Identifying School Travel Mode Choice Patterns in Mersin, Türkiye

Abstract

This study investigates the factors affecting the choice of school travel mode among students in Mersin, Türkiye, focusing on walking, private car, public transit and school bus. A two-step modeling approach was adopted. First, a latent class cluster analysis (LCCA) was applied to identify subgroups of students with similar characteristics. Then, separate multinomial logit (MNL) models were estimated for each cluster. The data come from the 2022 Urban Transport Master Plan household survey and include 2798 students from 2092 households. The results show that trip distance is the most consistent and significant factor across all clusters, as increasing distance makes students more likely to use motorized modes instead of walking. Gender also demonstrates a consistent influence in specific clusters, where male students are less likely to travel by private car. Similarly, residing in a single-family house consistently increases the likelihood of car use in multiple clusters. Conversely, the influence of household structure, parental education, income, and household size differs significantly between clusters, underlining the importance of considering group-level differences in school travel behavior. These findings suggest that policies aiming to promote sustainable school travel should be sensitive to the needs of different student groups. Integrating land use and transportation planning may help to support active and shared modes of travel.

1. Introduction

Children’s daily travel to school is a significant aspect of urban mobility, with far-reaching implications for traffic congestion, environmental sustainability, and public health. As cities expand and educational facilities are increasingly located outside traditional residential areas, how students commute to school has become a growing concern for urban planners and policymakers. While school travel behavior has been widely studied in high-income countries, research remains limited in contexts like Türkiye, where distinct urban dynamics, socioeconomic diversity, and evolving spatial patterns warrant closer attention. This study addresses this gap by examining the key factors influencing the choice of school travel mode in Türkiye, drawing on a broad set of individual, household, and environmental characteristics.

Children, especially those enrolled in schools, need transportation options similar to those of adults [1]. However, the decision-making processes related to children’s transportation needs differ from those of adults [2]. Generally, children rely on adults to decide where, when, and how they will travel [3,4]. Research shows that regular physical activity and the use of active transportation modes enhance children’s social well-being and cognitive abilities [5,6]. Active transportation involves traveling by modes dependent on human power, such as walking or cycling [7].

Biking and walking are encouraged around the world to help children become more active and reduce obesity. Including these activities in children’s daily routines, especially for school trips, has important health benefits [8,9]. Children who regularly walk or bike to school are more likely to maintain this healthy and active lifestyle by choosing sustainable travel practices as they grow older [10,11]. Cardiovascular diseases and other health issues in adulthood are often linked to a lack of physical activity during childhood and adolescence [12].

In the literature, it is increasingly evident that families are choosing larger, more dispersed schools over smaller, neighborhood-based ones. This leads to fewer opportunities for children’s active transportation and greater reliance on motorized vehicles [13]. This shift worsens problems such as air pollution, energy consumption, and traffic congestion during peak urban hours [14,15]. The trend is observed in both developed and developing countries [16]. At the same time, urban sprawl, along with concerns about transportation safety and declining walkability conditions for students, has contributed to a growing dependence on motorized modes of school travel in recent years.

McDonald et al. [17] stated that 10–14% of morning traffic in the U.S. is caused by parents dropping their children off at school. Addressing this issue could help reduce congestion and vehicle emissions, making roads safer for vulnerable users [18]. This effect becomes especially evident during school holidays, when peak-hour traffic volumes noticeably drop, particularly around school areas [19]. Increased vehicular traffic on school routes and concerns about traffic safety for pedestrians and cyclists have contributed to a decline in children using active transportation modes. This downward trend has been exacerbated by parents’ growing concerns about their children’s personal safety [20,21,22]. The proportion of school trips made by elementary and middle school students using active modes in the U.S. dropped from 48% in 1969 to 13% by 2009 [17]. Similarly, in Australia, the percentage of children using active and independent travel to school fell from 61% to 32% between 1991 and 2012 [23]. This phenomenon is observed not only globally but also in Türkiye. Urbanization and population growth have led authorities to invest in new schools, often in peripheral areas due to limited land in city centers. As a result, students increasingly rely on school buses, public transit, or parental accompaniment, contributing to a decline in public transit use and a rise in private car dependency in urban areas. However, unlike many developed countries, a significant proportion of middle and high school students in Türkiye—about 55%—still walk to school. Similar to other developing countries, school travel choices in Türkiye may be influenced by unique geographical and cultural factors, making them distinct from more developed regions [9].

To the best of our knowledge, only one recent study from Türkiye has examined school travel behavior [24], focusing on the impact of built environment characteristics on children’s choice to walk to school. However, that study was limited to a single travel mode and did not account for socio-demographic or household-level factors.

This study offers several contributions to the literature. First, it analyzes multiple school travel modes—walking, car, public transit, and school bus—in an integrated manner. Second, it considers a wide range of explanatory variables, including student age, gender, school grade, household size and income, siblings, parental employment and education, car ownership, and travel distance. Third, it is the first study in Türkiye to address the choice of school travel mode with this level of detail and breadth. It also employs a latent class approach to capture unobserved heterogeneity, which enhances the explanatory power of the analysis. Finally, it offers new insights from a developing country context, contributing to a more balanced understanding of global school travel behavior and informing evidence-based transport planning.

The remainder of this paper is organized as follows. Section 2 presents the literature review, summarizing key findings from previous research on the choice of school travel mode. Section 3 describes the study area and the dataset used in this research. Section 4 outlines the methodology, including the modeling approach and variable definitions. Section 5 presents the model’s results, followed by a discussion in Section 6. Finally, Section 7 provides the main conclusions and policy implications.

2. Literature Review

There is an extensive body of literature exploring the factors that influence children’s school travel, with most studies conducted in developed countries. The findings reveal that various elements play a role in shaping children’s transportation preferences. Table 1 provides a summary of selected studies, including key aspects such as study location, students’ age, modes of travel, and methodology.

Distance between home and school stands out as a pivotal factor influencing mode choice [8,25,26]. As the distance increases, the use of active transportation modes for school commutes tends to decrease [27,28,29]. Meanwhile, the use of school buses [29], cars [30], and public transit [2] has increased, along with parental driving [31]. Scheiner et al. [32] argued that the threshold for abandoning active modes depends on cultural norms and environmental conditions. In China and the U.S., walking is preferred up to 500 m [33,34], while in Germany, Müller et al. [35] reported a 1 km threshold. In Australia, Timperio et al. [36] emphasized that walking and biking are preferred by children of all age groups up to 800 m. Studies from Canada and the U.S. have identified an upper threshold for active travel at around 1.6 km [37,38].

The built environment is an important factor in the use of active transportation [39]. Pedestrian- and cyclist-friendly infrastructure has been shown to increase the use of active transportation modes [27,36]. In contrast, inadequate infrastructure—such as high-traffic streets and busy intersections—usually leads to more car use [40,41,42,43]. Conversely, Mandic et al. [31] concluded that a high intersection density in residential areas supports active school transportation.

Broberg and Sarjala [44] reported that as urbanization levels rise, public transit becomes a more common choice than active transportation. Similarly, Yarlagadda and Srinivasan [3] found that children living in areas with high employment rates are more likely to use public transit, although Braza et al. [45] argued that high employment rates support walking. Studies show mixed findings regarding land use and street density. Some suggest that a mixed land use and high street density encourage walking [37,46,47], while others report the opposite effect [39,41]. Features such as short block lengths, a high density of sidewalks, good street connectivity, and tree coverage along school routes have been found to encourage walking [24,34,39,48]. Although many aspects of residential land use support active transportation, Lin and He [8] found that residential density itself is not a significant factor in school travel.

Household socio-demographic characteristics—including income level [41,47], car ownership [27,28,29], the number of licensed drivers [13], work obligations, and travel behavior [3,49]—play a crucial role in determining the choice of school travel mode. Higher income levels and car ownership are generally associated with a greater likelihood that parents will drive their children to school [9,22,50]. Lidbe [51] indicated that in the U.S., children from low-income families tend to prefer taking the school bus. However, Lin and He [8] found no significant relationship between income and active transportation in their study in China

Other household dynamics also influence this decision. For example, Mitra and Buliung [27] found that if adults are at home during school commute times, children are more likely to be driven by private car. Li and Zhao [52] noted that a greater number of family members increases the chance of children being driven to school. Flexible working hours for parents also make it more likely that they will accompany their children to school [53]. McDonald [49] found that fathers’ commuting patterns do not strongly affect children’s active travel, whereas children whose mothers work in the morning are less likely to walk. Students with unemployed parents are more likely to walk [26,43].

The influence of children’s age [9,54,55] and gender [56,57,58,59] on mode choice has been explored in various studies. One exception is Lin and He [8], who found no association between age or gender and choice of school travel mode. As children grow older, they tend to become more independent, reducing the likelihood of being accompanied by parents and increasing the use of independent modes such as bicycles, buses, or walking [19,59]. Dias et al. [9] found that this independence leads older students to prefer walking as their primary mode of transportation. Males are generally more likely to use active or independent transportation modes compared to females [43,59,60,61], while females are more often driven to school by their parents [51,62]. McMillan et al. [63] noted that males tend to use active transportation more frequently at younger ages. Clifton [64] observed that young adolescents use active modes more often than older ones but quickly abandon them when driving becomes an option. McDonald [65] suggested that the combined effects of age and gender shape choices of school travel mode.

Martin et al. [66] observed that students from families with educated parents are less likely to use public transport or walk. Similarly, AlQuhtani [43] found that higher parental education levels are associated with lower levels of active commuting. However, Ozbil et al. [24] reported the opposite effect. It has also been noted that parents tend to be more cautious about the mode choice for young girls, although gender differences tend to narrow as children get older. McDonald [65] highlighted that a higher number of siblings reduces the reliance on cars for school travel.

Families are more likely to support active transportation in areas perceived as walkable [37,41,67]. In contrast, traffic [68,69] and crime [55,70] negatively impact the use of active modes. Van den Berg et al. [71] concluded that parental safety perceptions are shaped by children’s age, income, neighborhood infrastructure, and social cohesion. Pabayo et al. [72] observed that when older and younger siblings travel together, parental concerns decrease. Families worried about punctuality and comfort may prefer driving their children to school or using the school bus [73].

The literature shows that factors such as distance to school, household income, car ownership, and parental employment strongly affect the choice of school travel mode. Built environment features and safety concerns are also important, especially for younger students and girls. Although many studies in developed countries examine multiple travel modes, they typically do not account for unobserved heterogeneity in travel behavior. To address this gap and provide evidence from a developing country context, recent data from Mersin, Türkiye, were analyzed using a latent class modeling approach, which better captures variations in travel choices across different student groups.

Table 1. Summary of selected school trip studies.

Study	Country	Age Group	Active	Car	Public Transit	School Bus	Others	Method
Liu et al. [54]	China	6–18	x					GWLR
Dias et al. [9]	Sri Lanka	6–18	x	x	x	x	x	MNL, Mix logit
Li et al. [59]	China	6–18	x	x	x		x	MNL
Müller et al. [19]	Germany	10–19	x	x	x			Nested logit
Ozbil et al. [24]	Türkiye	12–14	x					Nominal. logistic reg.
Lin and He [8]	China	6–13	x					Binary logit
Scheiner et al. [32]	Germany	6–10	x	x	x			MNL
Distefano et al. [74]	Italy	3–11	x	x				SEM
Singh and Vasudevan [75]	India	5–15	x	x		x	x	MNL
Stark et al. [2]	Austria, Germany	12, 13	x	x	x			SEM
Ermagun and Samimi [16]	Iran	12–17	x	x	x	x		Copula-based joint
Woldeamanuel [76]	U.S.	12–16	x	x	x			Binary logit
Ermagun et al. [73]	Iran	12–17	x	x	x	x		Random forest
Li and Zhao [52]	China	13–15	x	x	x			MNL
Mitra and Buliung [77]	Canada	11, 14, 15	x	x	x	x		MNL
Guliani et al. [60]	Canada	10–12	x	x				SEM
Broberg and Sarjala [44]	Finland	11–14	x	x				MNL
Elias and Katoshevski-Cavari [78]	Israel	9–15	x	x		x		MNL
Noland et al. [79]	U.S.	3–14	x	x		x	x	Mix logit
Nevelsteen et al. [80]	Belgium	6–12	x	x	x			Logistic regression
He [81]	U.S.	5–18	x			x		MNL
Alemu and Tsutsumi [82]	Japan	15–18	x	x	x			MNL
Mitra et al. [83]	Canada	11–13	x	x				Binary logit
McDonald [84]	U.S.	5–13	x	x	x			MNL
Mota et al. [85]	Portugal	12–16	x	x	x			Logistic regression

Table 1. Summary of selected school trip studies.

Study	Country	Age Group	Active	Car	Public Transit	School Bus	Others	Method
Liu et al. [54]	China	6–18	x					GWLR
Dias et al. [9]	Sri Lanka	6–18	x	x	x	x	x	MNL, Mix logit
Li et al. [59]	China	6–18	x	x	x		x	MNL
Müller et al. [19]	Germany	10–19	x	x	x			Nested logit
Ozbil et al. [24]	Türkiye	12–14	x					Nominal. logistic reg.
Lin and He [8]	China	6–13	x					Binary logit
Scheiner et al. [32]	Germany	6–10	x	x	x			MNL
Distefano et al. [74]	Italy	3–11	x	x				SEM
Singh and Vasudevan [75]	India	5–15	x	x		x	x	MNL
Stark et al. [2]	Austria, Germany	12, 13	x	x	x			SEM
Ermagun and Samimi [16]	Iran	12–17	x	x	x	x		Copula-based joint
Woldeamanuel [76]	U.S.	12–16	x	x	x			Binary logit
Ermagun et al. [73]	Iran	12–17	x	x	x	x		Random forest
Li and Zhao [52]	China	13–15	x	x	x			MNL
Mitra and Buliung [77]	Canada	11, 14, 15	x	x	x	x		MNL
Guliani et al. [60]	Canada	10–12	x	x				SEM
Broberg and Sarjala [44]	Finland	11–14	x	x				MNL
Elias and Katoshevski-Cavari [78]	Israel	9–15	x	x		x		MNL
Noland et al. [79]	U.S.	3–14	x	x		x	x	Mix logit
Nevelsteen et al. [80]	Belgium	6–12	x	x	x			Logistic regression
He [81]	U.S.	5–18	x			x		MNL
Alemu and Tsutsumi [82]	Japan	15–18	x	x	x			MNL
Mitra et al. [83]	Canada	11–13	x	x				Binary logit
McDonald [84]	U.S.	5–13	x	x	x			MNL
Mota et al. [85]	Portugal	12–16	x	x	x			Logistic regression

3. Study Area and Data

Mersin City, situated on Türkiye’s southern coast, is the focus of this research, with a population slightly exceeding one million in 2022 [86]. During the late 20th century, Mersin City experienced rapid industrialization and urban growth, propelling it to become the 10th largest city in Türkiye. Consequently, both the general population and the student population saw significant increases. Over this period, the number of cars per thousand people surged from 69 to 150, while the average daily trips per person rose from 1.28 to 1.64 (see

Table 2. Basic indicators [87,88].

Indicator	2000	2022
Population	661,740	1,026,398
Trips per day	847,027	1,681,688
Trip per capita per day	1.28	1.64
Car ownership per 1000 population	69	150
Student population	146,977	224,724
Primary and middle school students	108,720	127,868
High school and university students	38,257	96,856

).

The dataset for this study is sourced from a travel survey carried out among residents of Mersin City as part of the Urban Transport Master Plan in 2022 [87]. The households were selected through random sampling, with approximately 3% of households surveyed in each of the 140 neighborhoods of Mersin City. All members of the selected 12,324 households were asked to report their one-day travel activities, resulting in a total of 48,832 trip records. Alongside travel details such as travel duration, starting and ending points, trip purpose, and mode of transportation, respondents were also asked about pertinent household characteristics, including household size, vehicle ownership, income, employment status, and housing situation. For the current research, while the travel survey covered all of Mersin City, the focal areas selected for this study specifically include the city center and the most urbanized regions of Mersin City, encompassing Akdeniz, Mezitli, Toroslar, and Yenişehir. Within these areas, middle school students (aged 11–14 years old) and high school students (aged 15–18 years old) were chosen as the target population.

Table 3. Summary of sample characteristics.

Category	Frequency	Percentage
Trip Mode
Walking	3062	55.3
Public transit	1417	25.6
School bus	806	14.6
Car	249	4.5
Trip Distance
<0.5 km	1953	35.3
0.5 km–1.5 km	1030	18.6
1.5 km–2.5 km	745	13.5
≥2.5 km	1806	32.6
Trip Direction
To school	2788	50.4
From school	2746	49.6
Student Gender
Male	2923	52.8
Female	2611	47.2
School Grade
Middle school	3326	60.1
High school	2208	39.9
Household Size
2	120	2.2
3	892	16.1
4	1845	33.3
5	1480	26.7
6+	1197	21.6
Housing Type
Apartment	3343	60.4
Single-family house	2083	37.6
Shanty	108	2.0
Parent Count
One	822	14.9
Two	4712	85.1
All Parents Employed
Yes	755	13.6
No	4779	86.4
Unemployment Mother
Yes	4634	83.7
No	900	16.3
Same-School Sibling
Yes	1182	21.4
No	4352	78.6
University Graduate Parent
Yes	822	14.9
No	4712	85.1
Car Ownership
0	3130	56.6
1	2260	40.8
2+	144	2.6
Household Income
<USD 500	220	4.0
USD 500–USD 999	1237	22.4
USD 1000–USD 1499	3225	58.3
USD 1500–USD 1999	656	11.9
>USD 2000	196	3.5

presents the descriptive statistics of the final dataset, which includes 2798 students from 2092 households living in the study area. In total, 5534 school trips were recorded, almost equally distributed between trips to school (50.4%) and from school (49.6%). Walking appears as the most used travel mode, with 55.3% of trips, followed by public transit (25.6%), school buses (14.6%), and private cars (4.5%). In Türkiye, school bus services are typically provided by private operators, and the monthly cost ranges from USD 100 to 150. A significant portion of walking trips occurred at very short or relatively long distances: 35.3% were shorter than 0.5 km, and 32.6% were longer than 2.5 km. Medium-range distances were less common, with 18.6% of trips between 0.5 and 1.49 km and 13.5% between 1.5 and 2.49 km. Trip distances were calculated by determining the shortest pedestrian path between each student’s home and school using a GIS-based network analysis. This analysis utilized the home and school locations obtained from the household travel survey, which was conducted as part of the Transport Master Plan, and the resulting shortest paths were accepted as the trip distances for this study. As shown in Figure 1, walking dominates at short distances: more than 90% of trips under 0.5 km and over 85% of those up to 1 km were made on foot. However, after 2 km, the walking rate drops significantly, falling below 40% and reaching around 15% at 4 km. These findings indicate that 2 km may represent a behavioral threshold for students’ willingness to walk.

The sample includes 52.8% male and 47.2% female students, with the majority (60.1%) attending middle school and the rest (39.9%) in high school. In terms of household size, families with four members are the most common (33.3%), followed by five-member (26.7%) and six-or-more-member (21.6%) households. Three-member households make up 16.1%, while only 2.2% of students live in two-person households.

Regarding housing conditions, 60.4% of students live in apartments, 37.6% in single-family houses, and 2.0% in shanty-type dwellings. A large proportion of students (85.1%) come from two-parent households. Only 13.6% of students reside in households where all parents are employed, and notably, 83.7% of mothers are not working. Additionally, just 14.9% of students have at least one parent with a university degree.

About 21.4% of students have siblings attending the same school. Car ownership is generally low; 56.6% of households have no vehicle, 40.8% own one car, and only 2.6% have two or more cars. The average monthly household income is USD 802. The majority of households (58.3%) fall within the USD 1000–1499 income range, followed by 22.4% in the USD 500–999 range. Only 4.0% of households earn less than USD 500, while 3.5% report an income higher than USD 2000.

4. Method of the Study

When individuals travel from one point to another, they tend to choose the mode that maximizes utility by evaluating factors such as cost, travel time, and other attributes (e.g., comfort, safety) associated with the available options [13]. The literature presents various approaches to modeling travel mode choices, with discrete choice models being a common method due to their closed-form expressions and interpretable results. Recently, studies have also incorporated machine learning and explainable AI to enhance their predictive and interpretative capabilities [89,90].

While multinomial logit (MNL) models are frequently used to evaluate multiple mode alternatives (see Table 1), their use is constrained by the Independence of Irrelevant Alternatives (IIA) assumption, which implies that unobserved factors are uncorrelated across choices [9]. To address this limitation and better capture unobserved heterogeneity, advanced models such as latent class and random parameter approaches have been adopted in some studies [91,92]. One advantage of the latent class model is that it identifies homogeneous subgroups without requiring assumptions on parameter distributions, unlike random parameter models that rely on specific distributional forms [93].

Building on this, our study adopted a two-step modeling approach to address unobserved heterogeneity in school travel behavior. First, latent class clustering (LCC) was utilized to group students with similar characteristics. Subsequently, a separate multinomial logit model was estimated for each identified class to pinpoint the factors influencing their school mode choices.

4.1. Latent Class Clustering

Latent class clustering (LCC) is a statistical method based on a probability model used to determine subgroups of datasets according to a latent categorical variable. LCC maximizes the homogeneity of the samples within classes and the heterogeneity between classes [94]. If a dataset with J categorical items is classified into N classes, the LCC model can be described by the following equation:

$$P\left(Y_i=y\right)=\sum _{n=1}^N{\gamma }_n\prod _{j=1}^J\prod _{r_j=1}^{R_j}{p}_{j,r_j│n}^{{I(}_{y_j=r_j})}$$

(1)

where $y_i$ represents item j of a response pattern y. $I(y_j=r_j)$ is an indicator function that equals 1 when the response variable $j=r_j$ and equals 0 otherwise. ${\gamma }_n$ is the probability of membership in latent class n, and $p_{j,r_j│n}^{{I(}_{y_j=r_j})}$ is the probability of response $r_j$ to item j, conditional on membership in latent class n. γ is a vector of latent class membership probabilities [95].

There is no requirement to specify the number of clusters in advance, as this can be statistically determined. Common criteria used for selecting the optimal number of classes include the Bayesian Information Criterion (BIC) and the Akaike Information Criterion (AIC). Lower values of AIC and BIC indicate a better model fit. Additionally, entropy is frequently employed to assess the quality of the clustering solution. Entropy values range from 0 to 1. Values closer to 1.0 indicate the significance of the model [96].

4.2. Multinomial Logit Model

For each cluster, MNL was utilized to investigate the significant factors influencing school trip mode choices among multiple modes as shown in Equation (2).

$$U_{in}=B_{\mathrm{İ}}{X}_{in}+{\epsilon }_{in}$$

(2)

where U_in determines the mode choice outcome i by individuals, X_in is a vector of observed characteristics (covariates), and ε_in is the random error term.

Table 4. Data’s variable specifications.

Dependent Variable	Levels
Mode	Walking *, Car, Public transit, School bus
Categorical Independent Variables	Levels
Student gender	Male, Female *
School grade	Middle school, High school *
Housing type	Shanty, Single-family house, Apartment *
Parent count	Two, One *
Trip direction	To school, From school *
Sibling in same school	No, Yes *
University graduate parent	No, Yes *
All parents employed	No, Yes *
Unemployed mother	No, Yes *
Car ownership	No, Yes *
Ordinal Independent Variables	Levels
Household income	<500$ *, 500$–999$, 1000$–1499$, 1500$–1999$, >2000$
Household size	2, 3, 4, 5, 6+
Same-school siblings	1, 2, 3, 4+
Number of cars	0, 1, 2+
Continuous Independent Variables
Student age
Trip distance

* Reference category in MNL model.

presents the variables included in the model, grouped by their type and listed with their respective levels and reference categories. The dependent variable is the choice of school trip mode, modeled across four alternatives: walking (reference), private car, public transit, and school bus. The explanatory variables comprise a mix of categorical, ordinal, and continuous measures. Categorical variables include student gender, school grade, housing type, parent count, trip direction (to or from school), presence of a sibling in the same school, parental education status e.g., whether a university graduate parent exists), and indicators of employment (e.g., whether all parents are employed, or the mother is unemployed). Ordinal variables reflect ordered categories such as household income, household size, number of same-school siblings, and number of cars in the household. The continuous variables, student age and home-to-school distance, were entered in their original metric form.

5. Results

Latent class cluster analysis (LCCA) was applied in the first stage to reduce preference heterogeneity within the sample. The number of clusters was determined using model fit criteria such as log-likelihood, the Bayesian Information Criterion (BIC), and entropy. The three-cluster solution, with a BIC score of 231,359 and a high entropy value of 0.991, was selected as the best-fitting model. Notably, the percentage decrease in BIC fell below 2% beyond the three-cluster model, indicating a diminishing improvement in model fit (see Figure 2).

Table 5. Overview of cluster characteristics.

	C1	C2	C3
Sample Size	812	1840	2882
Mode
Walking	44.6%	35.1%	71.2%
Car	10.9%	2.9%	3.7%
Public transport	22.7%	50.0%	10.9%
School service	21.8%	12.0%	14.2%
School grade
Middle school	54.7%	0.0%	100.0%
High school	45.3%	100.0%	0.0%
Student gender
Male	54.5%	54.4%	51.4%
Female	45.5%	45.7%	48.6%
Housing type
Shanty	2.2%	2.3%	1.7%
Single-family house	20.8%	37.2%	42.7%
Apartment	77.0%	60.5%	55.7%
Household size
2	5.6%	2.5%	1.0%
3	24.6%	18.0%	12.5%
4	40.3%	31.6%	32.5%
5	22.5%	26.1%	28.4%
6+	7.0%	21.8%	25.6%
Household income
<500 USD	13.4%	2.1%	2.5%
500–999 USD	5.1%	23.2%	26.7%
1000–1499 USD	40.5%	62.1%	60.8%
1500–1999 USD	28.4%	9.9%	8.5%
>2000 USD	12.7%	2.6%	1.6%
Parent count
One	66.7%	88.3%	88.4%
Two	33.3%	11.8%	11.6%
Unemployment mother
No	100.0%	2.3%	1.6%
Yes	0.0%	97.7%	98.4%
University graduate parent
No	61.7%	88.8%	89.4%
Yes	38.3%	11.2%	10.6%
Same-school sibling
No	82.6%	83.8%	74.2%
Yes	17.4%	16.2%	25.8%
Car ownership
No	40.0%	57.1%	60.9%
Yes	60.0%	42.9%	39.1%
Number of car
0	40.0%	57.1%	60.9%
1	51.3%	41.2%	37.6%
2+	8.7%	1.7%	1.5%
Student age
Mean	14.2	16.3	12.4
Trip distance (km)
Mean	2.9	3.6	1.4

presents an overview of the cluster characteristics. Cluster 1 is composed mostly of middle school students, with walking being the most common mode of travel (44.6%). A significant portion of households own a car (60.0%), and all mothers are employed. This cluster has the lowest household size. Household income generally falls within the middle range, and the proportion of university-educated parents is higher compared to the other clusters. Travel distances in this group range from short to medium.

Cluster 2 consists entirely of high school students, and public transportation stands out as the dominant travel mode (50.0%). Households in this cluster are larger than those in Cluster 1. Almost all of the mothers are unemployed. Income levels are concentrated in the low-to-middle range, and the share of university-educated parents is relatively low. On average, students in this cluster travel longer distances to school than those in the other groups.

Cluster 3 includes only middle school students; the vast majority of whom walk to school. This group has the lowest level of car ownership and the highest household size. Representing a low-income population, almost all mothers in this cluster are not working, and the proportion of university-educated parents is low. Students in this group have the shortest school travel distances among all clusters.

Multinomial logit (MNL) models, with walking mode set as the reference category, were estimated separately for each latent cluster. The results indicate that the variables influencing choice of school travel mode vary both within and across clusters. McFadden’s pseudo R-squared values for the models across all clusters range between 0.2 and 0.4, indicating an acceptable level of model fit [97] (see

Table 6. Multinomial logit models results by cluster.

Mode	Factors	C1			C2			C3
		β	Sig.	OR.	β	Sig.	OR.	β	Sig.	OR.
Car	Student gender = Male	−0.694	0.023	0.499	−1.174	0.038	0.309
	School grade = Middle school	1.389	0.020	4.010
	Trip distance	1.206	0.000	3.339	1.471	0.000	4.353	0.544	0.005	1.723
	Parent count = Two	−0.729	0.047	0.483	2.667	0.001	14.4
	Housing type = Single-family house	1.041	0.006	2.831	1.222	0.032	3.393
	Housing type = Shanty	22.652	0.000	6.88 × 109
	University graduate parent = No	−0.944	0.009	0.389
Public Transit	Student age							0.331	0	1.392
	Trip distance	1.317	0.000	3.731	1.405	0	4.075	1.031	0	2.804
	Parent count = Two	−1.071	0.000	0.343	1.383	0	3.985
	Household size				0.445	0.001	1.56
	Household income	0.296	0.029	1.344	−0.18	0.009	0.835
	Housing type = Single-family house				0.629	0.006	1.877
	Housing type = Shanty	21.296	0.000	1.77 × 109
	Number of cars				−17.775	0	1.91 × 10−8
School Bus	Student age	−0.373	0.001	0.689
	Trip distance	1.324	0.000	3.760	1.211	0	3.356	0.669	0	1.952
	Parent Count = Two	−0.704	0.019	0.495	1.22	0.006	3.387
	Household size	−0.598	0.001	0.550				−0.429	0.008	0.651
	Household income	0.368	0.005	1.445	−0.257	0.005	0.773
	Same school sibling = No							0.846	0.016	2.33
	University graduate parent = No							−1.135	0.011	0.322
	Car ownership = No				−3.099	0.014	0.045
	Household income * Number of cars				−1.219	0.025	0.296
Cox and Snell			0.596			0.502			0.303
Nagelkerke			0.648			0.572			0.476
McFadden			0.358			0.331			0.270

).

For Cluster 1, the likelihood of choosing a car, public transit, or school bus instead of walking changes with several factors. Male students are less likely to use a car compared to female students (OR = 0.499). Students in middle school are about four times more likely to use a car compared to those in high school (OR = 4.010). As the trip distance increases, the use of a car becomes more likely (OR = 3.339). Students living in a single-family house are more likely to travel by car (OR = 2.831). Having two parents (OR = 0.483) or having parents without a university degree (OR = 0.389) reduces the chance of using a car. Students living in a shanty have a much higher probability of using a car (OR ≈ 6.88 × 10⁹). For public transit, longer trip distances increase the likelihood of use (OR = 3.731), while having two parents lowers it (OR = 0.343). A higher household income increases the use of public transit (OR = 1.344). Students living in a shanty are also much more likely to use public transit (OR ≈ 1.77 × 10⁹). For school bus use, older students are less likely to choose it (OR = 0.689), while a greater distance increases the probability (OR = 3.760). Students with two parents (OR = 0.495) or from larger households (OR = 0.550) are less likely to use the school bus. A higher household income slightly increases the likelihood of school bus use (OR = 1.445).

In the second cluster (C2), the likelihood of choosing a car, public transit, or school bus instead of walking also varies with several factors. Male students are less likely to use a car compared to female students (OR = 0.309). Students living with both parents are much more likely to use a car (OR = 14.4). As the trip distance increases, the use of a car becomes more likely (OR = 4.353). Students living in a single-family house are more likely to travel by car (OR = 3.393). For public transit, longer trip distances increase the likelihood of use (OR = 4.075), and students living with both parents are also more likely to choose this mode (OR = 3.985). However, a higher household income slightly reduces the use of public transit (OR = 0.835). Students from larger households are more likely to use public transit (OR = 1.56), while an increase in the number of household cars greatly reduces the probability (OR ≈ 1.91 × 10⁻⁸). Students living in a single-family house are also more likely to use public transit (OR = 1.877). For school bus use, students living with both parents are more likely to choose this mode (OR = 3.387), and a greater distance again increases the probability (OR = 3.356). A higher household income slightly reduces the likelihood of school bus use (OR = 0.773). Students from larger households are less likely to use the school bus (OR = 0.651). The interaction between household income and number of cars also reduces the probability of choosing the school bus (OR = 0.296). Additionally, students from households without a car are much less likely to use the school bus (OR = 0.045).

In the third cluster (C3), the effects of most variables are weaker compared to the other clusters; however, trip distance still increases the likelihood of choosing all three motorized modes. Specifically, a greater distance raises the probability of using a private car (OR = 1.723), public transit (OR = 2.804), and school bus (OR = 1.952). As student age increases, the likelihood of using public transit also increases (OR = 1.392). School bus use is less likely among students whose parents have no university education (OR = 0.322). In contrast, students without a sibling attending the same school are more likely to use the school bus (OR = 2.33). Additionally, a larger household size slightly reduces the probability of choosing a school bus (OR = 0.651).

6. Discussions

The findings of this study show that school travel choices vary significantly among students, depending on individual, household, and environmental characteristics. Among all variables, trip distance emerges as the most consistent and influential factor across the three clusters. As the distance between home and school increases, students are more likely to shift from walking to motorized modes such as private car, public transit, or school bus. This pattern is especially clear in Cluster 2, where students travel longer distances and motorized modes are used more often. In contrast, students in Cluster 3 mostly walk to school, which can be explained by their short travel distances and very low levels of car ownership. In Cluster 1, walking is also relatively common, likely due to moderate travel distances and the younger age of the students. These findings are consistent with previous studies showing that distance is a key determinant in school travel behavior [8,25,26], with longer distances leading to a reduced use of active modes [27,28,29] and greater reliance on school buses [29], private cars [30], and public transit [2].

The effects of household structure vary between clusters. In Cluster 1, students living with both parents are less likely to use a private car, public transit, or school bus. This may be related to the fact that in this group, all the mothers are employed, and the students are in middle school; under these conditions, walking may be a more practical choice. On the other hand, in Cluster 2, where the mothers are mostly not working, and the students are in high school, living with both parents increases the likelihood of using all motorized modes. In such households, families may have a stronger influence on their children’s travel decisions and may encourage safer or more structured transportation options.

Household income also shows different effects depending on the cluster. In Cluster 1, a higher income increases the likelihood of using public transit and a school bus. This may reflect families’ preference for more secure and organized options. In Cluster 2, however, a higher income slightly reduces the use of these modes, possibly due to a tendency toward private car travel. In Cluster 3, income has no significant effect, likely because the group is more socioeconomically homogeneous, and income levels are generally low.

Some findings are specific to certain clusters. For example, in Cluster 3, students whose parents do not have a university degree are less likely to use the school bus. Also in this group, students without a sibling attending the same school are more likely to choose the school bus, which may reflect hesitation toward walking alone. In Cluster 2, a higher number of household cars drastically reduces public transit use, suggesting that private vehicles are a strong substitute. In both Clusters 1 and 2, living in a single-family house increases the probability of using a private car, but only in Cluster 2 is this associated with higher public transit use as well. Lastly, school bus use is associated with not having a sibling in the same school only in Cluster 3; this effect does not appear in the other groups.

These differences clearly show that student travel behavior cannot be explained by a single model. The results justify the use of a latent class approach in this study, as it allows for uncovering meaningful heterogeneity based on both social and spatial conditions. It is important to note that most previous studies have used single-level models based on the full sample, without identifying distinct clusters of students. These models generally assume that all students behave in a similar way. However, this study used a latent class approach, which helps to identify different groups of students based on their social and spatial characteristics. Thanks to this method, it was possible to see specific patterns in each group that would not be visible in a single model. For this reason, results from other studies that do not use this type of segmentation should be compared carefully, as their findings may reflect general averages rather than group-specific behaviors.

7. Conclusions

School trips represent a significant portion of urban travel, with walking being particularly crucial for sustainability and children’s health. Many studies emphasize the benefits of walking and highlight several key factors that influence how students travel to school. These include the distance between home and school, the quality of pedestrian and cycling infrastructure, and family characteristics such as income level, car ownership, and whether the parents are working. Additionally, the age and gender of the children are important, with older students and males more likely to choose active transportation modes like walking or biking. Understanding these factors is essential for developing strategies that promote healthier and more sustainable travel habits among students.

In the past few decades, students in Mersin, like other major cities in Türkiye, have become increasingly dependent on school buses and parental escorts. This shift has led to a decline in public transit use and a growing reliance on private cars. Although walking remains a common mode of travel for school trips in Mersin due to the city’s relatively low household income levels, this trend may change as income and car ownership rates rise in the future. Such changes could significantly alter students’ travel patterns, increasing the use of cars and private school buses. The current advantage of walking as a primary mode of transport may not last if school locations are not carefully planned and transportation facilities are not improved. In recent years, both public and private schools have chosen to build on the periphery of the city, where land costs are lower than in the city center. This trend reflects findings from previous studies, showing that families tend to prefer schools on the periphery over neighborhood-based locations, leading to longer travel distances and more motorized school trips. As a result, walking and public transport usage have decreased, leaving parents with limited options, like driving their children or relying on private school buses. This shift brings both monetary and non-monetary costs for families and society, including fuel expenses, time costs, air pollution, and increased accident risks. Additionally, students’ health and well-being may suffer. Policymakers should consider the long-term social impacts of school location decisions, not just the short-term investment costs.

While this study provides valuable insights into the factors affecting school travel mode choices, it is important to acknowledge some limitations. This study mainly focused on middle and high school students, so younger children and university students, who might have different travel behaviors, were not included. Also, the research was based on cross-sectional data, meaning it only looked at travel patterns at one point in time, not showing how these patterns might change over time. Moreover, even though key variables were considered, other factors like weather conditions, crime statistics, family perceptions about safety and convenience, and school policies were not included but could affect transportation choices. This research obtained data from a conventional home interview survey; therefore, the questions were limited to household profile, traveler profile, and trip-making behaviors. Future studies could include more demographic groups and variables and explore how facility location and land use planning affect school travel behavior.

It has been shown in the literature that the walkability of the physical environment has an important impact on the share of walking trips. In this study, it was not possible to calculate a walkability index covering the entire city due to data limitations. Therefore, the effects of physical factors such as sidewalk quality, land use diversity, or street connectivity on walking trip shares at the neighborhood level could not be evaluated. Previous studies have also pointed out that variables like housing density, perceived safety, and environmental quality play a role in shaping walkability. In our case, the dataset did not include information about these variables, and it was not possible to access reliable secondary data either. For example, data related to weather conditions, crime rates, or parental perceptions of safety were not available and thus could not be included in the analysis. It is recommended that future studies consider incorporating such factors, by using more detailed surveys or combining different data sources. In this way, more comprehensive modeling approaches, including microscopic or multi-level methods, could be applied to better understand the determinants of choices of school travel mode and to help develop planning strategies that support sustainable mobility.