementation should begin with installing air filtration systems in existing vehicles, followed by gradually replacing the fleet with electric alternatives through subsidized purchase programs [54]. The primary beneficiaries would be lower-income groups (53.9% of respondents earning <10,000 THB monthly) who rely heavily on public transportation during a crisis. To contextualize these income levels, 10,000 THB (approximately 308 USD) represents the minimum wage threshold in Thailand. For reference, typical local transportation costs in Chiang Rai include motorcycle taxi fares of 20–40 THB (0.60–1.25 USD) per trip, Song-Teaw (mini-bus) fares of 10–15 THB (0.30–0.45 USD), private car fuel costs of approximately 35–40 THB (1.08–1.23 USD) per liter of gasoline, and monthly motorcycle maintenance averaging 500–800 THB (15–25 USD). These transportation expenses can represent 15–25% of monthly income for the lowest-income respondents, making them particularly sensitive to mode choice decisions during an air quality crisis when health considerations may conflict with economic constraints [55].

5.2.2. Integrated Air Quality Monitoring and Transportation Information System

The finding that travel frequency becomes a significant factor during air quality crisis indicates the need for improved information systems to support decision-making. A comprehensive approach involving air quality monitoring stations integrated with transportation information platforms would enable travelers to make informed decisions during pollution events. Real-time air quality information has been shown to significantly influence travel decisions during pollution events in Beijing [56]. Such systems reduce health impacts through behavioral adaptation, with potential healthcare cost savings through decreased respiratory admissions during pollution events. Implementation would require inter-agency coordination within existing governance frameworks as documented in analyses of energy and environmental policy coordination in ASEAN countries [57].

5.2.3. Financial Support Mechanisms for Sustainable Mode Shifts

The analysis revealed significant financial influences on mode choice during an air quality crisis. Financial intervention strategies could include air quality-responsive fare systems with public transport fare reductions during severe pollution events, targeted transportation vouchers for lower-income households, and incentive programs for businesses providing alternative transportation options. Income levels significantly influence travel behavior during air pollution events [20]. Such measures would particularly benefit the 17.7% of respondents without vehicle ownership [58].

5.2.4. Active Transportation Infrastructure with Air Pollution Protection

The significant increase in alternative usage (from 6.4% to 13.8%) during an air quality crisis indicates substantial potential for growth with appropriate protective infrastructure. Infrastructure development could include separated bicycle and pedestrian pathways with vegetative barriers, enclosed air-filtered waiting areas at transportation nodes, and targeted subsidy programs for electric micromobility options. Vegetation barriers can reduce particulate matter exposure by 15–30% for active transportation users [58]. These measures would benefit the 68.2% of student respondents [24].

5.3. Limitations and Future Work

Several methodological limitations warrant acknowledgment. The probability sampling approach resulted in significant demographic skewing, with overrepresentation of younger respondents (85.9% under 30 years) and students (68.2%) compared to Chiang Rai’s general population [32]. This bias limits generalizability to older adults and working professionals. The cross-sectional design and reliance on stated preferences introduce potential recall bias regarding pre-crisis travel patterns. Methodologically, separate multinomial logit models lacked explicit interaction terms between environmental conditions and individual characteristics. The aggregation of taxi services and active transportation into the “Alternatives” category potentially masked distinct behavioral responses due to different pollution exposure profiles.

Future research should address these limitations through several approaches. Methodological improvements should incorporate explicit interaction terms to capture how environmental conditions moderate relationships between socioeconomic characteristics and mode choice. Mixed-method sampling strategies combining probability and non-probability techniques would improve demographic representation. Longitudinal studies tracking behavioral changes over multiple air quality crisis seasons would provide insights into adaptation patterns and the sustainability of behavioral changes. Alternative-specific data collection enabling comprehensive modeling of all transportation options is recommended. Future studies should examine taxi services and active transportation modes separately, investigate policy intervention effectiveness, and explore emerging transportation technologies and remote work options in addressing air quality crisis challenges.

6. Conclusions

This study examined travel mode choice adaptations during an air quality crisis in Chiang Rai, Thailand, revealing significant behavioral shifts toward enclosed transportation modes and shorter trips during the air quality crisis. The findings demonstrate systematic changes in transportation preferences, with motorcycle usage decreasing from 50.2% to 42.9% while private car usage increased from 30.3% to 34.7% during crisis periods. Financial considerations became increasingly influential in mode choice decisions during an environmental crisis, particularly among the younger, student-dominated sample population. While sample demographics limit generalizability to older adults and working professionals, the results provide valuable insights into an important demographic segment representing future transportation users. The research contributes to understanding how environmental crises impact urban mobility patterns and emphasizes the importance of integrating environmental and socioeconomic factors in transportation planning. The findings indicate that effective crisis-period transportation policies must address both infrastructure limitations and socioeconomic barriers to mode switching. Incorporating broader demographic representation will be essential for developing comprehensive transportation policies that can effectively respond to the air quality crisis across all population segments. The study establishes a foundation for creating more resilient urban transportation systems capable of adapting to environmental challenges while addressing equity considerations in policy development.