Applying discrete choice models to understand sport tourists’ heterogeneous preferences for Winter Olympic travel products

Abstract

This study aims to better understand how South Korean sport tourists show their preferences for Winter Olympic travel products. Comparing a series of discrete choice model (DCM) techniques, this study also investigates what attributes comprising PyeongChang Winter Olympic travel products are considered by sport tourists important and how they make trade-offs among those attributes for purchasing an optimal product. Study findings suggest that respondents place distinctive importance on each attribute to yield their greatest amount of utility through jointly considering a set of Olympic travel product attributes. Results also indicate that sport tourists put high willingness-to-pay values for multiple opportunities of attending skating competitions and popular tournaments while they show strong interests in more sightseeing and recreational experiences. Based on different DCM estimation results, several management strategies are presented for developing quality Olympic travel products.

Keywords

discrete choice model sport tourist travel product Winter Olympic Games

Introduction

Hosting mega sport events is believed to be a valuable opportunity for dramatically enhancing the host region’s reputation as an established international tourism destination (Fourie and Spronk, 2011; Li and Kaplanidou, 2013). The quality destination images distributed via different worldwide media channels promotes international tourists to visit the host sites (Kaplanidou and Vogt, 2007). For the immediate success of event operations, however, organizing officials are more interested in how much domestic sport tourists support those events by means of voluntarily visiting the host destination for the limited event duration rather than how much foreign money is obtained from international tourists after the events (Weed and Bull, 2009). In other words, the extent to which domestic sport tourists attend sporting venues can become a critical factor that determines whether the world-class event is successful or not (Hinch and Higham, 2004). This is simply because a larger proportion of stadium stands is filled with domestic tourists who show their strong willingness to purchase different travel products purposefully tailored for those mega sport events (Masterman, 2009).

Similar to any other ordinary consumers with budget and time constraints, sport tourists make decisions to choose an optimal travel product which likely results in the greatest amount of utility or satisfaction. To reach the uppermost limit of travel satisfaction, tourists tend to collectively consider different travel product attributes and make complicated trade-offs among those product characteristics (Lyu and Han, 2017). Accordingly, sport tourism businesses need to better understand what types of attributes comprising mega sport event travel products are regarded fundamental by their customers and how much they are willing to pay for particular products (Murphy et al., 2000). Mega sport event organizers also need to be better aware of how sport tourists hold different preference patterns to deliver optimal travel products to each targeted market segment (Kim and Chalip, 2004).

It is widely recognized that product is the first and most important component constituting the traditional marketing mix. Kotler et al. (2006) defined products as bundles offered to a market for attention, acquisition, use, or consumption, which involve various intangible elements like services, persons, places, organizations, and ideas, as well as physical objects. Lewis and Chambers (1999) divided products into three different categories: core, formal, and augmented products. Core products are associated with a variety of basic benefits or utilities of particular products which a consumer normally pursues; formal products encompass a combination of product attributes substantiating the relevant core products. Several benefits from travel products including relieving stress and experiencing novelty are believed as a typical element of core products, for example, while most package travel programs comprised of important attributes like accommodations, foods, and transportation can be classified into a type of formal products. Augmented products represent a mixture of different value-added features and benefits offered by product suppliers such as product warranty and after-purchase services.

Medlik and Middleton (1973) noted in their seminal work that travel product characterizes an amalgamation of multiple attributes including activities, amenities, and benefits that comprise the complete tourism experience. Many authors (e.g. Lehto et al., 2004; Murphy et al., 2000) have provided useful information regarding a range of product attributes which vary according to different travel typologies. Most typically, Oh et al. (1995) argued that a variety of consumptive services and hospitality experiences such as transportation, accommodations, and restaurants lie within the territory of the basic attributes consisting of general package travel products.

A handful of research efforts have been made to examine different elements that comprise particular travel product bundles customized for mega sport events. In conjunction with essential attributes of package travel products including accommodations and transportation, an effective bundling of sport event travel products embraces multiple attributes to meet the distinct needs for event spectatorship. Most sport event tourists, as a typical segment of passive sport tourists, tend to express their strong preferences for impressive experiences of attending popular tournaments and watching sporting spectacle on the spot (Gammon and Robinson, 1997; Gibson et al., 2003). In particular, the nationalistic fandom significantly affects sport tourists’ decisions for purchasing mega event travel products since they are eager to share unique subculture with others (Green, 2001). Weed and Bull (2009) similarly denoted that sport tourists visiting world-class sport event sites generally indicate stronger preferences for attending different venues where their national players or teams are expected to win medals.

Sport tourists tend to reveal a distinctive preference structure associated with “associated experiences,” which are sought to attain their higher levels of travel enjoyment (Weed, 2005). While addressing the “après ski” phenomenon as a case of these experiences pursued by ski tourists, Hudson (2000) noted that a large percentage of ski resorts have attempted to offer a wide range of recreation opportunities which may not be directly related to traditional skiing activities. Recreation participation in snow festivals and dogsledding events can be examples of these associated experiences. Sport tourists also showed their willingness to engage in a variety of recreational and cultural activities at or near travel destinations (Weed and Bull, 2009). García (2001) found that a large percentage of sport tourists visiting the 2000 Sydney Olympics purchased a travel product option for experiencing different types of cultural and recreational programs offered by host tourism organizations. In a similar vein, Kim and Chalip (2004) noted that international sport tourists indicated significant positive attitudes toward visiting diverse 2002 FIFA World Cup venues, provided that they had more opportunities to learn about Korean cultures during their trip.

The main purpose of this study is to provide useful opportunities for an improved awareness of sport tourists’ decisions for choosing mega sport event travel products and their heterogeneous preference patterns for those products. We also aim to identify particular attributes that are considered by South Korean sport tourists as important criteria for purchasing travel products specifically customized for 2018 PyeongChang Winter Olympic Games. To expand our knowledge on sport tourists’ inherent preference structures, we employ several discrete choice model (DCM) approaches—conditional logit (CL) models, random parameter logit (RPL) models, and latent class (LC) models (Hensher et al., 2005). Based on study findings, we will suggest practical strategies that help sport tourism businesses meet their customers’ needs and develop quality mega sport event travel products.

Methods

Discrete choice models

Consumers fail to maximize their utilities in real markets as a result of insufficient information as well as unobservable characteristics of alternative choices (Hensher et al., 2005). Random utility theory allows economists to incorporate these intangible errors in the utility function (McFadden, 1974). More explicitly, this theory has a premise that consumer utility is comprised of two different elements: systematic and random (Louviere et al., 2000). The latent utility U_ij for product j that consumer i obtains can be shown as

U_{i j} = V_{i j} + ε_{i j} = β X_{i j} + ε_{i j},

where V_ij represents the systematic element of utility assumed a linear index of several product attributes and ε_ij is the unobservable random error term. X_ij characterizes a mixture of different levels of each attribute; β is the vector of parameters related to X_ij.

With an assumption that the unobservable error terms are independently and identically Gumbel-distributed (IID), the predicted probability of choosing product j can be estimated using the CL model (McFadden, 1974), which is presented as

P_{i j} = \frac{exp (β X_{i j})}{\sum_{k \in C_{i}} exp (β X_{i k})} .

An important requirement of the CL model involves the fulfillment of the independence of irrelevant alternatives (IIA) property, which indicates that “the ratio of the probabilities of any two alternatives should be preserved despite the presence or absence of any other alternative within the set of alternatives” (Hensher et al., 2005: 519). However, the CL models tend to violate this restrictive IIA property in most choice situations where close substitutes are inevitably used for the choice alternatives.

Personal characteristics are included in the indirect utility functions to remedy the critical limitation of the CL models which assume respondents’ homogeneous preference structure (Louviere et al., 2000). Incorporating various sociodemographic features can be helpful to identify the source of heterogeneous preferences in the sampled population. The alternative specific constants (ASCs) are often applied to develop a series of interaction terms with sociodemographic variables. The extended CL model with interaction terms can be shown as $U_{i j} = V_{i j} + ε_{i j} = β_{n} X_{i j} + ε_{i j}$ , where β_n represents each parameter to be estimated for different attributes as well as a series of interaction terms between ASCs and sociodemographic factors.

A useful way for avoiding the IIA violation is the application of the RPL models (McFadden and Train, 2000). This model is free from the nonnegligible assumption because it takes the unobservable heterogeneity in the systematic element of the indirect utility function into account by treating parameter estimates of each attribute as random variables (Train, 2003). With a particular distributional assumption of each parameter estimate of random variables, the RPL model also provides information about whether respondents hold identical preferences for certain attributes (Hensher and Greene, 2003). The indirect utility function of this model can be expressed as $U_{i j} = V_{i j} + ε_{i j} = β_{h} X_{i j} + ε_{i j}$ , where $β_{h} = β_{m} + ν_{i}$ and $ν_{i} \sim N (0, \sum β_{m})$ ; β_m represents the mean values of parameter estimates and ν_i indicates the stochastic deviation of consumer preferences. With an assumption of ε_ij following the IID extreme value type I, the predicted probability of choosing product j can be shown as $L_{i j} = \frac{exp (β_{h} X_{i j})}{\sum_{k} exp (β_{h} X_{i k})}$ . Because β_h includes unknown random parameters, this equation cannot estimate the choice probability. The unconditional probability is integrated over all values of β_h using the maximum likelihood estimation like this equation, $P_{i j} = \int (\frac{exp (β_{h} X_{i j})}{\sum_{k} exp (β_{h} X_{i k})}) f (β | θ) d β .$ The probability can also be estimated by way of employing a simulation technique for given values of θ.

The LC model is an alternative strategy for relaxing the stringent IIA requirement. Unlike the RPL model, the LC model allows researcher to recognize how many market segments sharing homogeneous preferences exist by taking statistical criteria into account (Shen, 2009). Given that consumer i is a member of class s, the indirect utility function can be expressed as $U_{i j}_{| s} = V_{i j} + ε_{i j} = β_{s} X_{i j} + ε_{i j | s}$ and the predicted probability can be written as $P_{i | s (j)} = \frac{exp (β_{s} X_{i j})}{\sum_{k \in c_{i}} exp (β_{s} X_{i k})}$ , where β_s represents segment-specific parameter estimates. Beside the indirect utility functions, the LC model encompasses a set of class membership functions largely comprised of sociodemographic and attitudinal variables. The inclusion of those factors into the functions aims to assess their effects on the segment membership. The probability of becoming a member of segment s can be expressed as $P_{i | s} = \frac{exp (λ_{s} Z_{i})}{\sum_{s = 1}^{S} exp (λ_{s} Z_{i})}$ , where Z_i denotes a vector of sociodemographic features; λ_s, a vector of parameter estimates. Taken together, the joint probability can be shown using this equation, $P_{i j | s} = \sum_{s = 1}^{S} [\frac{exp (β_{s} X_{i j})}{\sum_{k = 1}^{K} exp (β_{s} X_{i k})}] [\frac{exp (λ_{s} Z_{i})}{\sum_{s = 1}^{S} exp (λ_{s} Z_{i})}] .$

Data collection

South Korean ski and snowboard tourists were purposefully selected as our study population since the winter sports participants are believed to indicate their stronger intentions to acquire a travel product for attending the 2018 PyeongChang Winter Olympic Games. With an aid of a South Korean online survey company, we randomly chose a sample comprised of 2000 respondents from slightly less than one million panel participants. We sent out our survey invitations at the end of March 2014. Of 1982 survey participants, 394 respondents reported their experiences of skiing or snowboarding trips during the past 12 months. After dropping 24 respondents who indicated their unwillingness to attend the Winter Olympic Games and 8 cases with insincere information, we included 362 surveys in the final data set.

Experiment design

Identifying appropriate choice attributes as well as allocating relevant levels for each attribute is the first DCM exercise to accurately understand respondents’ inherent tastes for particular hypothetical products (Lyu and Han, 2017). A review of earlier studies (e.g. Chalip and Leyns, 2002; Kim and Chalip, 2004) yielded eight different attributes comprising the hypothetical Olympic travel products. We also allotted three levels to each product attribute. The details about attributes and levels used for this study are presented in Table 1.

Table 1.

Detailed information about attributes and levels.

Attributes	Description	Levels (expected coefficient sign)
Transportation (TRANSPORTATION)	Types of transportation that a tourist moves to the host site and uses at the destination	Car driving* Tour bus (+) Express train + Tour bus (+)
Accommodation (ACCOMMODATIONS)	Types of accommodation facilities that a tourist uses at the destination	Homestay or guest houses* Condominiums (+) High-class resorts or hotels (+)
Meals offered (MEALS)	Types of meals that a tourist wants to be served during the trip	Breakfast* Breakfast + Dinner (+) Breakfast + Lunch + Dinner (+)
Spectating Korean team competitions (KOREAN TEAMS)	Attendance opportunities for skating tournaments that Korean players are expected to win gold medals	None* Speed skating only (+) Speed + Short-track skating (+)
Attending popular tournaments (POPULAR EVENTS)	Attendance opportunities for popular tournaments and events such as opening and closing ceremony, ice hockey, figure skating, etc.	Low* Medium (+) High (+)
Sightseeing programs (SIGHTSEEING PROGRAMS)	The number of programs that a tourist sightsees several attractions at or near the destination	None* 1 time (+) 2 times (+)
Skiing and snowboarding opportunities (SKIING/SNOWBOADING)	The number of programs that a tourist participates in skiing and snowboarding at or near the destination	None* 1 time (+) 2 times (+)
Product price (PRICE)	Price for a 5-day travel product	US$800* US$1000 (−) US$1200 (−)

* Base levels.

Given three levels of each attribute, a full factorial design results in an uncontrollable number of alternatives. Therefore, this study utilized a fractional factorial design with main effects to develop a reasonable number of paired comparisons. Despite the partial loss of statistical information, the fractional factorial design ensures the nature of orthogonality, suggesting that the variations of each level are uncorrelated. Using the SAS 9.2 macro recommended by Kuhfeld (2005), we randomly yielded 54 choice comparisons. To relieve respondents’ cognitive burden, we additionally employed a blocking strategy to split the paired choice sets into nine different questionnaire versions. Thus, only six paired comparisons were presented to each respondent.

We incorporated a non-choice alternative in all paired comparisons in that many people do not purchase any product in actual choice situations. Stated differently, a non-choice option (i.e. “I would not choose either product”) was employed plus the two different choice alternatives (i.e. “Travel product A” and “Travel product B”). We utilized different types of pictograms to improve respondents’ comprehension of complicated choice situations. Bateman et al. (2009) noted that the application of visualization techniques is beneficial in reducing response fatigue and decreasing reliance upon response heuristics. Figure 1 illustrates a choice comparison used for this study.

Figure 1.

An example of paired choice sets.

Results

Sample description

More than half of the 362 effective survey participants were males (57.2%). The average age of respondents was 37.4 years, while nearly 4 out of 10 were in their 30s (37.0%), followed by their 40s (29.9%), 20s (21.8%), and 50s (11.3%). Almost 8 out of 10 skiers and snowboarders (80.7%) reported being college or university graduates. The greatest portion of our respondents (16.3%) also exhibited their monthly household after-tax income range from US$5000 to US$6000, followed by the ranges of US$4000 to US$5000 (14.9%) and US$7000 to US$8000 (14.9%).

CL and extended CL model

While the standard CL models mostly violate the IIA requirement, the models are estimated to succinctly delineate respondents’ average preference structures. Since the CL model is based on an assumption of homogeneous preference patterns, a set of sociodemographic variables are often incorporated in the model using interaction terms with ASCs to capture the distinctive effects of personal characteristics on choice behaviors. A series of parameters in the deterministic component of both CL model and extended CL model with interactions were estimated by utilizing a maximum likelihood estimation technique. The results of the CL and the extended CL model estimations are illustrated in Table 2. Among the eight different attributes used for this study, seven attributes (i.e. ACCOMMODATIONS, MEALS, KOREAN TEAMS, POPULAR EVENTS, SIGHTSEEING PROGRAMS, SKIING/SNOWBOARDING, and PRICE) showed at least partially significant in the parameter coefficients of the two CL-related models. All coefficients revealed the same signs as our expectations with an exception of the “Speed skating only” level in the attribute of KOREAN TEAMS while they indicated an identical pattern of statistical significance across both models.

Table 2.

Results of the CL and the extended CL model estimations.

Attributes and levels		CL model		Extended CL model
Attributes and levels		Coef.	SE	Coef.	SE
ASC		3.4620***	0.2097	2.0648***	0.0002
TRANSPORTATION	Tour bus	0.0243	0.0436	0.0244	0.5461
TRANSPORTATION	Express train + Tour bus	−0.0048	0.0436	−0.0046	0.0436
ACCOMMODATIONS	Condominiums	0.0368	0.0438	0.0367	0.0436
ACCOMMODATIONS	High-class resorts or hotels	0.1237***	0.0435	0.1239***	0.0438
MEALS	Breakfast + Dinner	0.0561	0.0435	0.0563	0.0435
MEALS	Breakfast + Lunch + Dinner	0.0908**	0.0435	0.0909**	0.0435
KOREAN TEAMS	Speed skating only	−0.1025**	0.0434	−0.1025**	0.0435
KOREAN TEAMS	Speed + Short-track skating	0.4687***	0.0435	0.4691***	0.0435
POPULAR EVENTS	Medium	0.0123	0.0435	0.0124	0.0435
POPULAR EVENTS	High	0.2642***	0.0434	0.2645***	0.0435
SIGHTSEEING PROGRAMS		0.1133***	0.0377	0.1135***	0.0434
SKIING/SNOWBOARDING		0.1762***	0.0378	0.1765***	0.0377
PRICE		−0.0020***	0.0002	−0.0020***	0.0378
AGE × ASC				0.0050	0.0086
GENDER × ASC				−0.2285	0.1544
INCOME × ASC				0.0643*	0.0342
EDUCATION × ASC				0.4653**	0.1909
Log likelihood		−2996.35		−2988.85
McFadden $ρ^{2}$		0.158		0.160

Note: CL: conditional logit; SE: standard error; ASC: alternative specific constant.

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

The highly significant positive signs of the ASC coefficients in both models denoted that our respondents preferred choosing an option from the two travel products over the non-choice alternative. However, the negative coefficient signs of PRICE attribute indicated respondents’ aversion to higher prices for the mega event travel products. The positive coefficient sign of the “High” level in POPULAR EVENTS attribute suggested that sport tourists were interested in Olympic travel products covering more attendance opportunities for popular events and tournaments. It may be notable that the two levels of KOREAN TEAMS attribute demonstrated an opposite coefficient signal direction across both models. The significant positive sign of the “Speed skating + Short-track skating” level showed respondents’ preferences for the extensive opportunities to attend the two competitions. Nevertheless, the negative signal of the “Speed skating only” level revealed their strong distaste for spectating the competition alone, which is different from our previous expectations.

Respondents’ strong preferences for more opportunities to go sightseeing to diverse tourist attractions at or near the host site were substantiated through the significant positive coefficient sign on the attribute of SIGHTSEEING PROGRAMS. An identical preference pattern was witnessed in SKIING/SNOWBOARDING attribute. In other words, sport tourists were more in favor of frequent participation in their preferred activities during the Olympic travel period. The “High-class resorts or hotels” level in the attribute of ACCOMMODATIONS exhibited a positive coefficient sign, suggesting respondents’ preferences for the highest quality of accommodations. However, the attribute of TRANSPORTATION failed to show a significant result in both models, denoting that our respondents were insensitive to the means of transportation. Among four different interactions in the extended CL model, which provide an improved understanding of the effects between sociodemographic characteristics and ASC, two different interaction terms (i.e. INCOME × ASC and EDUCATION × ASC) were significant. The positive coefficient signs indicated that sport tourists were more likely to choose Olympic travel products as their income and education levels increased.

RPL and extended RPL model

Unlike the CL models that commonly contravene the IIA assumption in choice data, the RPL models are immune to the rigorous requirement because the models have an essential premise that respondents’ preferences for particular attributes are heterogeneous. The RPL models allow different parameters to be random variables and subsequently help researchers better understand the presence of preference heterogeneity (Train, 2003). For the stable parameterization of the RPL models, we employed the Halton Draws 500, which is the most popular maximum likelihood simulation estimation method (Bhat, 2001).

The results of the two different RPL-related model estimations are presented in Table 3. All significant coefficient signs representing the mean values of several parameter estimates were identical across both RPL models. The parameter estimates of the CL- and the RPL-related models revealed an analogous preference patterns while the absolute values of diverse significant parameters in the RPL-related models were greater compared to those in the CL-related models. Unlike the two CL-related models and the extended RPL model, the RPL model failed to indicate statistical significance for the level of “Breakfast + Lunch + Dinner” in MEALS attribute. This finding mirrors that our respondents were not sensitive to the specification of meals offered when choosing an Olympic travel product. We skip detailed explanations about other mean values of parameter estimates because the results of both CL- and RPL-related models can be interpreted in an identical manner.

Table 3.

Results of the RPL and the extended RPL model estimations.

Attributes and levels		RPL model		Extended RPL model
Attributes and levels		Coef.	SE	Coef.	SE
Mean parameters
ASC		6.8223***	0.5577	3.5958*	1.8598
TRANSPORTATION	Tour bus	−0.0162	0.0815	−0.0109	0.0820
TRANSPORTATION	Express train + Tour bus	−0.0212	0.0847	−0.0263	0.0855
ACCOMMODATIONS	Condominiums	0.1058	0.0757	0.1095	0.0763
ACCOMMODATIONS	High-class resorts or hotels	0.2405***	0.0757	0.2389***	0.0761
MEALS	Breakfast + Dinner	0.1054	0.0742	0.0998	0.0741
MEALS	Breakfast + Lunch + Dinner	0.1254	0.0855	0.1423*	0.0857
KOREAN TEAMS	Speed skating only	−0.1647**	0.0818	−0.1644**	0.0823
KOREAN TEAMS	Speed + Short-track skating	0.8241***	0.0873	0.8262***	0.0874
POPULAR EVENTS	Medium	0.0280	0.0769	0.0325	0.0779
POPULAR EVENTS	High	0.4751***	0.0829	0.4799***	0.0832
SIGHTSEEING PRGRAMS		0.2482***	0.0777	0.2436***	0.0783
SKIING/SNOWBOARDING		0.3395***	0.0687	0.3443***	0.0688
PRICE		−0.0038***	0.0004	−0.0039***	0.0004
ASC × AGE				0.0319	0.0320
ASC × GENDER				−0.4467	0.4868
ASC × INCOME				0.1254	0.1003
ASC × EDUCATION				0.7439	0.7742
Standard deviation parameters
ASC		3.2185***	0.3399	3.2199***	0.3592
TRANSPORTATION	Tour bus	0.6845***	0.1530	0.6810***	0.1521
TRANSPORTATION	Express train + Tour bus	0.7529***	0.1431	0.7776***	0.1407
ACCOMMODATIONS	Condominiums	0.2160	0.2058	0.2667	0.1856
ACCOMMODATIONS	High-class resorts or hotels	0.3375	0.2228	−0.3190*	0.1888
MEALS	Breakfast + Dinner	0.0142	0.1938	−0.0269	0.1855
MEALS	Breakfast + Lunch + Dinner	0.6535***	0.1366	0.6344***	0.1441
KOREAN TEAMS	Speed skating only	0.5231***	0.1573	0.5268***	0.1620
KOREAN TEAMS	Speed + Short-track skating	0.4679***	0.1575	0.4346**	0.1686
POPULAR EVENTS	Medium	0.4931***	0.1369	0.5184***	0.1364
POPULAR EVENTS	High	0.5168***	0.1416	0.5156***	0.1443
SIGHTSEEING PROGRAMS		0.5866***	0.1384	0.5910***	0.1504
SKIING/SNOWBOARDING		0.0155	0.3613	0.1012	0.2908
Log likelihood		−1695.49		−1692.28
McFadden $ρ^{2}$		0.129		0.130

Note: RPL: random parameter logit; CL: conditional logit; SE: standard error; ASC: alternative specific constant.

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

The standard deviation statistics of several random parameters showed a different pattern from the mean values of the estimated coefficients. The majority of standard deviation values were significant across both models. To be more specific, 9 out of 13 standard deviations of the random parameters were statistically significant in the RPL model, while 10 were significant in the extended RPL model. These results provide obvious evidence of respondents’ underlying heterogeneous preference structures concerning the Winter Olympic travel products. Nevertheless, the two RPL-related models failed to offer meaningful information on the heterogeneity source.

LC model

Since the results of the RPL-related model estimations revealed respondents’ heterogeneous preference, we estimated the LC model to more clearly elaborate the subgroup differences in preferences for the Olympic travel products. An important premise of the LC model is that consumers are comprised of several segments while each subgroup holds a homogeneous preference structure (Greene and Hensher, 2003). The segment classifications are endogenously made by identifying observable characteristics within the choice data, which often encompass respondents’ attitudes toward particular products and several sociodemographic features (Boxall and Adamowicz, 2002).

The optimal number of inherent classes embedded in the sample is commonly outlined by comparing multiple statistics such as log likelihood, Akaike Information Criterion (AIC), and Bayesian Information Criterion (BIC). Table 4 reports different useful statistics for the segment classification. The results of likelihood ratio (LR) tests suggested that there was a significant model improvement in the three- and four-class models. While the four-class model indicated the most desirable AIC statistic, the three-segment model showed the smallest BIC measure. Taken together, the parsimonious three-class model was chosen as the final LC model, which may provide more accurate information regarding respondents’ heterogeneous preferences for the Olympic travel products.

Table 4.

Statistical information for the number of latent classes.

Number of classes	Log likelihood	Number of parameters	AIC	BIC
1	−2996.35	14	6020.70	6115.64
2	−1742.35	33	3550.70	3679.12
3	−1679.97	52	3463.94	3666.31
4	−1643.17	71	3428.34	3704.65
5	−1626.30	90	3432.61	3782.86

Note: AIC: Akaike Information Criterion = −2(logL − K)/n; BIC: Bayesian Information Criterion = −(2logL − Klogn)/n, where K represents the number of parameters estimated.

The results of the LC model estimation (refer to Table 5) indicated that the largest portion of respondents fell within the class 2 (49.5%), followed by the class 1 (33.9%) and the class 3 (16.6%). Among a set of sociodemographic characteristics used to better understand the source of preference heterogeneity, the variable of INCOME was solely significant in the two class membership functions. In other words, the two segments of the classes 1 and 2 revealed a significant difference in household income level from the base group (i.e. class 3). It also seems to be interesting that the two coefficient signs of each class showed an opposite direction. This result suggests that respondents’ economic status played an important role in shaping their embedded preference hetergoneity. Nevertheless, other sociodemographic variables failed to reveal any statistical significance in the class membership functions.

Table 5.

Results of LC model estimations.

Attributes and levels		Class 1		Class 2		Class 3
Attributes and levels		Coef.	SE	Coef.	SE	Coef.	SE
Utility function
ASC		10.906***	1.506	2.421***	0.567	1.360	0.871
TRANSPORTATION	Tour bus	0.295*	0.170	−0.079	0.078	−0.328**	0.160
TRANSPORTATION	Express train + Tour bus	−0.438**	0.187	0.343***	0.091	−0.429**	0.186
ACCOMMODATIONS	Condominiums	−0.197	0.213	0.041	0.080	0.368**	0.168
ACCOMMODATIONS	High-class resorts or hotels	0.377**	0.162	0.090	0.075	0.105	0.164
MEALS	Breakfast + Dinner	−0.052	0.179	−0.030	0.081	0.454***	0.161
MEALS	Breakfast + Lunch + Dinner	0.307	0.210	0.078	0.080	0.066	0.163
KOREAN TEAMS	Speed skating only	−0.439*	0.230	0.044	0.074	−0.369**	0.164
KOREAN TEAMS	Speed + Short-track skating	0.782***	0.173	0.564***	0.075	0.414***	0.159
POPULAR EVENTS	Medium	−0.033	0.209	-0.026	0.076	0.022	0.162
POPULAR EVENTS	High	1.000***	0.263	0.123	0.075	0.469***	0.155
SIGHTSEEING PROGRAMS		0.518**	0.213	0.016	0.065	0.108	0.154
SKIING/SNOWBOARDING		0.340**	0.154	0.124**	0.063	0.472***	0.159
PRICE		−0.009***	0.002	0.001	0.001	−0.003***	0.001
Class membership function
AGE		0.010	0.021	0.011	0.019	Fixed parameters
GENDER		−0.216	0.380	−0.164	0.351
INCOME		−0.171**	0.086	0.146*	0.080
EDUCATION		0.007	0.469	0.582	0.428
Constant		1.044	1.263	−1.228	1.205
Log likelihood		−1679.97
LC probability		0.339		0.495		0.166

Note: LC: latent class; SE: standard error; ASC: alternative specific constant.

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

The utility functions provided further persuasive evidence of the heterogeneous preference structures inherent in the three segments. Among different choice attributes, TRANSPORTATION may be worth noting in the light of the fact that each class revealed distinctive preferences for several modes of transportation. More specifically, sport tourists belonging to the class 3 exhibited strong aversion to using tour buses and/or express trains, whereas the class 2 members showed greater tastes for those means of transportation compared to the base level of “car driving.” Contrastingly, the class 1 members indicated their interest in tour buses while they were unwilling to use express trains and tour buses together.

Model comparison

To select a best-performing model, we compared multiple models elaborated above. An optimal model is normally chosen by way of intricate trade-offs among several standards associated with fit indices, model parsimony, and efficient interpretation (Cameron and Trivedi, 2010). For a better comparison of model fit, we employed three different criteria: log likelihood, AIC, and BIC statistics. Several fit indices to compare our competing models are presented in Table 6.

Table 6.

Model comparison.

	CL model	Extended CL model	RPL model	Extended RPL model	LC model
Log likelihood	−2996.35	−2988.85	−1695.49	−1692.28	−1679.97
Degree of freedom	14	18	27	31	52
AIC	6020.70	6013.69	3444.98	3446.56	3463.94
BIC	6115.64	6135.77	3628.1	3656.80	3666.31

Note: CL: conditional logit; RPL: random parameter logit; LC: latent class; AIC: Akaike Information Criterion; BIC: Bayesian Information Criterion.

We firstly conducted a series of LR tests to distinguish a better model from the nested models. We excluded the CL-related models from the comparison procedure because these models are known to violate the strict IIA assumption. The results of the LR tests indicated that the extended RPL approach using interaction terms failed to significantly improve the model fit over the RPL model. The RPL model with no interactions was also better when taking its model parsimony into account. We then compared the two non-nested models (i.e. the RPL model and the LC model) as recommended by Ben-Akiva and Swait (1986). The less statistics of AIC and BIC clearly suggested that the RPL model outperformed the LC model. As a result, we chose the RPL model as the final model for this study.

Marginal willingness-to-pay values

The most notable merit of the DCM approach involves the provision of richer information about the marginal willing-to-pay (MWTP) values or implicit prices, helpful in assessing the relative importance among product attributes and ranking various management policies (Blamey et al., 1999). The MWTP values can be calculated by $- \frac{β_{k}}{β_{PRICE}}$ , where $β_{k}$ indicates different coefficients of the non-price attributes k and $β_{PRICE}$ represents the coefficient of the PRICE attribute (Hanley et al., 1998). Table 7 demonstrates several MWTP values derived from each model. As this study utilized an effect coding strategy, the MWTP values of each base level can be computed by the negative sum of the estimated MWTP values of other levels. For example, the MWTP value on the “Homestay or guest houses” level of ACCOMMODATIONS attribute in the CL model can be derived as −(US$18.4 + US$61.9) = −US$80.3.

Table 7.

Results of MWTP value computations.

Attributes	Levels	CL model	Extended CL model	RPL model	Extended RPL model	LC model
Attributes	Levels	CL model	Extended CL model	RPL model	Extended RPL model	Class 1	Class 2	Class 3	Weighted
TRANSPORTATION	Tour bus	12.1	12.2	−4.2	−2.8	33.3	−394.2	−112.7	−202.5
TRANSPORTATION	Express train + Tour bus	−2.4	−2.3	−5.5	−6.8	−49.3	1713.2	−147.4	806.8
ACCOMMODATIONS	Condominiums	18.4	18.3	27.6	28.2	−22.2	204.4	126.7	114.7
ACCOMMODATIONS	High-class resorts or hotels	61.9	61.9	62.7	61.4	42.5	448.6	36.2	242.5
MEALS	Breakfast + Dinner	28.1	28.1	27.5	25.7	−5.8	−149.2	156.0	−49.9
MEALS	Breakfast + Lunch + Dinner	45.5	45.4	32.7	36.6	34.5	390.9	22.8	209.0
KOREAN TEAMS	Speed skating only	−51.3	−51.2	−43.0	−42.3	−49.5	219.2	−126.9	70.7
KOREAN TEAMS	Speed + Short-track skating	234.5	234.5	214.9	212.4	88.1	2818.8	142.4	1448.8
POPULAR EVENTS	Medium	6.2	6.2	7.3	8.3	−3.7	−127.7	7.5	−63.2
POPULAR EVENTS	High	132.2	132.2	123.9	123.4	112.6	612.5	161.4	368.1
SIGHTSEEING PROGRAMS		56.7	56.7	64.7	62.6	58.3	78.0	37.3	64.6
SKIING/SNOWBOARDING		88.1	88.2	88.6	88.5	38.3	618.0	162.4	345.8

Note: MWTP: marginal willing-to-pay; CL: conditional logit; RPL: random parameter logit; LC: latent class. Unit: US$.

Unlike the CL- and RPL-related models, the LC model seemed to overestimate the MWTP values, which suggest its unstable parameterization. Among several MWTP calculation results, this study focused on the values computed based on the RPL model because it was already chosen as our final model. The MWTP values indicated that our respondents put the heaviest weight on the attribute of KOREAN TEAMS in the light of the fact that this attribute revealed the largest amount of the MWTP values (i.e. US$214.9). Furthermore, sport tourists were willing to pay US$88.6 and US$64.7 more for purchasing an Olympic travel product given a marginal change in the attributes of SKIING/SNOWBOARDING and SIGHTSEEING PROGRAMS.

Discussion

Study implications

Through comparing several DCM algorithms, we found the superiority of the RPL model in addressing sport tourists’ preferences for mega sport event travel products. There is much effort to assess what types of discrete choice algorithms including the RPL models and the LC models perform better (e.g. Greene and Hensher, 2003; Shen, 2009). Findings from this study suggest that the RPL-related models can be more effective than other models in elaborating respondents’ inherent preference patterns in our choice data, which is consistent with different earlier studies (e.g. Cerwick et al., 2014; Ek and Persson, 2014). Furthermore, we disclosed potential sources of preference heterogeneity embedded in our sample consisting of sport tourists, which support the importance of various market segmentation strategies. While the LC model was not chosen as the final model, the membership functions indicated an important clue that respondents’ income level acted as a critical determinant in formulating their heterogeneous Olympic travel product preferences. This finding provides a practical marketing suggestion that travel product planners need to implement various market segmentation strategies developed based on their customers’ economic status.

The MWTP values provided useful opportunities for an improved awareness of the relative importance of different choice attributes comprising the hypothetical Olympic travel product. Respondents revealed their largest amount of the MWTP value on the attribute of KOREAN TEAMS. This result reflects that they put the heaviest weight on multiple opportunities for attending speed and short-track skating venues. This preference orientation may be attributed to respondents’ distinctive travel motivations for watching their national teams or players anticipated to win medals. As discussed Green and Chalip (1998), many mega sport event tourists tend to boast their innate patriotism through voluntarily supporting national teams at event venues.

A noteworthy finding can be the significant negative coefficient signs on the “Speed skating only” level in the attribute of KOREAN TEAMS, which was an opposite signal of the highest level of “Speed skating + Short-track skating.” This result described our respondents’ strong aversion to spectating speed skating tournaments only. The extraordinary preference pattern may result from the mediocre accomplishment of their national speed skating players during the 2014 Sochi Olympic Games. Nevertheless, respondents’ strong tastes for spectating the two skating competitions indicates their high expectations for the national short-track skating players that have won 21 Olympic gold medals since the 1992 Albertville Games.

Another interesting finding is that our respondents put the second heaviest weight on various opportunities for attending popular tournaments and events such as opening and closing ceremonies, ice hockey, and figure skating. This result is consistent with previous studies (e.g. Filo et al., 2013) demonstrating that most sport tourists tend to visit mega sport event destinations for spectating popular tournaments and experiencing sporting spectacle. These findings suggest useful managerial implications that tourism businesses need to include a variety of opportunities for attending popular competitions when designing mega sport event travel products. The active collaborations with event organizing committees in the earlier stage of travel product planning are also required to successfully acquire sufficient event and tournament tickets (Bramwell, 1997).

This study contributed to an improved insight into sport tourists’ unique preference structure closely related to their “associated experiences” (Weed, 2008). Results of the MWTP value calculation indicated that our respondents displayed their interests in more opportunities for sightseeing to different tourist attractions close to the Olympic hosting region. They showed their willingness to pay US$64.7 more, given that an additional sightseeing program was incorporated in the travel product packages. Respondents were also willing to pay US$88.6 more for one increased skiing or snowboarding opportunity at or near the destination. These results indicate that sport tourists were eager to engage in their favorite activities even while traveling a mega sport event destination, which may be clearly different from other types of tourists (Green and Chalip, 1998). These findings suggest an important management implication that adding more opportunities that promote sport tourists to visit tourist attractions and participate in preferred activities can be effective in enhancing overall quality of Olympic tourism experiences. Sport tourism businesses need to incorporate a variety of recreational experiences into their travel products to meet customer desires.

Results indicate that respondents revealed their strong preferences for the highest quality of accommodation facilities. This finding seems to be associated with the economic status of our sample in that a large portion of ski and snowboard tourists in South Korea was commonly classified into the wealthier class (Korea Ministry of Culture, Sports, and Tourism, 2018). This result also indicates that sport tourists placed more importance on such conspicuous consumption, which is dependent upon their prestige motivators (Weed, 2008). Accordingly, sport tourism businesses may benefit from combining luxurious accommodations in their mega sport event travel products. Nevertheless, both CL- and RPL-related models failed to denote significant coefficient signs on the attribute of TRANSPORTATION. This result suggests that sport tourists were not sensitive to several types of public transportation such as tour buses and express trains when choosing their preferred Olympic travel products. It may be because our respondents were accustomed to using their private vehicles for traveling to a domestic destination in everyday lives.

Study limitations and future studies

While the LC model was not chosen as the final model, we disclosed that respondents’ income level played a significant role in determining their heterogeneous preference styles. Incorporating a range of attitudinal characteristics into the model can be effective in identifying other source of preference heterogeneity. Consumer attitudes are known to significantly affect potential group diversity in general product tastes (Greene and Hensher, 2003).

Many previous studies utilizing the choice contexts have substantiated the methodological robustness. Nevertheless, special caveats are needed to interpret our study findings because we recruited the stated preference technique assuming that respondents’ preference formation can be portrayed through using the hypothetical choice situations. Further research applying other methods mixing revealed and stated preference approaches may be beneficial to rectify diverse limitations resulting from the hypothetical market settings. Caution is also required since we did not check the internal validity by using a series of additional choice questions, which is recommended by Green and Srinivasan (1990). Finally, this study is limited to generalize several findings to other populations due to its sample of ski and snowboard tourists in South Korea. Adapting our study context to international sport tourists may be helpful to enhance knowledge on their decision-making mechanisms for purchasing mega sport event travel products.

Footnotes

Declaration of conflicting interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This work was supported by a research grant from the College of Culture and Sports, Korea University.

ORCID iD

Seong Ok Lyu

References

Bateman

Day

Jones

, et al. (2009) Reducing gain–loss asymmetry: a virtual reality choice experiment valuing land use change. Journal of Environmental Economics and Management 58(1): 106–118.

Ben-Akiva

Swait

(1986) The Akaike likelihood ratio index. Transportation Science 20(2): 133–136.

Bhat

(2001) Quasi-random maximum simulated likelihood estimation of the mixed multinomial logit model. Transportation Research Part B: Methodological 35(7): 677–693.

Blamey

Gordon

Chapman

(1999) Choice modelling: assessing the environmental values of water supply options. Australian Journal of Agricultural and Resource Economics 43(3): 337–357.

Boxall

Adamowicz

(2002) Understanding heterogeneous preferences in random utility models: a latent class approach. Environmental and Resource Economics 23(4): 421–446.

Bramwell

(1997) Strategic planning before and after a mega-event. Tourism Management 18(3): 167–176.

Cameron

Trivedi

(2010) Microeconometrics Using STATA. College Station: STATA Press.

Cerwick

Gkritza

Shaheed

, et al. (2014) A comparison of the mixed logit and latent class methods for crash severity analysis. Analytic Methods in Accident Research 3: 11–27.

Chalip

Leyns

(2002) Local business leveraging of a sport event: managing an event for economic benefit. Journal of Sport Management 16: 132–158.

10.

Persson

(2014) Wind farms—Where and how to place them? A choice experiment approach to measure consumer preferences for characteristics of wind farm establishments in Sweden. Ecological Economics 105: 193–203.

11.

Filo

Chen

King

, et al. (2013) Sport tourists’ involvement with a destination: a stage-based examination. Journal of Hospitality & Tourism Research 37(1): 100–124.

12.

Fourie

Spronk

(2011) South African mega-sport events and their impact on tourism. Journal of Sport & Tourism 16(1): 75–97.

13.

Gammon

Robinson

(1997) Sport and tourism: a conceptual framework. Journal of Sport Tourism 4(3): 11–18.

14.

García

(2001) Enhancing sport marketing through cultural and arts programs: lessons from the Sydney 2000 Olympic arts festivals. Sport Management Review 4(2): 193–219.

15.

Gibson

Willming

Holdnak

(2003) Small-scale event sport tourism: fans as tourists. Tourism Management 24(2): 181–190.

16.

Green

(2001) Leveraging subculture and identity to promote sport events. Sport Management Review 4(1): 1–19.

17.

Green

Chalip

(1998) Sport tourism as the celebration of subculture. Annals of Tourism Research 25(2): 275–291.

18.

Green

Srinivasan

(1990) Conjoint analysis in marketing: new developments with implications for research and practice. Journal of Marketing 54(4): 3–19.

19.

Greene

Hensher

(2003) A latent class model for discrete choice analysis: contrasts with mixed logit. Transportation Research Part B: Methodological 37(8): 681–698.

20.

Hanley

Wright

Adamowicz

(1998) Using choice experiments to value the environment. Environmental and Resource Economics 11(3–4): 413–428.

21.

Hensher

Greene

(2003) The mixed logit model: the state of practice. Transportation 30(2): 133–176.

22.

Hensher

Rose

Greene

(2005) Applied Choice Analysis: A Primer. New York: Cambridge University Press.

23.

Hinch

Higham

JES

(2004) Sport Tourism Development. Clevedon: Channel View Publications.

24.

Hudson

(2000) Snow Business. London, UK: Cassell.

25.

Kaplanidou

Vogt

(2007) The interrelationship between sport event and destination image and sport tourists’ behaviours. Journal of Sport & Tourism 12(3-4): 183–206.

26.

Kim

N-S

Chalip

(2004) Why travel to the FIFA world cup? Effects of motives, background, interest, and constraints. Tourism Management 25(6): 695–707.

27.

Korea Ministry of Culture, Sports and Tourism (2018) 2017 White book of leisure in Korea. Available at: http://kostat.go.kr/portal/korea/kor_pi/8/6/2/index.board?bmode=read&aSeq=365695&pageNo=16&rowNum=10&amSeq=&sTarget=&sTxt= (accessed on 2 December, 2018).

28.

Kotler

Bowen

Makens

(2006) Marketing for Hospitality and Tourism. Upper Saddle River: Prentice Hall.

29.

Kuhfeld

(2005) Marketing Research Methods in SAS: Experimental Design, Efficiency, Coding and Choice Designs. Cary: SAS Institute Inc.

30.

Lehto

Cai

O’Leary

, et al. (2004) Tourist shopping preferences and expenditure behaviours: the case of the Taiwanese outbound market. Journal of Vacation Marketing 10(4): 320–332.

31.

Lewis

Chambers

(1999) Marketing Leadership in Hospitality: Foundations and Practices. New York: John Wiley and Sons.

32.

Kaplanidou

(2013) The impact of the 2008 Beijing Olympic games on China’s destination brand: a US-based examination. Journal of Hospitality & Tourism Research 37(2): 237–261.

33.

Louviere

Hensher

Swait

(2000) Stated Choice Methods: Analysis and Application. Cambridge: Cambridge University Press.

34.

Lyu

Han

(2017) Assessing preferences for mega sports event travel products: a choice experimental approach. Current Issues in Tourism 20(7): 740–758.

35.

Masterman

(2009) Strategic Sports Event Management: Olympic Edition. Oxford, UK: Butterworth-Heinemann.

36.

McFadden

(1974) Conditional logit analysis of qualitative choice behavior. In Zarembka

(ed) Frontiers in Economics. New York: Academic Press, pp. 105–142.

37.

McFadden

Train

(2000) Mixed MNL models for discrete response. Journal of Applied Econometrics 15: 447–470.

38.

Medlik

Middleton

VTC

(1973) The tourism product and its marketing implications. International Tourism Quarterly 3: 28–35.

39.

Murphy

Pritchard

Smith

(2000) The destination product and its impact on traveller perceptions. Tourism Management 21(1): 43–52.

40.

Uysal

Weaver

(1995) Product bundles and market segments based on travel motivations: a canonical correlation approach. International Journal of Hospitality Management 14(2): 123–137.

41.

Shen

(2009) Latent class model or mixed logit model? A comparison by transport mode choice data. Applied Economics 41(22): 2915–2924.

42.

Train

(2003) Discrete Choice Methods with Simulation. Cambridge, UK: Cambridge University Press.

43.

Weed

(2005) Sports tourism theory and method: concepts, issues and epistemologies. European Sport Management Quarterly 5(3): 229–242.

44.

Weed

(2008) Olympic Tourism. Oxford, UK: Butterworth-Heinemann.

45.

Weed

Bull

(2009) Sport Tourism: Participants, Policies and Providers. Oxford, UK: Butterworth-Heinemann.