1 Introduction

In recent years, the growing aging of society, the reduction in population, and the outflow of young people to big cities are common issues in rural cities in Japan. Also the decrease in the number of public transport passengers and the resultant loss of income for public transport companies are serious problems in those cities. However, the role of public transport still remains an important one, as it is a mode of transportation that is suitable for an aging society that is experiencing a reduction in population (Burkhardt 2003; Fujiyama 2009; Katakami et al. 2014). Therefore, public transport authorities and related organizations require marketing strategies that maintain public transport services for vulnerable users. However, such strategies are always discussed based mainly on information related to income and cost, without consideration of the nature of trips made by public transport passengers. According to interviews with employees of public transport authorities in Kochi City (conducted by the authors in January 2014), the outcomes of such discussions always result in plans for service cuts, such as to the number of routes, and tram and bus frequencies. Public transport authorities also say that it is difficult to win back passengers once they stop using public transport after a cut in services. To discuss the above issue, understanding the behavior of public transport passengers and the characteristics of public transport usage based on transport data is the first step in gaining fundamental information that is vital for any discussion in rural cities. Because the number of passengers tends to decrease in rural cities in Japan, this study focuses on the reaction of public transport passengers in case that the service level has been changed by public transport authorities to revitalization of public transport in those cities.

To compile such information, many researchers have suggested and evaluated public transport policies including integrated services such as flexible transport services (Mulley and Nelson 2009; Mulley and Daniels 2012), a seamless transfer system (Chowdhury and Ceder 2016), an integrated fare system including free public transport (Witte et al. 2006; Thogersen 2009; Grange et al. 2013; Wang et al. 2015; Cools et al. 2016), services focusing on social welfare (Holmgren 2014; Chalak et al. 2016), an integration or optimization by combining all-stop, short-turn, and express transit services (Tetreault and EI-Geneidy 2010; Leiva et al. 2010; Cortes et al. 2011; Chiraphadhanakul and Barnhart 2013; Larrain et al. 2015; Zhang et al. 2016), and so on. Most of these studies have some merit in investigating how to manage public transport authorities and in understanding typical travel patterns of public transport passengers. However, when analyses of the behavior of public transport passengers and the characteristics of public transport usage are used, the data collected from interviews or questionnaire surveys show the preferred modes of transportation of respondents but not their characteristics (including variations in travel patterns) that need to be considered in order to discuss the routes and timetables of public transport so as to achieve administrative improvement. To resolve the above issues when analyzing public transport in rural cities, public transport usage should be monitored to understand passengers’ characteristics, which is information conductive for improving the quality of services. In answer to requests for the collection of such data, there are several monitoring devices used in public transport in many cities, such as sensors that automatically count the number of boarding and alighting passengers (Tamin 1997; Kita and Tsukioka 2005; Zhang and Teng 2013). While data from these devices can be used for analyzing variations in the number of passengers, it is not directly possible to analyze the origin–destination (OD) of passengers, information that is important in the study of how to best manage public transport. However, smart card fare-collection systems that directly measure OD between boarding and alighting stops have been widely introduced in public transport systems, even in small cities. In recent years, data on trips collected from smart cards have proved useful in understanding the characteristics of passengers’ behavior (Pelletier et al. 2011). Many researchers have analyzed the trip patterns and trip regularity of passengers so as to understand fundamental passengers’ behavior, as smart card data record passengers’ behavior patterns by using continuous measurement data (Chapleau and Chu 2007; Morency et al. 2007; Devillaine et al. 2012; Asakura et al. 2012; Nishiuchi et al. 2013; Ma et al. 2013; Kieu et al. 2014). Smart card data have also been analyzed to provide details on passengers’ behavior for public transport planning (Kusakabe et al. 2010; Kurauchi et al. 2014; van der Hurk et al. 2015) and the marketing aspect (Seaborn et al. 2009; Páez et al. 2011; Asakura et al. 2012). Even if Grange et al. (2013) and Wang et al. (2015) successfully estimated passengers’ price elasticity under fare change of public transport based on smart card data, they analyzed the aggregate number of passengers. Therefore, it has still some space to improved evaluation of passengers’ behavior using smart card data. Such information on passengers’ behavior from smart card data is fundamental to understanding the usage characteristics of public transport from the marketing strategy point of view. Therefore, it is important to understand the current situation of public transport usage in rural cities.

With the aim of contributing to the revitalization of public transport, Nishiuchi et al. (2014) analyzed variations in time series data of OD between tram stops (tram OD) in Kochi City using DESUCA smart card data. The analysis used a state-space model and the OD volume was divided into trend components, daily variation components, and auto regression (AR) components. The results show the characteristics of the time series number of tram OD passengers, such as that the number of passengers on Monday is low but it is high on Friday. However, this study did not clarify the characteristics of daily variation, and therefore, it does not provide knowledge that can directly contribute to knowing what types of tram OD pairs should be focused on to implement revitalization measures by public transport authorities. The present paper analyzes the impact of the reduction in tram services in the Kochi network implemented on 1 November 2012. First, this paper develops a model to estimate the probability of change in the number of passengers after 1 November 2012, and it then examines what types of tram OD pairs significantly change the number of passengers. Moreover, this study focuses on ascertaining what improvements should be implemented by public transport authorities to increase usage of the tram network in Kochi City.

2 Study area and smart card data

2.1 Study area and tram network

The study area in this paper focuses on the tram network, mainly in Kochi City, as shown in the map in Fig. 1. Kochi Prefecture is located on Shikoku Island in the southern part of Japan (colored red in Fig. 1). The population in residential areas within 500 m of tram stops has been estimated as approximately 154,000 people (Kochi Prefecture 2011). The tram network in Kochi City basically consists of an east–west line and a south-north line bisecting each other like a cross. The center of the cross is called the Harimayabashi Tram Stop, where passengers can transfer between the two lines (see Fig. 2). In this tram network, a reduction in service was implemented on 1 November 2012, at different rates of reduction according to the region served by the network, as shown in Fig. 2. In addition to that, service reduction was conducted during off peak hours. The western part of the city is 40% cut off frequency from 9:20 to 15:20 on weekdays and from 8:55 to 19:00 on weekends and holidays. The eastern part of the city is 30% cut off frequency from 10:30 to 16:00 on all days. And the center of the city is 20% cut off frequency.

Fig. 1
figure 1

Kochi prefecture and map of bus/tram stops

Full size image
Fig. 2
figure 2

(map courtesy Tosadenki Railway Co., Ltd.)

Tram network and frequency reduction rate by region

Full size image

2.2 Smart Card DESUCA and data collection

The smart card for this region is called DESUCA, and data results for this card are analyzed in this paper. DESUCA is used for automatic fare collection for public transportation, as is currently commonly found in the world. The DESUCA system started operation in January 2009 and is operated by joint funding from several public transportation companies. According to interviews with staff conducted in January 2014, the current patronage of smart card users on public transportation is almost 65% on weekdays. As bus and tram fares in Kochi City are dependent on the distance traveled, smart card users are required to touch the card on the fare-collection device as they board and alight the bus/tram. Therefore, the DESUCA database records trip information as passengers pay their fares (see Table 1), and the number of transactions per day reflects around 10,000 passengers’ trips for weekdays and around 2000–5000 passengers’ trips for weekends and holidays (see Fig. 3).

Table 1 Collected DESUCA data information
Full size table
Fig. 3
figure 3

Number of smart card transactions per day

Full size image

On the other hand, the DESUCA data collected in this paper does not record information of the routes of all trams. That is, the transfer behavior could not be identified without some assumptions such as “passengers with transfer may appear on raw data within 30 min from the latest (previous) record” which is not a guaranteed description of the passengers’ behavior. Therefore, considering transfer behavior is out of our study. Another limitation of this data set is that the card type that can indicate the age group of the passengers is not recorded for this study. Therefore, the analysis below could not consider the difference of trip characteristics by age or type of passengers. As Table 1 shows, DESUCA data was collected for four months for this study. This was also a limitation of withdrawal of the data from the database because the database is not designed for withdrawing the data to understand passengers’ trip characteristics but only for payment confirmation. It is not possible to use the trip data for a longer period in this paper. Therefore, the analysis for a longer period should be conducted after new and longer data is collected by the public transport authority.

This study analyzed trip data for one tram from its origin stop to its destination stop (tram OD) recorded as smart card data without considering the transfer behavior of tram passengers, as the recorded smart card data does not include whether the passengers transferred tram lines. In addition, the reduction in tram frequency was implemented only for the tram network. Therefore, this study focuses on passengers’ trips for each tram trip. Most of the passengers were low frequency passengers, as shown in Fig. 4. This means there are a limited number of samples if we select only frequent passengers, which in turn means that there is the difficulty in investigating what kinds of passengers were affected in terms of tram-use characteristics versus tram frequency reductions. In this study, we decided to analyze the aggregated number of tram OD passengers instead of using individual trip chain data based on card ID information so as to focus on wider tram OD pairs in the study site. Therefore, it should be noted that the analysis using an aggregated number of tram OD passengers, considering details in travel behavior such as transfer, individual travel frequency, age of the passengers and so on is a limitation of this analysis. In addition, generally the level of OD volumes is not high enough because the value of many ODs might be zero or a small value. Therefore, in this study, we focused on 235 tram OD pairs out of 5852 tram OD pairs that made more than 1000 OD trips during a four-month period under the assumption that the mean number of tram OD trips per day is 10.

Fig. 4
figure 4

Distribution of number of trip days per month

Full size image

3 Methodology

3.1 Cox proportional hazard model for survival time analysis

Survival time analysis is a method of analysis that focuses on the relationship between the times of occurrences of two events, for example the occurrence of a disease and the death of a patient in the medical field, and the failure of a product in the engineering field. The Cox proportional hazard model is classified as a semi-non-parametric model that can introduce a survival function by using the Kaplan–Meier method without the assumption of a probability distribution. The Cox proportional hazard model is shown in Eq. (1).

$$\lambda_{k} (t) = \lambda_{0} (t)\exp \left( {\sum\limits_{i = 1}^{m} {\beta_{i} z_{i} } } \right)$$
(1)

\(\lambda_{k} (t)\): probability of reduction in number of passengers at time t for tram OD pair k; \(\lambda_{0} (t)\): baseline hazard function; \(\beta_{i}\): parameter for covariance i; \(z_{i}\): covariates.

This study is focusing on which OD pairs’ number of passengers are affected by changes in tram service reduction using smart card data. The smart card data records use information continually with the card ID. It is continuously using mark the same ID once it is provided to a card. Therefore, the model should be considered advantageous for smart card data as time series data characteristics. In addition to that the method is to be desirable to explain factors of time series variations. In this sense, survival time analysis can consider both aspects mentioned above rather than other proposed methods related to transport such as several data mining techniques or behavior choice models and so on. In this study, as it is directed at the transition in the number of users, it is not possible to apply a survival time analysis in the same way as in existing studies because some of the passengers will continue to use public transport even if the service of the tram is cut. In other words, it is difficult to directly measure the stop of public transport use or zero value in number of tram OD passengers. Instead, the definition of death in survival time analysis is defined as standard value using the mean value of the number of tram OD passengers of the first week after the service cut started. The number of the periods that the value of the number of tram OD passengers exceeds the standard value in the analysis during the data collection period is totaled. Then the condition of the OD is judged whether OD pairs survive or not. Therefore, this study analyzes the probability of a reduction in tram OD passengers as the survival time for each tram OD pair, that is, which tram OD pair can maintain a certain number of passengers after frequent reductions in service made by public transport authorities.

In addition, the covariates of Eq. (1) are defined in the following section as there may be several factors that contribute to explain the changes in the number of tram OD passengers. That is why it was decided to apply the proportional hazard model of Cox, which is capable of handling the multiple factors as covariates to understand the factors to explain time series variation of the number of tram OD passengers.

3.2 Definition of Cox proportional hazard model

As mentioned in the preceding section, this study analyzes which tram OD pairs can maintain a number of passengers or which will find it difficult to maintain a number of passengers after a reduction in the frequency of tram services.

3.2.1 Definition of reduction probability of tram OD passengers

\(\lambda_{k} (t)\) is defined as the probability of a reduction in the number of passengers per tram OD pair k at time t. Specifically, the meaning of the reduction is the proportion of the number of weeks for which the number of tram OD passengers is lower than the standard value of one OD pair. The concept of the definition of reduction is shown in Fig. 5. The Cox proportional hazard model must be set for the period until the end of the target event. This study defines the period as 8 weeks, which is the period for which data are available after the start of reduced tram services by public transport authorities. During the 8 weeks, the number of weeks for which the number of tram OD passengers is lower than the standard value is used to aggregate data to estimate the probability of a reduction in tram OD passengers for each tram OD pair. The standard value is calculated as the mean number of tram OD passengers in the first week after the start of operation of the reduced service. Therefore, the data set for \(\lambda_{k} (t)\), which is the probability of a reduction in tram OD passengers at time t, is prepared by aggregating the weeks that the number of tram OD passengers is low over 7 weeks, following the concept shown in Fig. 5 to obtain a proportion. The result of aggregation of the number of OD pairs which were reduced in tram OD passengers shows that it was 167 OD pairs out of 235 OD pairs. And this paper assumes that the trend of the number of tram OD passengers may be reduced immediately after tram frequency reduction, rather than the general trend of reduction of the number of tram passengers as the impact of service cut by public transport authorities.

Fig. 5
figure 5

Concept of definition of probability of a reduction in tram OD passengers

Full size image

3.2.2 Definition of the baseline hazard function

The baseline hazard function is assumed as the trend of the probability values of subjects’ occurrence which is not affected by changing the values of the covariance vector z. In other words, the baseline hazard function represented as probability values can be estimated by zero value of covariance vectors z. This means that the function is not dependent on each individual subject of analysis. In this study, we set one function for all tram OD pairs as describing a common trend of probability.

3.2.3 Data formulation for covariance vectors

Covariates vector z in Eq. (1) is formulated based on the frequency change between before and after the reduction in tram service as policy variables, the number of passenger data of each tram OD pair such as mean value by time and the coefficient of variance. The variable of the number of passengers is formulated based on the state-space model. The state-space model is well known for analyzing time series data by dividing fluctuations into multiple components, and understanding factors of fluctuations by estimating unmeasurable states such as trend, season, weekly variation, and daily variation. The model has two main equations, one is the state equation (Eq. (2)) and the other is the measurement equation (Eq. (3)). \(y_{n}\) in Eq. (3) is defined as the measured time series data, and it is composed of four variables where Tn is the trend variation component, Dn is the cyclical variation component, ARn is the auto-regression variation component, εn is the error component, and k, p and m are the period of calculation time series data as shown in Eqs. (4)–(7). Factors of time series data variation are identified by analyzing all time-series data of all components. The same approach is already used to analyze time series data of traffic volume (Sasaki et al. 2012) and traffic state (Lu and Zhou 2014).

$$x_{n} = F_{n} x_{n - 1} + G_{n} v_{n}$$
(2)
$$y_{n} = Hx_{n - 1} + \omega_{n}$$
(3)
$$y_{n} = T_{n} + D_{n} + AR_{n} + \varepsilon_{n}$$
(4)
$$\varepsilon_{n} \sim N(0,\sigma^{2} )$$
$$T_{n} = \sum\limits_{i = 1}^{k} {c_{i}^{(k)} t_{n - i} + v_{n1} }$$
(5)
$$\nu_{n1} \sim N(0,\tau_{1}^{2} )$$
$$D_{n} = - \;\sum\limits_{i = 1}^{q - 1} {D_{n - i} + \nu_{n2} }$$
(6)
$$\nu_{n2} \sim N(0,\tau_{2}^{2} )$$
$$AR_{n} = \sum\limits_{i = 1}^{m} {\alpha_{i} AR_{n - i} + \nu_{n3} }$$
(7)
$$\nu_{n3} \sim N(0,\tau_{3}^{2} ).$$

In this paper, the time series data of the number of tram OD passengers from DESCA data are broken down by the state-space model to describe the characteristics of fluctuations in the four variations of trend (kind of seasonal) variation, daily variation and AR variation as colors, noise and error variation Eqs. (4)–(7). Figure 6 shows an example of the results of applying a state-space model to time series tram OD passengers based on DESCA data (Nishiuchi et al. 2014). Statistics such as the mean and coefficient of variance calculated from these divided time series data are used for the variables of covariance to describe the usage condition of each tram OD pair. These statistical data and service level information of public transport are prepared as variables of the covariate vector z.

Fig. 6
figure 6

Example of the state‑space model application for the number of tram OD passengers (upper left: trend variation component, upper right: cyclical variation component, bottom left: auto-regression variation component, bottom right: error component)

Full size image

4 Parameter estimation for the Cox proportional hazard model

4.1 Development of a Cox proportional hazard model for OD analysis

4.1.1 Parameter estimation for the baseline hazard function

Figure 7 shows the estimation results of the baseline hazard function. An exponential approximation is applied to decide the parameter of the function. From Fig. 7 and Eq. (8), the probability of maintaining the same number of passengers decreases over time. This means that the total number of passengers for the entire tram network in Kochi is basically decreasing.

Fig. 7
figure 7

Estimated function of the baseline hazard function

Full size image
$$\lambda_{0} (t) = 0.7222e^{ - 0.116t} .$$
(8)

4.1.2 Estimation of parameters for covariates

Table 2 shows the estimation results of the parameters for the covariate vector. From Table 2, the sign of each parameter can be understood, and a statistical test shows the significance for all explanatory variables that are at least 5% of significance. Therefore, the defined covariates are a factor explaining the probability of the reduction in the number of tram OD passengers.

Table 2 Estimation of parameters for covariance
Full size table

From the results, the change in tram frequency between before and after the reduction in tram services is negative. This means that the measure adopted by public transport authorities is a factor behind the reduction in the number of tram OD passengers. On the other hand, the value of the parameter shows that the impact of this factor is limited. Therefore, the measure itself does not affect the trend in the number of tram OD passengers 8 weeks after the introduction of the measure. The fluctuation in the number of tram OD passengers for both weekdays and weekends, including holidays, is also a significant factor for the explained variable, and the value of the parameter is positive. This means that an OD pair that has a higher number of tram OD passengers contributes to maintaining the number of passengers for the entire network. In addition, the variability of the fluctuation component for weekends and holidays, and special days such as rainy days (more than 1 mm/h in a day) significantly affects the explained variable. If the variance is higher, the OD pair contributes to an increase in the number of tram OD passengers. Using the estimated model, we examine the number of tram OD passengers using a survival time analysis.

5 Understanding changes in the number of tram OD passengers

5.1 Analysis of changes in the number of tram OD passengers

This section presents an analysis of which tram OD pairs maintain the same number of passengers during the 8 weeks following the reduction in service using the developed Cox proportional hazard model. Figure 8 shows the estimation results of probability of the reduction in tram OD passengers compared to the mean value for 235 target OD pairs. The results show that tram OD pair travel from a suburban area to a city center yields a lower value of probability than an OD pair travel from a city center. This tendency is examined on the supposition that a passenger who travels from a suburban area may use a tram to commute to work and school. Therefore, the probability is estimated as lower than for other tram OD pairs. In addition to that, Table 2 shows that the estimated values of parameters for the averaged number of tram OD passengers on a weekday is higher than the one on weekends. From the estimation results, the number of tram OD passengers on weekdays may affect the probability of the reduction in tram OD passengers rather than the one of weekends. Commuters on OD pair travel from a suburban area tend to travel on weekdays. Then the probability of the reduction in tram OD pairs on those OD pairs may estimate lower values.

Fig. 8
figure 8

Distribution of probability of a reduction in the number of tram OD passengers

Full size image

From the tendency mentioned above, public transport authorities should discuss maintaining commuting usage by increasing the convenience level for those passengers. On the other hand, for passengers who travel from a city center, the results yield a higher probability than the previous OD pair. Therefore, public transport authorities should consider how to increase the number of OD passengers. In particular, they should focus on the fact that the characteristics of the coefficient of variance on weekends and holidays, and rainy days can contribute to an increase in the number of tram OD passengers, based on the parameters of the developed model. Therefore, public transport authorities should consider holding promotions for transport usage on rainy days and re-examine point service for smart card use on weekends and holidays. To confirm its necessity, a detailed analysis using individual trip chain data including transfer and usage of busses should be made by combining other data sources including census survey data as conducted by Kusakabe and Asakura (2014) and Devillaine et al. (2012), and also statistical information of the area (quality of services, car ownership, income) (Paulley et al. 2006), as well as other analysis techniques.

5.2 Sensitivity analysis

To understand the change in the probability of a reduction in tram OD passengers, sensitivity analysis is implemented using the model developed in the previous section. The change in the reduction probability of passengers is confirmed by changing tram frequency and the mean number of tram OD passengers. Sensitivity analysis here is assumed by analyzing a typical tram OD pair, and therefore, mean values for other variables are given in this calculation.

5.2.1 Impact of change in tram frequency

Figure 9 shows the tendency of the probability of a reduction in tram OD passengers in a case by changing the value for the difference of tram frequency between before and after the reduction in the tram frequency variable. The value of the probability of a reduction in tram OD passengers is distributed around 90%, even if the value for the gap becomes larger. This shows that the impact of a change in tram frequency does not affect the probability of a reduction in tram OD passengers. On the other hand, the probability in the case of an increase in tram frequency does not affect the change in the number of tram OD passengers either. Therefore, changing tram frequency is not a suitable measure to increase tram usage in Kochi City, regardless of the fact that the number of tram OD passengers in the whole network in Kochi City is basically decreasing. In addition to that, it should be noted that this paper analyzed the tendency in case of focusing on time series data immediately after the reduction of tram frequency. That means the slope of the function possibly changes if smart card data collected for a longer period is analyzed for modeling.

Fig. 9
figure 9

Sensitivity analysis for change in the value of the gap in tram frequency

Full size image

5.2.2 Impact of the change in the number of tram OD passengers

This section shows the results of a sensitivity analysis of the number of tram OD passengers on weekends and holidays to ascertain which number of trips on weekends and holidays is more sensitive to a change in the number of tram OD passengers. Figure 10 shows the difference in the probability of a reduction in tram OD passengers by changing the number of tram OD trips on weekends and holidays. To maintain the level of the number of tram OD passengers, the probability of a reduction in tram OD passengers can be 0% if the mean number of tram OD passengers on weekends and holidays increases by just one trip/day, regardless of the mean number of tram OD passengers on weekdays needing to increase by two trips on average. Therefore, the results show that the measures that need to be taken on weekends and holidays to increase the number of OD passengers by public transport authorities have significant impact on public transport usage, such as incentives to use public transport by smart card on weekends and holidays. Smart card data in this study area can also be analyzed based on the card ID including card type, trip chain in the bus network and so on. Therefore, more detailed analysis focusing on individual trip characteristics should be conducted to understand what kind of passengers can keep their usage of public transport.

Fig. 10
figure 10

Sensitivity analysis for value of average number of tram OD passengers

Full size image

6 Conclusions

This paper analyzed how public transport authorities can maintain the number of tram OD passengers after they reduce services. The Cox proportional hazard model for survival time was used to develop the model used in this paper. To include specific fluctuation components of time series data of the number of tram OD passengers, the state-space model was also used to formulate the explanatory variables for covariance of the Cox proportional hazard model. The estimated parameters showed a change in tram frequency before and after the reduction in services, the mean number of tram OD passengers on weekends and holidays, the coefficient of the variation in the number of tram OD passengers on weekends and holidays, and rainy days are significant factors. In addition, the impact of the change is not significant enough to explain the change in the number of tram OD passengers. An estimation of probability of the reductions in the number of tram OD passengers after the introduction of a reduced service using the proposed model also showed the tendency that tram OD pairs from a suburban area to a city center can maintain their number of passengers regardless of tram OD pairs from city center. A sensitivity analysis of the proposed model shows that a change in tram frequency does not directly effect a change in the number of tram OD passengers. However, measures to increase the mean number of tram OD passengers on weekends and holidays might have a significant impact in reducing the probability of a reduction in the number of passengers. These perceptions could be providing more specific information based on trip characteristics recorded by smart card data. Analysis results might suggest what kind of rearrangement is required in terms of the characteristics of the location of a tram OD pair. For example, the authority may consider spatial trip characteristics to promote ridership of public transport such as serve high frequent tram in central city for short distance travelers. Also the authority may consider the elasticity of trip reduction in case they change tram service on weekends and holidays with more detailed analysis focusing on individual travel behavior characteristics.

Tasks for further study include considering usage of bus networks by smart card holders and also transfer behavior such as from bus to tram to consider the chain of public transport users. In addition, because the data set used in this study was not included card type data due to the problem of data withdraw from DESUCA database, individual behavior data created from smart card data such as card type by age group, frequency of public transport use, and use of commuter pass should also be considered. This information should be combined with other data sources such as geographical information including car ownership, income, and so on, including time series data from other transport mode such as variation of traffic volume to discuss the impact considering multiple transport mode, as the results of this analysis showed different characteristics according to the geographical condition of OD pairs. Furthermore, since this paper analyzed only a short period just after tram frequency was reduced, an analysis using smart card data collected over a longer period, such as several years, should be considered. Such research would yield more specific information for discussing rebuilding and improving public transport networks to maintain public transport operations in rural areas of Japan.