Valuations on quality of service for intercity travels using high-speed rail

Abstract

There have been extensive studies on the monetary valuations for the transit service attributes such as fares, frequency, and in- and out-of-vehicle time. On the contrary, five categories of the elements relating to tangible, comfort, convenience, customer care, and safety and security have not been sufficiently evaluated. This paper explored the valuations based on the service improvements of Korean high-speed rail. The survey for attribute selection identified 11 user-tailored items for the betterment. Factors pertinent to comfort were predominantly preferred. Subsequently, choice experiments were conducted in the form of train-type choices. Binary logit models were then specified and estimated using the data collected. Monetary values were obtained in both individual attributes and a service bundle. Policy implications were also investigated in terms of price discriminations. The direction for future studies, particularly in examining user heterogeneity, was finally suggested.

Keywords:

• Monetary valuations
• rail quality of service
• high-speed rail
• intercity travels
• choice experiments
• price discriminations

Introduction

High-speed rail services in Korea have drastically changed the intercity transport system of the country. Daily patronage in 2014 was around 150,000, which is approximately three times greater than that in 2004, when the service was launched. Its comparative advantage is particularly noticeable in long-distance journeys. For example, the modal share of railways for the Seoul–Busan trips, which are 424 km long and have the largest intercity ridership in Korea, has increased from 38% in 2004 to 69% in 2014. Air share, in contrast, has dropped from 32 to 15% during the same period. This rapid travel system has also contributed to the growth of the regional economy. Land values for the beneficiary cities have shown higher rate of increase than those of non-serviced towns. Creation of new jobs in tertiary employment for cities with high-speed rail stations has also exceeded the national average (Korea Transport Institute 2014).

However, customers require better quality of service for this most expensively priced ground travel mode in Korea, even though they are generally satisfied with travel time savings by this train (Chang et al. 2011). Hence, the operator, or Korail, is considering user-tailored service improvements relating to the types of tangible, comfort, convenience, safety and security, and customer care. These target elements are beyond the conventional transit service factors such as fares, frequency, and in- and out-of-vehicle time. Thus, quality of service in this paper represents the five categories of elements other than the traditional attributes.

Better quality of service is more attractive to potential passengers and can generate greater patronage, but it normally entails higher costs. In particular, quality of service depends to a great extent on the operating decisions made by the transit agency within its budget constraints. Thus, the identification of user-preferred service attributes and its corresponding valuation can be a prerequisite for making the decision, which is the primary motivation of this paper.

Literature, however, does not supply sufficient evidence on the monetary values for the service attributes. Three observations pertinent to the purpose of this paper are noted. First, extensive studies can be found that deal with the measurement for transit services, but they normally involve the performance assessment with grades or scales to monitor and control the quality of service provided to the passengers; see, for example, Chang et al. (2011), Kittelson & Associates, Inc., Parsons Brinckerhoff, KFH Group, Inc., and Texas A&M Transportation Institute (2013), Nathanail (2008), and; Wiley, Maoh, and Kanaroglou (2011). Second, the existing research is normally based on urban (Hensher, Rose, and Collins 2011), suburban (Basu and Hunt 2012; Lu, Fowkes, and Wardman 2008; Pepper, Spitz, and Adler 2003; Wardman and Whelan 2001), or both (Polydoropoulou and Ben-Akiva 2001; Wardman and Murphy 2015) traffic flows, but it is not very easy to find studies that treat intercity travels in the context of quality of service. Finally, indeed a few studies (Pepper, Spitz, and Adler 2003; Wardman and Murphy 2015; Wardman and Whelan 2001) report money values for several attributes, the majority (Basu and Hunt 2012; Hensher, Rose, and Collins 2011; Li and Hensher 2011; Lu, Fowkes, and Wardman 2008; Polydoropoulou and Ben-Akiva 2001; Wardman and Whelan 2011) however focuses on evaluating crowding inside buses and/or trains. This paper, on the other hand, aims at comprehensive coverage of components in valuing quality of service for intercity rail travel in Korea.

The next section describes the methodology adopted in this paper. The platform is the two-stage survey including attribute selections and contingent choices. Subsequently, valuations on service improvement are presented. Starting from the modeling framework employed, the report for results and the discussion regarding policy implications are the order of this section. Finally, the concluding remarks contain the summary of research and the direction for future studies.

Survey

This section describes the two-stage survey adopted in this paper. The survey for attribute selection identifies passengers’ preferences on the factors affecting quality of service for intercity rail travels. The subsequent choice experiments supply the input data for monetary valuations on the selected elements.

Attribute selection

There are lots of factors affecting rail quality of service, as shown in Table 1. The list is heavily dependent upon Chang et al. (2011) with the addition of the studies in Table 2. It would be impractical to consider all involved items. This research thus decided to select representative service attributes. The survey for the selection of items was separately conducted for professionals and the public. This design was proposed out of the hypothesis that the two groups could show different preferences. Sixty-one experts of rail planners and engineers were interviewed. Two hundred and eighty-five laymen were surveyed through internet-based questionnaire. Each respondent was asked to select 10 most important attributes of quality of service for intercity rail travel. The priority was the factors affecting travel comfort. Eleven asterisked items were selected for valuation based on the total votes. The 11th preferred element, or back-seat screen, was added since a marginal difference was recorded between the 10th and 11th attributes. There were no significant differences in preferences between professionals and ordinary travelers against the hypothesis, even though the order of importance in the attributes was somewhat dissimilar. It should also be noted that respondents may differently interpret the service items since the screening was administered in a descriptive manner, particularly in the web-based survey. Thus, the resulting selection may be biased, which is one of the limitations of this research.

Table 1. Candidate of service attributes for valuation.

Service type	Attributes
Tangible	Train exterior appearance, train interior appearance, in-train cleanliness*, wornness of rolling stock, and staff uniform and appearance
Comfort	Seat availability*, seat armrest, seating direction*, seat spacing*, seat with a view, seating layout*, mechanical vibration*, mechanical noise, man-made noise*, temperature control*, lighting quality, in-train stench, air quality, and width of gangway
Convenience	In-train café, luggage storage, facilities for the disabled, on-train announcements, plug sockets for laptops or mobiles*, free or cheap Wi-Fi*, and seat-back screen*
Safety and security	Safety equipment (e.g. portable fire extinguisher and emergency hammer), luggage security, and on-board security guards
Customer care	Staff courtesy, staffs on train, and complimentary meal

Note: *Represents 11 most preferred attributes.

Table 2. Some collections of related studies.

	Mode	Methodology		Traffic flows	Trip purpose	Attributes valued
		Survey	Model
Wardman and Murphy (2015)	Train	SP (contingent choice) + RP (CCTV footage)	Discrete choice models (MNL)	Urban and suburban	Commuting, business, and others	Crowding and seating layout
Basu and Hunt (2012)	Train	SP (contingent choice)	Discrete choice models (MNL, ML)	Suburban	Not available	Crowding
Hensher, Rose, and Collins (2011)	Metro	SP (contingent choice)	Discrete choice models (ECL)	Urban	Commuting	Crowding
Lu, Fowkes, and Wardman (2008)	Train	SP (contingent choice)	Discrete choice models (MNL, HL)	Suburban	Commuting, business, education, and leisure	Crowding and punctuality
Pepper, Spitz, and Adler (2003)	Train	SP (contingent ranking)	Not explicit	Suburban	Commuting	Seat material, train type, seating direction, system seating capacity, overhead racks, seating comfort, and seating location
Polydoropoulou and Ben-Akiva (2001)	Mass transit	RP + SP (contingent choice)	Discrete choice models (NL)	Urban and suburban	Commuting, education, shopping, and other	Travel time reliability and seat availability
Wardman and Whelan (2001)	Train	RP + SP (contingent choice)	Discrete choice models	Suburban	Not available	Comfort, layout, ride, noise, ventilation, ambience, and crowding

Notes: ECL = error component logit; HL = heteroskedastic logit; MNL = multinomial logit; ML = mixed logit; NL = nested logit.

Mass transit in Polydoropoulou and Ben-Akiva (2001) represents any possible combination of heavy rail, enhanced bus, light rail, and suburban rail.

Only those studies published in journals since 2000 were listed. Even though each manuscript includes diverse references, most of them are unpublished articles or proceedings of which original texts are hard to obtain. Review papers (e.g. Li and Hensher 2011; Wardman and Whelan 2011) are also excluded from the table.

Contingent choices

The questionnaire comprised travel characteristics, socioeconomic properties of respondents, and stated preference surveys, as shown in Table 3. It was a person-to-person interview done in August 2014. Four hundred high-speed rail travelers were surveyed. Their socioeconomic properties are summarized in Table 4.

Table 3. Survey components.

Survey	Components
Socioeconomic properties	Age, gender, occupation, income, etc.
Travel characteristics	Trip purpose, origin/destination, access mode, ticket class, etc.
Stated preferences	Choice experiments by scenario

Table 4. Socioeconomic characteristics of samples.

	Number of samples (%)
Total	400 (100.0)
Age
20s	105 (26.3)
30s	105 (26.3)
40s	108 (27.0)
50s	82 (20.5)
Gender
Male	212 (53.0)
Female	188 (47.0)
Occupation
Professionals or tertiary employments	196 (49.0)
Primary or secondary employments	67 (16.8)
Students	67 (16.8)
Others	70 (17.5)
Monthly income (10000 wons)
<300	61 (15.3)
300–400	86 (21.5)
400–500	88 (22.0)
500–600	70 (17.5)
600	89 (22.3)
Not available	6 (1.5)

Choice experiments are the primary platform for the survey in this paper. This stated preference technique is popularly used to monetize non-market goods. Most related studies have also adopted this approach, as shown in Table 2. Figure 1 is an example of contingent choices conducted in this paper. It was a photograph-based exercise with subsidiary artist impressions. Experiments using tables (Chang 2010; Pepper, Spitz, and Adler 2003; Polydoropoulou and Ben-Akiva 2001) or diagrams (Hensher, Rose, and Collins 2011; Wardman and Murphy 2015) are found in the literature, but the realistic images can more effectively convey the survey information to respondents. Comparative images with or without service improvements were located in the middle. Auxiliary texts were also added to explain the differences in quality of service. The experiment is hence based on an abstract choice context involving a within-mode comparison of different train alternatives. This design could excessively deliver the merit and demerit of service attributes. The simplified design however is helpful in placing a clear emphasis on the quality of service.

Figure 1. Example of choice experiments (seat spacing).

Fares that were the payment vehicle in this research were supplied on the bottom. The core of choice experiments would be the setup of the levels for this bottom-line item. An open-ended question was applied to draw the temporary levels of fares for this experiment. The preliminary survey contributed to range the upper and lower limits of levels. The final levels, as shown in Table 5, were determined after a series of pilot surveys done in July 2014. Since there are only two alternatives that contain the varying levels of attribute each, the full factorial design was adopted. In general, an experimental design aims at maximizing the amount of information on the parameters of a distribution under study. Two desirable properties of estimates for the parameter would be consistency and efficiency. Consistent designs are congruous with real preferences while efficient designs seek to minimize the standard errors of the parameter estimates. Not just the factorial approach but also methods that require prior information over the parameter such as D-efficient designs have been found in the literature; see, for example, Rose and Bliemer (2009). However, no universal method for optimal design has been established yet. The examination for the optimality is beyond the scope of this paper. Empirical treatment for an acceptable design based on the full factorial design was considered in three ways. The first is to increase the sample size, which has the effect of reducing both bias and variance. The 400 samples collected in this study represented the first property. Second, since excessively high or low levels inflate the variance of estimates, keeping the level away from the tails of a distribution can be helpful for good designs. A series of pilot interviews were endeavored along with the preliminary survey to reduce the extreme values for levels. The final treatment is to involve the ratio of parameters because the bias in its estimates is substantially smaller than that of the individual parameters (Hanemann and Kanninen 1999). The proposed marginal rate of substitution that will be discussed in the next section addressed this requirement.

Table 5. Levels of fares in the choice experiments.

Service attribute	Level 1 (0.5μ)	Level 2 (μ)	Level 3 (μ + σ)	Level 4 (μ + 2σ)	Level 5 (μ + 3σ)
Seat spacing	1900	3800	6800	9900	12,900
In-train cleanness	1300	2700	6600	10,500	14,400
Seat availability	3100	6200	13,100	20,100	27,000
Free or cheap Wi-Fi	800	1600	4100	6600	9100
Plug sockets for laptops or mobiles	700	1400	4200	7100	9900
Seating direction	700	1400	3400	5500	7500
Temperature control	1000	2000	5300	8500	11,700
Seating layout	1700	3300	7500	11,700	15,900
Mechanical vibration	1000	2000	4600	7100	9700
Man-made noise	1200	2500	6200	9800	13,500
Seat-back screen	1300	2500	9100	15,700	22,400
Overall	14,700	22,000	29,300	44,600	59,900

Notes: The values are expressed in Korean won where 1000 wons approximately equate to one USD. μ and σ represent the mean and standard deviation for the amount of fare-increase calculated from the result of preliminary survey.

In the actual survey, the set of levels was replicated to check variability and consistency in answers. Each respondent was then given a randomly selected questionnaire. Blocking, in the form of on/off peak spreading, was also considered to remove the effect of nuisance factors.

The methodology of survey adopted in this paper belongs to the stated preference technique, and hence biases such as hypotheticality, neutrality, and judgment are inevitable during the interview. However, the propensity to which respondents strategically behave to achieve a more desired outcome must be controlled. This issue has also been discussed in the valuation of rail quality of service (Lu, Fowkes, and Wardman 2008; Wardman and Whelan 2001). A useful measure for relaxation is to keep the property of incentive compatibility (Carson and Groves 2007). Since the valuation in this study deals with a public good with coercive payment, the responses are incentive compatible. This means that the problem of strategic responses does not seriously affect the result.

Valuation

Binary logit model

Respondents were asked to select the preferred option between two discrete alternatives in the experiment of contingent choices. The binary logit model is thus a reasonable framework to deal with the data collected; note that most studies in Table 2 have also adopted the logit-based random utility model.

Model of service attribute

Consider a service attribute that is associated with the travel option m as faced by the decision-maker n, namely . This attribute is an element that forms a service bundle . The utility of alternative m with service attribute for individual n, or , is given by(1)

where and are the deterministic and stochastic components of the utility, respectively; represents fares of alternative m perceived by n; and is the vector of parameters that should be estimated.

Two comments on Equation (1(1) ) are added. First, is a dummy variable. It takes value 1 for the improved rolling stock and 0 otherwise. The variable can be understood as an alternative specific constant. All else being equal, the constant in the logit model reflects the mean of the difference between the error terms, that is, the gap in the utility between the alternatives. Two options in the choice exercise were only disparate in the service attribute. Thus, the utility difference is caused by the service factor. The estimate for this dummy hence expresses the service improvement. Second, in general, when a new variable is added to the utility function, an interaction effect with the existing factors can appear. For example, the confounding effect between the variables of quality of service and travel costs may be expected. The effect is not considered in this section, but will be discussed more in details at conclusion.

The linear-in-parameters utilities are popularly applied to guarantee the global solution for parameters.(2)

The values of service attribute () are then calculated by the marginal rate of substitution.(3)

The utility function discussed thus far just contained fares and service element. As mentioned, this specification is useful for placing a clear emphasis on the service attribute of rolling stocks. However, the benefit of train improvements in terms of quality of service can vary according to the travelers’ characteristics. Changes in the specification of model can represent this variation.(4)

where is the k^th socioeconomic feature of traveler n and β^k is the corresponding parameter.

β^k reflects the differences in preferences for train type as the function of . For any given individual, is in fact constant. It would make no sense to define these variables in both alternatives because all that matters is the effect of on the relative utility. Thus, is specified as an alternative specific variable in the utility function. As a result, the simple marginal rate of substitution between the variables of fare and attribute is the willingness-to-pay of the reference group. That is, the amount is only associated with the benchmark trip-makers who do not bear the set of features . However, the extended specification is useful in addressing the benefit by user group, specified by . As β^k/β^c is the marginal effect in willingness-to-pay of people associated with relative to the reference group, (β^k/β^c) × μ^k is the corresponding utility difference of the stratum to the reference group where μ^k is the mean of computed across the entire sample. Thus, the average money value for service attribute using the extended formulation is calculated by(5)

Model of service bundle

It would be more common that a few service attributes are improved together as a package rather than a single element is just upgraded. The deterministic utility for the case of service bundle is expressed by replacing with in Equations (2(2) ) and (4(4) ). The corresponding values of service bundle, or and , are also evaluated by substituting β^b for β^a in equations (3(3) ) and (5(5) ) where β^b is the parameter for .

The value for the improvement in each attribute in a package sums to an amount which is considerably different from the amount that the same respondent assigns to the bundle. Namely, there is an upper limit on how much people are willing to pay for the betterment no matter how many there are (Steer Davies Gleave 2000). The precise causes for the effect still remain unclear, even though interactions, budget constraints, halo effects, and the artificial nature of stated preference exercises are often discussed in the literature (Wardman and Whelan 2001). Treatment to mitigate the packaging effect has also not been established. The strategy adopted in this paper is, based on the assumption that the valuations of specific attributes are unbiased in relation to each other, to rescale the values so that they sum to a separately estimated amount of the entire bundle.(6)

where and are the values of a service attribute rescaled by the capping values and , respectively.

Results

Speciation and estimation

Two groups of binary logit models were specified (Tables 6 and 7) and estimated. One set was the models of 11 different attributes and the other was those of service bundle. The service bundle was the package that included all 11 elements. The bundle does not have to accommodate all 11 items, but can embrace some of them or more depending on research purposes. The models for specific attributes only considered the basic variables of fares and service elements. The bundle model, on the other hand, included the basic factors and also the socioeconomic covariates for travelers. This design does not mean that only the package model can have the socioeconomic variables, but should be understood as an illustration for incorporating covariates in the process of modeling. Two points about the bundle model to be noted are: first, some other variables such as gender and age were initially considered but were finally excluded because they did not show statistical significance; and second, the business–nonbusiness split of which market segmentation is useful to check the variations in behavior by trip purpose was also applied.

Table 6. Specification of the binary logit model for service attribute.

Variables	Improved rolling stocks	Unimproved rolling stocks
Fares	Current fares plus added amount given (1000 wons)	Current fares (1000 wons)
Service attribute	1	0

Table 7. Specification of the binary logit model for service bundle.

Variables	Improved rolling stocks	Unimproved rolling stocks
Fares for business trips	Current fares plus added amounts given for business trips (1000 wons); 0 otherwise	Current fares for business trips (1000 wons); 0 otherwise
Fares for non-business trips	Current fares plus added amounts given for non-business trips (1000 wons); 0 otherwise	Current fares for non-business trips (1000 wons); 0 otherwise
Service bundle	1	0
Travel time for business trips	In-vehicle time given for business trips (min); 0 otherwise	In-vehicle time given for business trips (min); 0 otherwise
Travel time for non-business trips	In-vehicle time given for non-business trips (min); 0 otherwise	In-vehicle time given for non-business trips (min); 0 otherwise
Number of uses	Number of rail uses per month	0
Occupation	1 if the respondent is a professional or employee in tertiary industries; 0 otherwise	0
Income	1 if the respondent’s monthly income is higher than 5,000,000 wons; 0 otherwise	0

Tables 8 show the 11 estimation results for the binary logit model of individual service attributes. Although they were simplified analyses just considering the basic specification, the 11 models show the reasonable goodness-of-fit. The formal significance indices of the likelihood ratio statistics indicate that the null hypothesis – that all the parameters are zero – can be rejected at the 0.01 level of significance. The values of ρ², though they are an informal goodness-of-fit index, are also at an empirically acceptable level. Not just the overall goodness-of-fit but also the reasonableness of each parameter should be checked. All the coefficients in the model estimates have the expected signs. That is, the coefficients for travel impedance and quality of service were negatively and positively estimated, respectively. The t-statistics for independent variables across all 11 models also show the satisfactory precision of estimated coefficients.

Table 8. Estimation results for binary logit model of service attribute.

Variables	Seat spacing	In-train cleanness	Seat availability	Free or cheap Wi-Fi	Plug sockets for laptops or mobiles
Fares	−0.12552***	−0.12266***	−0.08069***	−0.09098***	−0.12485***
Service attribute	1.058185***	1.548122***	1.903760***	0.427638**	0.843513***
Summary statistics
Number of observations	368	368	368	368	368
	24.41440***	60.13960***	78.46800***	7.21480***	22.33340***
ρ^2	0.04786	0.11788	0.15400	0.01414	0.04378
Variables	Seating direction	Temperature control	Seating layout	Mechanical vibration	Man-made noise	Seat-back screen
Fares	−0.16535***	−0.10851***	−0.14005***	−0.20613***	−0.13099***	−0.10273***
Service attribute	0.760024***	0.902360***	0.616091***	1.198647***	1.559619***	0.414546**
Summary statistics
Number of observations	368	368	368	368	368	368
	17.37520***	22.57380***	61.24460***	40.03560***	64.17460***	68.35860***
ρ^2	0.03406	0.04425	0.12005	0.07848	0.12579	0.13400

Notes: L(0) is the value of the log likelihood function when all parameters are zero and is that at its maximum; and .

^** 5% significance level

^*** 1% significance level.

In Table 9, the estimation result for the model of service bundle is given. The two results also show the satisfactory goodness-of-fit in terms of the sign and variance of variables and the entire significance of the model in both the formal and informal measures. The extended model, however, performs slightly better. A remarkable observation is the minus coefficient for the variable of number of uses for high-speed rail. The result is read as frequent users are relatively not very keen on the improvements in quality of service. The coefficient of this variable can logically take either sign. However, the finding would mean that the critical determinant for existing users to choose travel modes is not the service attributes that are dealt with in this study but other properties such as travel time savings for spatial interactions. In contrast, the service betterment would be appealing to attract new patronage since the negative sign can indirectly refer to the positive impact on non-frequent riders.

Table 9. Estimation results for binary logit model of service bundle.

Variables	Basic binary logit model	Extended binary logit model
Fares for business trips	−0.06347***	−0.06989***
Fares for nonbusiness trips	−0.08136***	−0.08052***
Service bundle	1.29201***	0.80295**
Travel time for business trips		−0.02417***
Travel time for non-business trips		−0.01412**
Number of uses		−0.12906*
Occupation		0.75776**
Income		0.80714***
Summary statistics
Number of observations	335	335
	150.2572***	165.107***
ρ^2	0.32355	0.35552

^* 10% significance level

^** 5% significance level

^*** 1% significance level.

Values

Table 10 summarizes the values of travel time and service bundle computed from the parameters estimated in this study. The weighted averages for the values of time and quality of service are also given in the table. The official share in trip purposes of Korea (Korea Ministry of Land, Infrastructure and Transport 2013) was applied to this computation; the ratio is business 38.4 and nonbusiness 61.6%, respectively. The values of travel time for business and non-business trips estimated by this study is relatively less and greater than those by the recommendations of the Korean government.

Table 10. Values of travel time and service bundle (2014 prices).

	Basic binary logit model			Extended binary logit model
	Business	Nonbusiness	Average	Business	Nonbusiness	Average
Value of travel time (wons/hour)
Estimates of this research				20,754	10,524	14,452
Recommendations of the Korean government (2011 prices)				20,718	5621	11,419
Value of service bundle (wons/trip)
Reference group				11,489	9972
Relative difference in willingness-to-pay
Who use high-speed rail				−3697	−1997
Who are professionals or employee in tertiary industries				8443	3358
Whose monthly income is over five million wons				4733	3765
Overall	20,358	15,881	17,600	20,968	15,097	17,351

Several comments are given on the results in Table 10. First, and show very close amounts. This outcome can indicate the robustness of the estimates. Second, journey purpose has an expected influence on the values while the gap is narrower in the monetary values for quality of service. However, the absolute difference in the value of time gets bigger in the long-distance travels whereas the divergence becomes smaller in the short-distance trips. This is because the service quality is measured by the unit of trip, while the total money values of time is calculated by multiplying travel times consumed. Third, the extended model can supply the user specific amounts of willingness-to-pay, applying the corresponding mean of samples shown in Table 11. Diverse combinations in addressing willingness-to-pay can be explored. The highest group in business trips, for example, is professionals or employee in tertiary industries whose monthly income is over five million wons. Their amounts reach 24,665 wons. In contrast, trip-makers who frequently use high-speed rail, whose jobs are beyond professionals or employee in tertiary industries, and whose monthly income is below five million wons are only willing to pay 7792 wons on train improvement. This value is around 30% of that of the highest respondents.

Table 11. Sample mean by socioeconomic characteristic.

	Business trips	Nonbusiness trips
Number of high-speed rail uses per month	2.00205	1.24615
Ratio of travelers whose occupations are professionals or employee in tertiary industries	0.77869	0.35681
Ratio of travelers whose monthly income is over five million wons	0.40984	0.37559

The valuations on the individual attributes are arranged in Table 12. The table contains the set of monetary quantities in both per trip and per passenger-kilometer. The unscaled per trip values were obtained from the estimates in Table 8 applying Equation (3(3) ). The weighted average of in Table 10, or 17,600 wons, was used as the capping value to evaluate the scaled amounts; note that the business–nonbusiness segmentation was designed to the bundle model, while the service attribute models were based on total trips. can also be applied in this calculation without any loss of generality. Another weighted average in Table 13 representing the mean distance traveled, which was computed using the responses of samples, was applied for evaluating the value per passenger-kilometer.

Table 12. Values of service attributes (2014 prices in wons).

Service attribute	Values per trip (unscaled)	Values per trip (scaled)	Values per passenger-kilometer
Seat spacing	8430	1559	6.52
In-train cleanliness	12,621	2334	9.77
Seat availability	23,594	4363	18.26
Free or cheap Wi-Fi	4700	869	3.64
Plug sockets for laptops or mobiles	6756	1249	5.23
Seating direction	4597	850	3.56
Temperature control	8316	1538	6.44
Seating layout	4399	814	3.40
Mechanical vibration	5815	1075	4.50
Man-made noise	11,906	2202	9.21
Seat-back screen	4035	746	3.12
Total	95,171	17,600

Table 13. Calculation of average distance traveled.

Origin–destination	Distance in kilometers	Number of samples	Passenger-kilometers
Seoul–Cheonan	96.0	91	8736
Seoul–Daejeon	159.8	78	12,464.4
Seoul–East Daegu	293.1	88	25,792.8
Seoul–Busan	423.8	78	33,056.4
Weighted average	239.0

Two important findings are noted in the result of Table 12. Above all, the capping value is just 18% of the simple sum of the individual worth unscaled. This gap demonstrates the need for the maximum willingness-to-pay in the stated preference interview. Subsequently, the attributes in the category of comfort show higher values than those of tangible and convenience defined in Table 1. This consequence is consistent with the outcome of the survey for attribute selection. The priority hence can be given to the group of comfort in quality of service when an operator of railways supplies the improvement of rolling stocks.

Policy implications

This section illustrates two policy implications from the results of this paper. They are all related to price discrimination. One is by the difference in travel times and the other is that in quality of service.

Premium fares and discounted charges by travel time

High-speed rail service in Korea is supplied by the very similar level of ticket prices, even though significant differences in travel time are posted in the time-table. The gap in travel time is mainly attributed to the different portions of usages for the high-speed tracks. Namely, some trains still run a considerable part of their route on the conventional tracks upgraded (Chang and Lee 2008). The diverse skip-stop operation is another cause for the difference. This study has proposed premium fares and discounted charges for services requiring longer and shorter travel times relative to those of most frequent operation, respectively. The cases of the longest and shortest travel times by origin-destination pair are exemplified in Table 14. The weighted value of travel time shown in Table 10 was applied to this evaluation.

Table 14. Price discrimination by travel time.

Origin–destination	Travel time	Changes in fares (2014 prices in wons)	Remarks
Seoul–Cheonan	0 h 33 min	1129	Premium
	0 h 38 min	0	Reference
Seoul–Daejeon	0 h 54 min	1355	Premium
	1 h 00 min	0	Reference
	1 h 36 min	−8128	Discount
Seoul–East Daegu	1 h 44 min	1129	Premium
	1 h 49 min	0	Reference
	2 h 27 min	−8579	Discount
Seoul–Busan	2 h 17 min	4967	Premium
	2 h 39 min	0	Reference
	3 h 21 min	−9482	Discount

Note: Only two different types of travel times are supplied to the origin–destination pair Seoul–Cheonan.

The reasonableness of first-class fares

Currently, first-class service is available on four of the 18 coaches in Korean high-speed rail. Six of the 11 attributes valued in this paper are provided: seat spacing, in-train cleanliness, plug sockets for laptops or mobiles, seating direction, seating layout, and man-made noise. Table 15 compares the reasonableness of first-class fares with respect to the willingness-to-pay of travelers. People indeed have the intention of paying extra money for first-class service. The amount however is below the price levied. Thus, it can be suggested that first-class service can be expanded subject to lowering the premium. The financial viability should also be checked before the expansion. This is because the improvement itself needs further investment costs and may reduce ridership due to the increases in fares.

Table 15. Willingness-to-pay on first-class service (2014 prices in wons).

Origin–destination	Current premium	Willingness-to-pay
Seoul–Cheonan	5200	3619
Seoul–Daejeon	8800	6024
Seoul–East Daegu	15,800	11,049
Seoul–Busan	21,300	15,976

Conclusion

This study was conducted to monetize the improvement on quality of service for intercity rail travel in Korea. The survey for attribute selection identified 11 most preferred factors for quality of service. Travelers’ preferences on the betterment of rolling stocks were collected through the application of choice experiments. The data were applied into the binary logit model for train type choices. The values in both individual attributes and service bundle were estimated. In particular, the maximum willingness-to-pay was considered as the capping value for scaling the individual money amounts. Policy implications in terms of price discrimination were also investigated. These suggestions can demonstrate the reasonableness and flexibility of the result by this paper.

This paper can be a useful reference for pertinent studies in general as well as the pricing policy of rail operators. However, greater examination on travelers’ taste variations can provide more opportunity to appreciate the values of transit quality of service. This is simply because the value or importance that individuals place on each attribute is expected to vary depending on the decision-maker. This study incorporated market segmentation in trip purpose, but the trial could partly capture the heterogeneity. The limitation was resulted from the methodology adopted in this research. The survey and theory of the binary logit model are discrete nature-based. The estimated model however implicitly assumed the same model structure and values of the unobserved parameters for all members of population represented by the sample collected. Two approaches can be found in the literature (e.g. Train 2009, 42–45) to account for taste variations. First, taste variations that vary systematically with respect to observed variables can be incorporated in the logit model. For example, willingness-to-pay on the service improvement may depend on travel time spent. This can be addressed by adding an interaction term as the product of the variable for service feature as well as for travel time. In contrast, heterogeneity that varies with unobserved variables cannot be handled in the standard logit framework. Mixed logit models in this case can be applied to capture the random taste variations. These explorations are thus the useful themes for follow-up studies.

Disclosure statement

No potential conflict of interest was reported by the authors.

Acknowledgement

The authors are grateful to Mr. Tufail Ahmad for his useful comments for this study.

Appendix

Contingent choices conducted in this study

Just one combination of fares for the alternatives is noted as an illustration (see Figures 2–11).

Figure 2. In-train cleanness.

Figure 3. Seat availability.

Figure 4. Free or cheap Wi-Fi.

Figure 5. Plug sockets for laptops or mobiles.

Figure 6. Seating direction.

Figure 7. Temperature control.

Figure 8. Seating layout.

Figure 9. Mechanical vibration.

Figure 10. Man-made noise.

Figure 11. Seat-back screen.