Classification of Unrealized Trips and Identification of Mobility Difficulties: Urban–Rural Differences in Japan

Abstract

For various reasons, individuals sometimes cannot go on trips. These are known as unrealized trips. In this study, we focus on discretionary trips, such as those for shopping and leisure, and analyze the factors that lead to their lack of realization due to insufficient mobility and transportation services. Based on survey data collected in Japan from populations of various sizes, K-prototypes clustering was conducted, and respondents were classified into three groups: those who only travel by private car, those with multiple transportation options, and transportation-disadvantaged individuals. Notably, the transportation-disadvantaged group exhibited a higher frequency of unrealized trips and was present in both urban and rural areas. Estimation results of the zero-inflated bivariate ordered probit (ZIBOP) model revealed that younger and lower-income individuals were more likely to experience unrealized trips. Moreover, railway frequency and last service time of bus were significant factors in urban areas, while bus service levels played a key role in rural areas. These findings suggest that “transportation-disadvantaged” individuals should be defined not only as those without a car or driver’s license but also as those who lack alternative mobility options.

1. Introduction

In daily life, desired trips are widely realized. However, some are not carried out due to various reasons. The occurrence of these unrealized travel demands contributes to social exclusion and negatively affects psychological well-being [1]. For instance, households lacking transportation means often face more severe restrictions on activities and report lower levels of mental health [2].

Public transportation services and related policies are typically planned based on observed (realized) travel patterns, while latent travel demand is almost entirely ignored [3]. Latent demand can be categorized into six levels [4], ranging from Realized Activity and Travel Patterns (Level 1) to Unimagined Activities (Level 6), in which individuals are not even aware of their latent desire to travel. Among Levels 2 to 5, a key boundary lies between Levels 3 and 4: whether a trip was planned or not. If a trip is planned but not realized, the cause is often an external constraint, particularly one related to transportation, making this type of unrealized demand a key target for transportation policy intervention. In contrast, if a trip is desired but abandoned before planning, individual-level factors often play a greater role. These include sociodemographic characteristics (e.g., age), individual ability (e.g., holding a driver’s license, health status), and household context (e.g., household size).

More than half of existing studies have attempted to analyze the relationship between unrealized trips and transportation constraints, primarily focusing on trips that were planned but not realized [5]. However, many of these also include individual-level factors as explanatory variables, effectively analyzing cases where trips were never even attempted, i.e., abandoned desires. This complicates interpretation, as such situations may extend beyond what transportation policy can effectively address.

To mitigate this challenge, previous studies have narrowed their sample populations to those with specific attributes to simplify interpretation. The most common target group is older adults. Across these studies, two factors are consistently associated with a higher frequency of unrealized trips: health status and access to private vehicles (vehicle ownership or possession of a driver’s license) [3,6,7,8,9,10,11,12,13,14]. Other significant factors which increase the number of unrealized trips are the convenience of public transport [9,10,12,15], low income [7,9,12], living alone in rural areas [16], and living with children [7]. However, recent studies suggest that mobility services via smartphones [17] and the future availability of autonomous vehicles [18] have the potential to reduce unrealized travel needs.

In addition to older adults, existing studies have also focused on other demographic groups such as people with disabilities [15,19,20,21], women [22,23], and students or younger adults [24,25]. A survey conducted in New Jersey targeting people with disabilities found that the proportion of respondents dissatisfied with public transportation services around their homes was 47%. The main reasons for dissatisfaction were “services not being available when needed” (23%) and “accessible service not being available near home” (20%), indicating that accessibility issues were more significant than cost [19]. In Hennepin County, Minnesota, a study focusing on developmental disabilities revealed that 46% reported unrealized necessary trips and that the presence of public transportation (bus stops) in the neighborhood reduced the likelihood of unrealized trips [20].

Regarding women, it has been found that their daily travel is often restricted by harassment and safety concerns [22], while long-distance travel is constrained by household and work responsibilities, with many also reporting anxiety about leaving home [23].

For students, a study of seven university campuses in Toronto analyzed, using a multivariate logistic regression model, the factors discouraging participation in campus activities [24]. The results showed that students with longer commutes to campus or those who were employed participated less frequently. Another study, also in Toronto, investigated changes in the travel behavior of young adults collected in 2019 and 2022 [25]. Focusing on two life events, entering employment after completing education and residential relocation, it found that individuals who shifted from active modes to public transportation after moving tended to participate in fewer activities and experience transport-related social exclusion.

In summary, these groups share similar challenges with older adults, but they also tend to drive less frequently, making them more sensitive to factors such as the accessibility and cost of public transportation and the safety of walking environments.

While studies consistently show a strong relationship between transportation issues and unrealized trips across all demographic groups, relatively few have focused exclusively on transport-related constraints. Although there are clearly differences in service levels between urban and rural areas, comparative studies are limited and mostly conducted in North American contexts [16,26,27].

Japan, the focus of this study, is facing both a rapidly aging population and population decline while maintaining a high urbanization rate of over 90% [28]. These demographic trends are accelerating the closure of rural railway and bus lines [29], further widening the service gap in public transportation between urban and rural areas. Although it is already clear that the lack of transportation services in rural areas, where older adults are more concentrated, contributes significantly to unrealized trips [30], evaluating the legitimacy of rural service downsizing requires a detailed comparison between urban and rural regions to determine the extent to which service level disparities impact the frequency of unrealized trips.

Turning to trip purposes, studies about healthcare-related trips show that lack of access to transportation means, especially private vehicles, is a major reason for cancelations [31,32,33,34,35,36]. A survey conducted among low-income immigrants in suburban New York City found that about one-quarter of the participants reported transportation problems as a reason for missing or rescheduling clinic appointments [32]. Moreover, a study using data from the U.S. National Health Interview Survey (1997–2017) reported that in 2017, 5.8 million Americans (1.8%) delayed medical care each year due to a lack of available transportation [33]. In addition, when access by private car is restricted, patients with cancer may face more serious medical challenges, such as interruptions in radiation therapy [34].

However, at least in Japan, transportation constraints often have less impact on participation in mandatory activities such as commuting or medical visits, as they tend to involve institutional or social support [37]. In contrast, discretionary trips, such as those for shopping and leisure (e.g., sport, vacation, hobbies outside the home) are less likely to be supported and can be easily canceled. The cancelation of these discretionary trips raises serious concerns about increased social exclusion and psychological distress [1,38,39], because shopping provides opportunities for social connection and interaction within the community [40], and leisure activities help alleviate stress and fatigue, promote physical and mental recovery, and foster relationships with family and friends [41,42,43,44].

Based on the above background, we summarize research gaps below:

• Most previous studies focus on specific groups such as older adults. However, inconvenient public transportation can lead to unrealized trips even for those without mobility impairments. Therefore, it is inappropriate to restrict analyzing transportation constraints to specific groups. Furthermore, there is a need to differentiate between individuals who “planned a trip but could not realize” it and those who “gave up on the trip entirely” and to construct models that identify factors specific to the former group.
• Many studies are limited to either rural areas or specific urban regions, and few comparisons between urban and rural areas exist. Since public transportation plays different roles in urban and rural settings, separate models for these two areas are necessary to compare their respective impacts on unrealized trips.
• The existing literature lacks sufficient insight into discretionary trips, such as those for shopping and leisure activities. Although some studies focus solely on the latter [45,46,47], shopping and leisure trips are often correlated in terms of frequency and behavior. Thus, they should be jointly evaluated within a single model.

Considering these research gaps, we aim to reveal how the lack of transportation options contributes to the frequency of unrealized shopping and leisure trips, as well as to identify the extent to which transport-related variables affect this outcome. A wide-area survey was conducted in Japan, and K-prototypes clustering was used to identify groups that consistently experience a higher number of unrealized trips due to limited transport access. Individuals who do not have problems with transportation access, or who have already “given up” on making trips for personal reasons, are likely to report zero unrealized trips due to transport constraints. Therefore, zero values were expected to be dominant in the data. To handle this, we applied a zero-inflated bivariate ordered probit (ZIBOP) model separately to respondents living in urban and rural areas.

2. Materials and Methods

Using the online survey data described in this section, this study analyzes unrealized shopping and leisure trips through the process outlined in Figure 1. First, K-prototypes clustering is conducted to identify the group of respondents with the highest average frequency of unrealized trips. Respondents in this cluster are then further divided into urban and rural subgroups, to which the ZIBOP model is applied. K-prototypes is a relatively simple method, and the application of the ZIBOP model also follows existing literature [48]. However, by comparing the classified clusters in terms of unrealized trip counts and then applying the ZIBOP model specifically to the cluster with the highest number of unrealized trips, the combination of these two methods forms a coherent analytical set. This integrated approach can be regarded as a unique contribution of this study.

2.1. Survey Outline

This study uses online survey data that includes individuals’ access to transportation from home and their engagement in daily shopping and leisure activities. The survey was conducted between 13 and 19 September 2023, targeting individuals aged 18 and older. The survey area comprised seven prefectures in Japan’s Chubu and Hokuriku regions (Niigata, Toyama, Ishikawa, Gifu, Shizuoka, Aichi, and Mie), as illustrated in Figure 2. These seven prefectures fall under the jurisdiction of two Regional Development Bureaus (the Chubu Regional Development Bureau and the Hokuriku Regional Development Bureau) of the Ministry of Land, Infrastructure, Transport and Tourism (MLIT) [49]. These regions collectively include 224 municipalities, ranging from large metropolitan areas such as Nagoya, Niigata, Shizuoka, and Hamamatsu City to small municipalities with fewer than 20,000 residents. This diversity makes the dataset suitable for comparison across different population scales. The reason for selecting this region is that it allows for an easy comparison of rail accessibility. In Japan’s major metropolitan areas, Tokyo and Osaka, the railway networks are highly developed, providing high accessibility even to suburban area. In contrast, in this region, car dependency is relatively high, and only Nagoya and its surrounding areas can be assumed to offer convenient living without a car.

To ensure sample balance, quotas were set based on region (Figure 2a), municipality population size (Figure 2b), age group (18–29, 30–49, 50–64, and over 65 years old), and gender (male or female). Sampling was conducted to maintain proportional representation across these categories. In total, valid responses were collected from 5976 individuals. The criteria for population classification are based on the municipal classifications defined by the Local Autonomy Act [50]. Classes 1 and 2 are designated as “ordinance-designated cities”, which are delegated significant administrative authority from the prefectural government and can independently carry out urban planning and related functions. Cities in Class 3 are eligible to become “core cities”, serving as regional hubs with partial authority, similarly to ordinance-designated cities. Class 4 is defined as “other cities,” while Classes 5 and 6 represent “towns and villages”, where urban planning and other tasks are managed by the prefectural government.

The survey consisted of items regarding individual and household characteristics, service levels of public transportation from the respondent’s home, and the number of realized and unrealized trips for shopping and leisure made during the last month. Shopping trips were defined as outings to purchase essential goods such as groceries and daily necessities. Leisure trips referred to outside activities involving hobbies, social interactions, or recreation. Respondents were explicitly instructed to answer questions regarding public transportation, shopping, and leisure trips based on their own perceptions. To focus on transportation-related constraints, the following phrase was used to ask about unrealized trips: “In the last month, how often did you intend to go shopping but could not do so due to lack of available transportation?” The same question was posed for leisure.

2.2. Descriptive Statistics

Table 1. Descriptive statistics (N = 5976).

Variable	Percentage (%)	Mean	SD
Female	48.8
Age		48.62	17.04
Having a driving license	90.3
Occupation
Employee	49.3
Part-time	15.2
Housemaker	13.8
Student	4.1
Unemployed	16.4
Others	1.0
Household size		2.75	1.34
Annual household income (million JPY)		562.58	402.49
Number of private cars in the household		1.56	1.02
Distance to the closest station (km)		2.76	3.61
Distance to the closest bus stop (km)		1.07	2.21
Mean rail frequency at the closest station in peak hours (rail/h)		4.55	4.38
Mean bus frequency at the closest station in peak hours (bus/h)		2.75	2.92
Time of final train at the closest station (hour)		22.32	1.96
Time of final bus at the closest bus stop (hour)		20.76	2.07
Distance to the primary shopping destination (km)		2.92	3.98
Distance to the primary leisure destination (km)		7.32	7.13

provides a summary of the descriptive statistics. The average age of respondents was approximately 48 years, and more than 90% held a driver’s license. Approximately 50% were employed, while 13.8% were full-time homemakers, 4.1% were students, and 16.4% were unemployed. The average household size was 2.75 people, whereas the average number of private vehicles owned was 1.56, suggesting that many households shared a single vehicle among multiple members.

Regarding access to public transport, the average distance to the nearest railway station was 2.762 km, which was approximately 1.7 km farther than the nearest bus stop. However, the average number of services per day and the latest available departure time indicated that rail services were of a higher quality than bus services. With respect to trip destinations, the average distance to frequently visited leisure locations was over 7 km, which was more than twice as far as that to shopping locations.

Figure 3 shows the distribution of trip frequencies for shopping and leisure by the primary mode of transportation. The primary mode is defined as that most frequently used to travel to the most frequently visited locations. For shopping, those who used private vehicles (either self-driven or driven by others) tended to make trips 1 to 4 times per week. In contrast, individuals using rail or bus services reported a higher frequency of shopping trips, with many traveling 5 or more times per week. For leisure trips, over 60% of respondents reported traveling less than once per week regardless of the mode of transportation. Differences across transport modes were smaller for leisure, but bus users exhibited a relatively higher frequency of 5 or more trips per week.

Figure 4 presents the distribution of unrealized shopping and leisure trips by primary mode of transportation. For all purposes and modes, over half of the respondents reported zero unrealized trips per month. Those who drove themselves had notably lower frequencies of unrealized trips, while those who relied on public transport or rides from others experienced unrealized trips more frequently. This suggests that individuals who reside beyond walking distance to destinations and who do not own a private vehicle are particularly vulnerable to unrealized trips and thus merit closer analysis.

Table 2. Cross-tabulation of unrealized shopping and leisure trips (N = 5976).

Frequency per Month [N = 5976]		Leisure
		0	1	2	3	4	5 or More
Shopping	0	4418	104	30	12	12	10
	1	107	118	32	7	4	11
	2	89	66	97	22	15	20
	3	53	23	61	66	33	30
	4	29	12	31	36	50	21
	5 or more	57	21	36	58	52	133

shows a cross-tabulation of unrealized shopping and leisure trip frequencies for all samples. The data indicates a high degree of similarity between these two types of trips. For example, among individuals with two unrealized shopping trips per month, the most common corresponding value for leisure was also two. These results justify the use of a bivariate model that explicitly accounts for correlation between trip purposes.

2.3. K-Prototypes Clustering

K-prototypes clustering [51] is a non-hierarchical clustering method suitable for mixed scale data with numerical and categorical variables that partitions data into a predefined number of clusters, k. The dissimilarity between an observation and a cluster prototype is defined as the squared Euclidean distance for numerical variables and a simple matching measure for categorical variables, balanced by a weighting parameter $\gamma$. Similarly to K-means, the algorithm iteratively assigns observations to clusters and updates the prototypes until the total within-cluster dissimilarity is minimized. The clustering was implemented using R version 4.3.3, and the explanatory variables in this study were as follows:

• Possession of a driver’s license (binary);
• The number of private cars in the household divided by the number of adults (aged 18+) in the household;
• Distance to the closest station (km);
• Distance to the closest bus stop (km);
• Mean rail frequency at the closest station in peak hours (rail/h);
• Mean bus frequency at the closest bus stop in peak hours (bus/h);
• Time of final train at the closest station (hour);
• Time of final bus at the closest bus stop (hour);
• Distance to the primary shopping destination (km);
• Distance to the primary leisure destination (km).

Due to inconsistencies between the reported number of cars and the number of adults in 63 responses, the clustering analysis was performed using data from 5913 valid respondents. Numerical variables are applied after standardization. Furthermore, because the K-prototypes algorithm depends on the selection of initial centroids, choosing them only once at random increases the risk of converging to a local optimum. In this study, we mitigated this risk by randomly selecting the initial centroids 50 times and adopting the solution with the minimum within-cluster sum of squares (WSS), thereby enhancing the stability of the clustering results.

2.4. Zero-Inflated Bivariate Ordered Probit (ZIBOP) Model

The ZIBOP model consists of zero-inflated (ZI) part and bivariate ordered probit (BOP) part [48,52]. The two observed ordinal outcomes for individual $i$ (=1, …, N) are $y_{1i}$ and $y_{2i}$. The both-zero outcome is defined as inflation at the points ($y_{1i}$, $y_{2i}$) = (0,0). Both–zero outcome means that the number of unrealized trips is zero for shopping and leisure activities. The probability of both-zero outcome is represented by a binomial probit model as given below:

$$p_{zero}\left(s_i=1\right)=1-p_{zero}\left(s_i=0\right)=\mathsf{\Phi }\left({\displaystyle \frac{{\mathit{z}}_{\mathit{i}}’\mathit{\gamma }}{\sigma }}\right)$$

(1)

where $s_i$ denotes an indicator equal to 1 if the individual never experiences unrealized trips for both shopping and leisure, and 0 otherwise. Moreover, $\mathsf{\Phi }(·)$ represents the cumulative distribution function (CDF) of the standard normal distribution, ${\mathit{z}}_{\mathit{i}}’$ denotes a covariate vector, and $\mathit{\gamma }$ is the unknown parameter vector. $\sigma$ is the standard deviation of the error term, which is fixed at one.

Consequently, the ZIBOP model can be written as given below:

$$\mathrm{P}\mathrm{r}(y_{1i}=j,y_{2i}=k)=\left\{\begin{array}{ll}{p}_{zero}\left(s_i=1\right)+p_{zero}\left(s_i=0\right)p_{freq}\left(\left.y_{1i}=0,y_{2i}=0\right|s_i=0\right),& for \left(j,k\right)=\left(0,0\right)\\ p_{zero}\left(s_i=0\right)p_{freq}\left(\left.y_{1i}=j,y_{2i}=k\right|s_i=0\right),& for \left(j,k\right)\ne \left(0,0\right)\end{array}\right.$$

(2)

For j = 0, 1, …, J and k = 0, 1, …, K, $p_{freq}$ represents the probability of frequency of unrealized trips, captured by BOP model on the basis of the CDF of the standard bivariate normal distribution, ${\mathsf{\Phi }}_2(·,·)$, as given below.

$$\begin{array}{ll}{p}_{freq}(y_{1i}=j,y_{2i}& =k|s_i=0)\\ & ={\mathsf{\Phi }}_2\left({\displaystyle \frac{{\alpha }_{1j+1}-{\mathit{x}}_{1i}^{’}{\mathit{\beta }}_1}{{\sigma }_1}},{\displaystyle \frac{{\alpha }_{2k+1}-{\mathit{x}}_{2i}^{’}{\mathit{\beta }}_2}{{\sigma }_2}},{\rho }_{12}\right)-{\mathsf{\Phi }}_2\left({\displaystyle \frac{{\alpha }_{1j}-{\mathit{x}}_{1i}^{’}{\mathit{\beta }}_1}{{\sigma }_1}},{\displaystyle \frac{{\alpha }_{2k+1}-{\mathit{x}}_{2i}^{’}{\mathit{\beta }}_2}{{\sigma }_2}},{\rho }_{12}\right)\\ & -{\mathsf{\Phi }}_2\left({\displaystyle \frac{{\alpha }_{1j+1}-{\mathit{x}}_{1i}^{’}{\mathit{\beta }}_1}{{\sigma }_1}},{\displaystyle \frac{{\alpha }_{2k}-{\mathit{x}}_{2i}^{’}{\mathit{\beta }}_2}{{\sigma }_2}},{\rho }_{12}\right)+{\mathsf{\Phi }}_2\left({\displaystyle \frac{{\alpha }_{1j}-{\mathit{x}}_{1i}^{’}{\mathit{\beta }}_1}{{\sigma }_1}},{\displaystyle \frac{{\alpha }_{2k}-{\mathit{x}}_{2i}^{’}{\mathit{\beta }}_2}{{\sigma }_2}},{\rho }_{12}\right)\end{array}$$

(3)

For individual $i$, ${\mathit{x}}_{1i}^{’}$ and ${\mathit{x}}_{2i}^{’}$ stand for explanatory variable vectors, ${\mathit{\beta }}_1$ and ${\mathit{\beta }}_2$ are unknown parameter vectors, and ${\alpha }_{1j}$ and ${\alpha }_{2k}$ are threshold parameters. Additionally, ${\alpha }_{10}$ and ${\alpha }_{20}$ are $-\infty$ for j and k, ${\alpha }_{1J+1}$ and ${\alpha }_{2K+1}$ are $\infty$. ${\rho }_{12}$ is the correlation coefficient between the two error terms, and ${\sigma }_1$ and ${\sigma }_2$ are standard deviations of the error terms and are both set to one.

The likelihood function for estimation can then be formulated as given below:

$$\begin{array}{c}L={\displaystyle \prod _{i=1}^N}\left[{\left\{p_{zero}\left(s_i=1\right)+p_{zero}\left(s_i=0\right)p_{freq}\left(\left.y_{1i}=0,y_{2i}=0\right|s_i=0\right)\right\}}^{d_{ijk}}\right.\\ \left.{\left\{p_{zero}\left(s_i=0\right)p_{freq}\left(\left.y_{1i}=j,y_{2i}=k\right|s_i=0\right)\right\}}^{1-d_{ijk}}\right]\end{array}$$

(4)

where $d_{ijk}$ = 1 if both $y_{1i}$ and $y_{2i}$ are zero, and $d_{ijk}$ = 0 otherwise.

3. Results

3.1. K-Prototypes Clustering

To determine the optimal number of clusters, we applied the elbow method and silhouette method for values of k ranging from 2 to 8. In the elbow method, the reduction in the WSS did not show a distinct inflection point as the number of clusters increased (Figure S1). This made it difficult to identify the optimal k based solely on the standard elbow criterion. The silhouette method recorded the highest silhouette score at k = 8 (Figure S2). However, since ten explanatory variables are used, high numbers of clusters raise concerns of overfitting and make the interpretation more difficult. The purpose of clustering in this study is not to produce fine-grained classifications based on travel mode or access distance, but rather to distinguish between transportation-disadvantaged and other groups. From this perspective, among the smaller numbers of clusters, k = 3 was judged to be the most appropriate. This is because the differences between cluster characteristics are clear, k = 3 achieves a sufficiently high silhouette score, and in the cases of k = 4 to 6, there was still only one transportation-disadvantaged cluster, which had a sample size similar to that in the k = 3 solution. This can be seen by the plot of the first and second principal components from the principal component analysis (PCA) of the explanatory variables (Figure S3).

The weight parameter $\gamma$ is typically set by dividing the average variance of all numerical variables by the average dissimilarity of the categorical variables (In this case, $\gamma =5.82$). However, in this setting, the license variable which is the only categorical variable has relatively low importance (Table S1). This fact can be seen from the result of K-means without license is similar (Table S2). This is possibly due to having the only categorical variable among ten explanatory variables. To appropriately account for the license’s importance, we used a value of $\gamma$ doubled ($\gamma =11.64$).

Table 3. Statistics of each cluster.

	Cluster 1: Private Cars Only		Cluster 2: Both Private Cars and PT		Cluster 3: Transportation-Disadvantaged
Samples	2321		3033		559
Variables	Mean	SD	Mean	SD	Mean	SD
Driving license (1 = yes; 0 = no)	1.00	0.03	1.00	0.00	0.00	0.00
Number of private cars in the household/ Number of adults (aged 18+) in the household	0.83	0.36	0.62	0.35	0.32	0.35
Distance to the closest railway station (km)	4.11	5.14	1.92	2.08	2.24	3.13
Distance to the closest bus stop (km)	1.62	3.43	0.71	1.09	0.87	1.86
Mean rail frequency at the closest station in peak hours (rail/h)	2.21	1.80	6.26	4.77	5.21	5.24
Mean bus frequency at the closest station in peak hours (bus/h)	1.26	1.17	3.83	3.28	3.08	3.16
Time of final train at the closest station (hour)	21.3	2.29	23.2	0.86	22.1	2.28
Time of final bus at the closest bus stop (hour)	19.2	1.89	21.9	1.20	20.9	2.23
Distance to the primary shopping destination (km)	4.65	5.28	1.72	1.81	2.33	3.70
Distance to the primary leisure destination (km)	11.1	8.80	5.70	3.70	5.72	7.22
Unrealized shopping trips per month	0.88	2.44	0.73	3.22	1.42	3.17
Unrealized leisure trips per month	0.65	1.87	0.54	2.81	1.00	2.77

summarizes the descriptive statistics for the three clusters. Cluster 1 consists of respondents who hold a driver’s license and, on average, have the highest number of private vehicles per adult household member. However, they have the poorest access to public transportation in terms of distance to the nearest station or bus stop, service frequency, and the latest available departure times. The distances to frequently visited shopping and leisure destinations are also the longest. This cluster can be described as “private cars only.”

Cluster 2, which includes the largest proportion of respondents (51.3%), has the lowest average number of unrealized trips. Members of this group hold driver’s licenses and enjoy relatively good access to both private cars and public transportation, as well as shorter distances to destinations. Thus, Cluster 2 is characterized as “both private cars and public transport (PT).”

Cluster 3 consists entirely of respondents who do not hold a driver’s license and have the lowest number of vehicles per adult household member. These individuals are more likely to depend on others for rides or to use alternative transportation modes, and they report the highest average number of unrealized trips. Although their access to public transport and destinations is not as poor as that of Cluster 1, the relatively high standard deviations, particularly in service frequency and latest departure times, suggest a variability in public transport accessibility even within this group. This cluster is considered the “transportation-disadvantaged” group.

To better understand the travel behaviors of and transportation challenges for each group, we analyzed the cluster distribution by municipality population size (Figure 5). As population size increases, the proportion of Cluster 2 also increases, while Cluster 1 becomes less dominant. This naturally reflects the fact that larger cities tend to offer more convenient public transport and a greater number of nearby facilities. However, the share of Cluster 3, the “transportation-disadvantaged” group, is also higher in more populated areas.

3.2. Zero-Inflated Bivariate Ordered Probit (ZIBOP) Model

Cluster 3, the “transportation-disadvantaged” group, reported the highest number of unrealized trips and demonstrated variability in transportation service levels, necessitating more detailed analysis. Therefore, we divided this group into two subgroups based on their place of residence: urban residents and rural residents (those living in municipalities with populations of 200,000 or more and of less than 200,000, respectively). We then applied the ZIBOP model to each subgroup (

Table 4. Estimation results of ZIBOP model (Cluster 3: “transportation-disadvantaged”).

Zero-Inflated Component (Binary Probit Model)	Urban				Rural
Variable	Parameter				Parameter
Constant	0.696		**		0.267
Gender = female					−0.515		*
Young (≤29 years old)	−1.125		**		−0.843		**
Occupation = unemployed					0.710		**
Occupation = part-time worker	−0.318
Occupation = housemaker	−0.531		(*)
Household size ≥ 3	−0.434		(*)
Annual household income ≤ JPY 2.5 million	−0.550		*		−0.261
Bivariate ordered probit component
Variable	Shopping		Leisure		Shopping		Leisure
Mean rail frequency at peak hours (rail/h)	−0.038	*	−0.057	**
Mean rail frequency at peak hours ≥ 4 rail/h					−0.237
Mean bus frequency at peak hours (bus/h)							−0.072	*
Final railway service time ≤ 7 pm							0.514	**
Final bus service time (h)					−0.077	**
Final bus service time ≤ 7 pm	0.778	**	0.648	*
Distance to the primary destination ≤ 0.5 km			−0.827	**	−0.833	**	−0.869	*
Threshold 1 (0–1/month)	−0.246		−0.174		−2.549	**	−0.483	*
Threshold 2 (1–2/month)	0.092	**	0.173	**	−2.114	**	−0.030	**
Threshold 3 (2–3/month)	0.511	**	0.545	**	−1.715	**	0.429	**
Threshold 4 (3–4/month)	0.836	**	0.841	**	−1.216	**	0.809	**
Threshold 5 (4–5 or more/month)	1.094	**	1.120	**	−0.853	**	1.048	*
Correlation	0.684 **				0.661 **
Sample size	349				210
Initial log-likelihood	−1492.6				−898.1
Final log-likelihood	−591.3				−445.0
Rho-squared	0.604				0.504
Adjusted rho-squared	0.589				0.480
AIC	1226.7				934.0
BIC	1311.5				1007.7

(*) p < 0.1; * p < 0.05; ** p < 0.01.

). This classification is based on the populations of municipalities required for designation as “core cities,” which are regional hub cities under Japan’s Local Autonomy Act. These cities have broader administrative authority and a regional hub function and tend to have higher levels of public transport infrastructure compared with other municipalities.

Furthermore, the clustering results support this classification. In Figure 5, the result showed that in cities with populations of 200,000 or more, the proportion of “Cluster 1: private cars only” is lower than that of “Cluster 2: both private cars and PT”, whereas in municipalities with populations below 200,000, the opposite pattern is observed. These results suggest that the threshold produces meaningful differences in transport accessibility, thereby supporting the validity of this classification.

In the ZI component, younger individuals in both urban and rural models are more likely to experience unrealized trips. In the urban model, individuals who are homemakers, live in households of three or more members, or belong to lower income groups tend to report more unrealized trips. However, in the rural model, women are more likely to report unrealized trips, while unemployed individuals are less likely.

Regarding the BOP component, multiple variables related to transportation service levels are statistically significant in both areas. Overall, a decline in service level contributes to an increase in the number of unrealized trips. In the urban model, a higher frequency of railway services at the nearest station is associated with a reduction in unrealized trips. Conversely, if the last bus departs before 7:00 p.m., this early cutoff significantly increases the frequency of unrealized trips. Additionally, for leisure trips specifically, unrealized trips are less likely to occur when the destination is within 500 m.

In the rural model, different variables are found to be significant compared to those in the urban model. For shopping purposes, a later final bus service time reduces the frequency of unrealized trips. For leisure purposes, a greater number of bus services is associated with fewer unrealized trips, while an early last train service time (before 7:00 p.m.) increases the frequency of these trips. Furthermore, differences in distance to the destination have a significant impact, with 0.5 km serving as a critical threshold.

4. Discussion

The results of K-prototypes reveal that individuals’ modes of transportation can be clearly categorized into three types based on transportation access and driver’s license ownership. Notably, Cluster 2, “both private cars and public transport”, reports the lowest number of unrealized trips. This finding aligns with previous research (focused primarily on older adults) showing that having multiple mobility options is directly linked to higher trip realization rates [9]. This highlights the importance of mobility policy developing a transportation environment that does not depend on single mode.

The results in Figure 5 indicate a trend in which the proportion of “transportation-disadvantaged” individuals (Cluster 3) increases in more populous areas. This suggests that urban areas tend to have more residents without driver’s licenses. Some studies have shown that license ownership significantly affects trip realization [3,6,7,8,9,10,11,12,13,14], and these results emphasize that individuals can still become transportation-disadvantaged even while living in urban areas. This is especially true in the urban regions in this study, which have a higher share of private cars compared to Japan’s major metropolitan areas (e.g., Tokyo, Osaka). In these regions, residents living in public transport-deprived areas tend to experience unrealized trips.

The results of the ZIBOP model indicate that, across both urban and rural contexts, younger individuals are more likely to experience unrealized trips, an aspect that has not been widely discussed in previous research. When comparing realized trip frequency between younger individuals and older adults (65+), shopping trips are more frequently realized among older adults (the share making shopping trips three times or more per week is 40.7% for younger people vs. 48.2% for older adults), while leisure trips are more frequently realized among younger people (the share making leisure trips once or more per week is 39.2% vs. 26.9%). From this, it can be inferred that older adults prioritize shopping as a routine activity. In other words, they may primarily visit a limited number of nearby stores within walking distance, which helps reduce unrealized trips. In contrast, some young individuals may treat shopping as a selective and irregular activity, choosing stores that meet their conditions from a wide range of options. For example, previous studies have shown that younger individuals tend to seek trends and variety in clothing purchases [53]. Moreover, recent research in Japan has revealed that younger adults place greater emphasis on product assortment, brand, and affordability [54], which limits the range of stores and products that meet their preferences. As a result, their trips are more easily hindered by the inconvenience of public transportation or longer travel distances. However, since this study did not investigate the number or diversity of stores that respondents intended to visit, testing this hypothesis remains a task for future research.

Regarding leisure, the higher realized frequency among younger people suggests that they also more often “want to go” and “plan to go,” which leads to a higher number of unrealized trips. Conversely, older adults have relatively lower demand for leisure activities and may rarely plan such trips in the first place, resulting in fewer unrealized trips.

Previous studies have shown that younger generations prefer to live in central areas with good accessibility and frequently use public transport, walk, or cycle [55,56]. This supports the hypothesis that unrealized trips are more frequent among younger people precisely because they seek both convenience and diversity in shopping and leisure opportunities. Moreover, living in areas with poor accessibility is particularly disadvantageous for younger people, who are more inclined to rely on public transport and active modes; therefore, service levels and operating hours impose stronger constraints on them compared to older adults, leading to a higher incidence of unrealized trips.

In the urban model, homemakers, those living in households with three or more members, and low-income individuals were more likely to experience unrealized trips. Homemakers travel more frequently than other groups (in Cluster 3, 46.1% of respondents reported shopping three times or more per week, whereas this proportion rises to 54.5% among homemakers). They are overwhelmingly female (98%) in Cluster 3. Females often take on responsibilities such as picking up children, shopping, and running household errands, which frequently forces them into complex trip chains [57]. Such circumstances make them particularly prone to time poverty [58]. Moreover, because they tend to depend more heavily than men on public transportation, this may help explain the higher number of unrealized trips reported in this group. In the rural model, the negative values observed for female may be partly explained by the burdens of complex trip chains and the resulting time poverty.

In the rural model, the final railway service time was significant for leisure, whereas the final bus service time was significant for shopping. While it cannot be definitively stated, a possible explanation lies in the differences in access distance and activity time between them. Leisure activities tend to involve longer travel distances on average and may take place at night (e.g., taking a train to attend a sporting event), making them more sensitive to early final train schedules. In contrast, shopping generally requires shorter access distances and is often completed within the rural area. In such cases, given that rural areas typically have smaller railway network and final train times do not vary within the area, differences in the final bus service time are likely to exert a stronger influence.

It is noteworthy that variables were found to significantly differ between urban and rural areas in the BOP model. In urban areas, the number of train services was a key factor, while in rural areas, the last service time and the frequency of buses were the most significant variables. This contrast reflects differing transport infrastructures, which are rail-based in urban settings and bus-reliant in rural ones. In urban contexts, extended service hours are essential for enabling continued participation in leisure activities, and in rural areas, whether a bus is available at night becomes critical.

These findings redefine the concept of “transportation disadvantaged” by isolating Cluster 3. The ZIBOP model demonstrates that transportation disadvantage is not synonymous with simply older adults but rather arises from the interaction between individual attributes (especially younger adults) and public transportation constraints. Unrealized trips can be regarded as a form of social exclusion. Many previous studies have focused exclusively on older adults, implicitly assuming that they are the primary representatives of the “transportation disadvantaged.” While this assumption holds when considering personal factors such as health status, the impact of transportation constraints appears to be stronger among younger populations.

From the perspective of mobility justice, our findings also reveal structural inequities in how transit services are distributed between urban and rural areas. Rural municipalities generally have smaller fiscal capacity, which limits the budget that can be allocated to public transportation. Combined with longer route distances and lower population densities, this often results in fewer services per line compared to urban areas. Consequently, despite residents’ equal right to mobility, rural inhabitants face disproportionately restricted access to essential shopping and leisure opportunities. This highlights how institutional and structural constraints reinforce spatial inequities in mobility.

A key insight for ensuring equity in urban planning lies in the fact that younger people frequently use public transport and active modes. Thus, the findings provide supporting evidence for advancing urban development strategies such as Transit-Oriented Development (TOD) and walkable cities. Furthermore, the result that transportation constraints differ between urban and rural contexts reinforces the need for optimizing mobility services (e.g., demand-responsive transit, MaaS) according to regional structure.

By considering unrealized trips, the combination of realized trips and unrealized trips more accurately reflects individuals’ true mobility needs. Quantifying this demand offers a methodological contribution that goes beyond conventional trip-based analyses.

5. Conclusions

In this study, we analyzed the factors contributing to unrealized discretionary trips. Using variables related to transportation accessibility, K-prototypes algorithm was conducted, classifying respondents into three groups. The transportation-disadvantaged group was found to have a higher frequency of unrealized trips. We then applied the ZIBOP model separately for urban and rural residents. The results showed that, in urban areas, unrealized trips were more likely among younger individuals, homemakers, those living in households with three or more members, and low-income individuals. In contrast, in rural areas, women were more likely to experience unrealized trips, while unemployed individuals were less likely to do so.

In both urban and rural models, variables representing transportation service levels such as frequency and the last rail and bus service times, as well as distances to destinations, were significantly associated with unrealized trip frequency. In urban areas, the number of train departures was a key determinant, while in rural areas, bus service frequency and timing had stronger effects.

Since younger individuals and non-license holders are more likely to experience unrealized trips, the development of shopping facilities within walking and cycling distance should be promoted as a long-term urban policy. In addition, because train frequency was found to be an important factor in urban areas, extending night-time services and optimizing timetables could serve as effective short-term measures. In rural areas, where bus-related variables are more critical, the introduction of demand-responsive transit (DRT) services would be a viable short-term solution.

This study has several limitations. First, unrealized trips were measured solely based on respondents’ subjective recall, which may be prone to biases. Nonetheless, unrealized trips are inherently unobservable in behavioral datasets, as they represent trips that did not occur. In addition, the measurement of unrealized trips in this study may include both cases where individuals were physically unable to travel and cases where they actually traveled to some extent but still felt a subjective sense of insufficiency, wishing to travel more. To clearly distinguish these factors from transportation constraints, it is necessary to add items to the questionnaire where respondents describe or select reasons for unrealized trips. However, the regional distribution of transportation service levels collected in this study’s questionnaire (e.g., the distribution of municipalities by mean rail frequency) has been confirmed to be similar to external objective data. In addition, the clustering results show that unrealized trips are significantly more frequent among the “transport-disadvantaged” group with restricted access to transportation. This suggests that, at least in part, the indicator used in this study reflects transportation constraints.

Although the survey states “due to lack of available transportation,” psychological and motivational factors such as fatigue can also contribute to unrealized trips. However, since the survey design did not include questions on psychological or motivational reasons distinct from transport constraints, we were unable to incorporate psychological constraints as explanatory variables, which represents a limitation of this study. Nevertheless, the zero-inflated component of the ZIBOP model statistically separates individuals with low travel motivation, who are less likely to have unrealized trips, from those constrained by transportation conditions.

Regarding trip frequency, the analysis assumes that shopping and leisure trips are made separately. However, some respondents may conduct both activities in a single trip but have reported only one purpose, potentially underestimating the true frequency.

In the structure of this study’s model, the possibility of endogeneity cannot be completely ruled out. One potential source is a bidirectional causal relationship, for example, the possibility that a higher likelihood of unrealized trips could lead to lower income.

In future work, Global Sensitivity Analysis (GSA) could be applied to avoid overstating transportation as the primary determinant by strengthening the assessment of the relative importance and robustness of explanatory variables [59,60]. GSA is a methodology that quantifies the extent to which each input variable contributes to the variance of model outputs while accounting for uncertainty across the entire input domain. It is distinguished from local sensitivity analysis by its explicit treatment of nonlinearities and interactions among variables (e.g., income and transport service levels). In other words, GSA can reveal what proportion of the variance in unrealized trip probabilities is attributable to each explanatory factor. Representative techniques include the Elementary Effects method for identifying dominant factors, variance-based methods for variance decomposition, and density-based methods for comparing probability distributions, often combined with sampling strategies such as Latin Hypercube Sampling.

The analysis focused on transportation-related factors, while other constraints, such as health conditions or caregiving responsibilities, were not fully considered. Unrealized trips sometimes result from a combination of these constraints, meaning that the model has limitations in explanatory power.

In future work, adding survey questions that directly ask about the reasons for unrealized trips, as well as linking behavioral data with survey responses, may enable a clearer distinction between transport constraints and psychological or motivational factors.

In this study, we focused on answering the question of how much public transportation constraints affect the increase in unrealized trips by limiting the scope of application of the ZIBOP model to Cluster 3. One reason for this is that applying the ZIBOP model to clusters with few unrealized trips results in fewer significant explanatory variables and complicates the interpretation of the results. In the future, it will be desirable to select models and variables that can be compared between clusters.

In addition, analyzing temporal changes in accessibility and transportation services at the regional level (and their effects on realized and unrealized trip frequencies) would valuably extend this research. Finally, by integrating the model developed in this study with regional statistics, it is possible to estimate the frequency of unrealized trips at the municipal level. These estimates can then be compared with indicators such as life satisfaction or accessibility scores to better understand the relationship between them.