What residents of potential Olympic cities want: using conjoint analysis to deal with dominant and heterogeneous preferences

ABSTRACT Understanding residents’ preferences for mega-sports events is a hot topic in tourism and event research. Thus far, most studies have assumed homogeneous preferences, and linear utility additivity of all event attributes when measuring residents’ support. This study uses an adaptive choice-based conjoint analysis combined with hierarchical Bayesian estimation. In contrast to previous research, the method considers the entire spectrum of residents’ preferences at the individual level and identifies non-substitutable dominant preferences (must-haves and unacceptables). A survey with 687 residents was conducted in Tyrol, an Austrian state with a remarkable Olympic tradition. The findings of this study extend the current literature by showing that the usually applied assumptions (homogeneous and linear additive substitutable preferences) are violated with substantial consequences. A simulation allows transforming the survey results into hypothetical referendum support rates for different feasible concepts of the Olympic Games.


Introduction
For several decades, the International Olympic Committee (IOC) was positioned to choose the next Olympic city from a group of top-qualified and ambitious candidates since many countries submitted bids for hosting the Olympic Games (Preuss & Schnitzer, 2015). This situation, however, changed fundamentally, notably by an increase in withdrawn bids, several rejected Olympic referenda and an overall decrease in these cities' residents' demand for the Olympic Games since the early 2000s (Könecke & De Nooij, 2017;Maennig, 2017). Thus, analysing residents' expectations and residents' support (Johnston, Naylor, Dickson, et al., 2021b) has become a hot topic in sports and tourism research (Bakhsh et al., 2018;Chen et al., 2019;Santos et al., 2019;Scott et al., 2018).
Owing to the methods used in previous studies, such as linear regressions (Bakhsh et al., 2018), factor analyses (Santos et al., 2019), or structural models (Pappas, 2014;Prayag et al., 2013), the results were based on two assumptions. First, it was assumed that an aggregated utility function was sufficient to represent the preferences of the entire population (Kurscheidt & Prüschenk, 2020;Wicker et al., 2017), implying homogeneous preferences for all residents. Second, the used empirical methods (Fredline et al., 2008;Gursoy et al., 2017) had the underlying assumption of a substitution effect for individual attributes of an event, indicating that higher utility levels of an attribute can compensate for the lower ones of another attribute. This assumption is The study contributes to a theoretical enhancement of the current research on residents' preferences and support for mega-sports events. It focuses on individual preferences instead of relying on aggregated ones and allows the existence of dominant preferences. Moreover, the results of the ACBC are transformed into support rates for various feasible concepts of the Olympic Games. This allows a transfer of the residents' preference structure to specific design possibilities for the Olympic Games. This provides a valuable link for communication between researchers and practitioners of mega-events.
However, the previously applied methods cannot identify necessary attributes, either in being present or absent, to consider supporting the Olympic Games. Moreover, studies commonly assumed homogeneous preferences due to the estimation of average relationships (e.g. using linear or ordered logistic regressions). Using discrete choice experiments, more specifically, the ACBC combined with a hierarchical Bayesian estimation allows for the existence of both dominant and heterogeneous preferences (Chapman et al., 2009;Johnson & Orme, 2007).
While the method has already found its way into tourism economics (Kemperman, 2021), studies using choice experiments on sports settings are limited. To date, discrete choice experiments have been used to assess tourists' preferences for travel products (Lyu & Han, 2017), to optimize ticket price strategies (Lee & Kang, 2011;Woratschek et al., 2020), or to assess participants' preferences for small-scale sports events (Fotiadis et al., 2016). The method's full potential by providing insights into the residents' individual preferences for mega-sports events, an essential driver of tourism (Fourie & Santana-Gallego, 2011), and tourism development has not yet been part of current research.
Choice experiments have their origin in market research (McFadden, 1986). The idea is that respondents choose their preferred out of at least two different product choices, that is, choices of travel packages or, in this survey, feasible concepts for the Olympic Games (Choe et al., 2020;Lyu & Han, 2017). The method allows the prediction of respondents' preferences based on their decision making (Train, 2009).
The most crucial part of the proposed method is to describe the concept for the Olympic Games with a small number of relevant characteristics for the design of the Games, called attributes (e.g. environmental planning) and their corresponding levels (e.g. eco-friendly, standard) (Deal, 2005). The attributes and levels must be carefully selected to represent the potential impact variables. The following literature review is the primary source of the attributes used in this experiment.

Support for mega-sports events
Literature on sports tourism has identified a remarkable variety of variables that impact event support, with the largest share being economic ones (Flyvbjerg et al., 2016;Khraiche & Alakshendra, 2020). Some studies have systematically reviewed the essential impact variables (Mair et al., 2021;Ribeiro et al., 2020). Research has indicated that an expected increase in tourism, economic growth, and boosted employment opportunities positively affect event support (Lin & Lu, 2018;Ritchie et al., 2009;Wood & Meng, 2021). The same holds for expected beneficial long-term legacies, such as (new) sports and transport facilities and infrastructural, social, and environmental surplus (Gratton & Preuss, 2008;Pereira, 2018;Preuss, 2019). One of the most fundamental predicted benefits is destination promotion (Bowersox, 2016;Kaplanidou et al., 2016).
However, staging a tourist event is accompanied by a financial burden, primarily caused by immense investment costs of required infrastructure, often followed by price and tax increases (Kim et al., 2006;Scheu & Preuss, 2018;Wang & Bao, 2018). The potential hosts are often concerned about financial risks (Kurscheidt & Prüschenk, 2020).
While an event is primarily evaluated economically, various studies highlight the importance of intangible determinants (Arnegger & Herz, 2016;Dolan et al., 2019;Porter & Fletcher, 2008). Negatively perceived environmental attributes can be ecological destruction, increased pollution, or the loss of cultural resources (Gursoy et al., 2011;Müller, 2012). In contrast, incentives to save local and historical resources, construct energy-efficient infrastructures, and waste management can positively impact event support (Karadakis & Kaplanidou, 2012;Kim et al., 2006;Yuen, 2008). Regarding socio-cultural effects, the literature shows that community attachment, social inclusion, national pride, and cultural exchange positively support such events Shipway & Brown, 2007). Negatively linked are crime, perceived security risk, or overcrowding (Choe et al., 2020;Zhou & Ap, 2009).
All attributes in this choice experiment rely on the literature review and are related to the recommendations of the Olympic Agenda. The Olympic Agenda 2020 (IOC, 2014) is a strategic repositioning for the organizers of the Olympic Games. The Agenda is focused on a straightforward and transparent planning process, increased sustainability, and decreased financial expenditures (IOC, 2014(IOC, , 2021. Thus, the expertise of the IOC can be used adequately when designing the discrete choice experiment.

Survey area
Tyrol (Austria) was ideal for observing residents' expectations and support for mega-sports events, as Innsbruck (Tyrol) hosted the Olympic Winter Games in 1964 and 1976, the Winter Paralympics in 1984and 1988Youth Olympic Winter Games (Schnitzer et al., 2016. The survey gathered residents' preferences for the 2026 Olympic Winter Games six months before the public referendum. Contrary to the opinion polls forecast, estimating a support rate of 72% (Gallup Institut, 2017), the referendum was rejected at 53.3% in October 2017 (Government of Tyrol, 2017). This evident discrepancy between the opinion poll and the referendum outcome can be traced to various reasons (Erb et al., 2002;Whitehead et al., 2016). However, the focus of this study is not on reducing this evident bias, although estimating residents' preferences without assuming homogeneous and additively substitutable preferences may reduce such discrepancies.

Sample size and data collection
With a discrete choice experiment, the preferences of 687 locals six months before the public referendum on the attempt to bid for the 2026 Olympic Winter Games in May 2017 were collected. The experiment was implemented based on extensive literature research. The comprehensibility of the selected attributes and levels was discussed in three focus groups of five people each; students and experts in the field. Fifty-eight subjects voluntarily participated in a pretest. The final questionnaire was addressed to 850 people who received a link for the experiment via e-mail.
The sample was stratified into respondents living in the potential host city Innsbruck, residents in a municipality providing sports venues and infrastructures for the event, and residents of municipalities not hosting any part of the Olympic Games. This distinction might be necessary as studies (Ritchie et al., 2020) pointed to differences in support of mega-events between host and nonhost municipalities since the Olympic Games are followed by different benefits and costs for the host municipality, partly involved municipalities and those not involved in the holding of the event. A total of 771 participants started the experiment, of which 687 fully completed the questionnaire, resulting in a response rate of 80%. Orme and Chrzan (2017) recommend an optimal sample size of 300 for the discrete choice experiment, which is in line with Akis et al. (1996), suggesting a sample size of 385 for the parameter setting. Accordingly, a significance level of 5% and a confidence level of 95% were used. Thus, the final sample size is appropriate for the planned analyses.

Survey design
ACBC provides respondents with feasible product concepts; in this study, these are potential concepts of the Olympics. The selected attributes and levels are demonstrated in Table 1 (and in the  Supplementary Material Table A.1.).
Destination promotion (low, medium, high) is used to proxy potential economic benefits (Yuen, 2008). The infrastructure construction strategy (minimal renovation, sustainable renovation, or new construction) indicates the expected costs of hosting the Games. The same is true for the Olympic Villages' (OV) construction strategy; both attributes require necessary financial investments to host the Games.
The survey included the attribute host cities since the Olympic Agenda 2020 offers cities the possibility to collaborate for the 2026 Olympic Winter Games. Respondents could specify between Tyrol (AT) as a host, a collaboration of Tyrol (AT) and South Tyrol (IT), or the cooperation of the European Region [Tyrol (AT), South Tyrol (IT), and Trento (IT)] as host destinations. Since the European Region, especially South Tyrol (IT), has a unique historical significance due to its German minority, the attribute host cities are also partly a proxy for national pride (Khaptsova & Fruchtmann, 2020).
The attribute environmental planning offered respondents two levels; standard (as usual) or ecofriendly. The attribute subsequent use of the Olympic Village is a proxy for the cities' potential social development as it could either be reused as public housing (social reuse) or sold after the event (cost-saving). The respondents' strategic choice of infrastructure construction strategy directly impacts long-term legacies. Minimal renovation would be much cheaper than sustainable renovation or new construction. However, it provides minor sustainable long-term event legacies (e.g. public transport systems and sports facilities), repeatedly highlighted as an essential aspect of event support in previous research (Gold & Gold, 2009;Pereira, 2018).
The choice of an economic or sustainable construction of the Olympic Village provides information about the relevance of the costs and the importance of event legacies. While selling the Olympic Village would significantly decrease the hosts' financial burden, reuse as public housing impacts social development. A decrease in costs is also true for the attribute host cities, indicating a reduction in the financial burden for each potential host city.
By multiplying the corresponding levels of the six attributes (3 3 × 2 3 ), the analysis considered 216 different Olympic concepts in the full factorial design.

Empirical analysis
ACBC is theoretically based on the random utility maximization theory (Manski, 2001;McFadden, 1973), assuming that decision-making is rational, temporally stable, and based on transitive preferences. Respondents choose their preferred product from at least two different product choices (feasible concepts for the Olympic Games). The method allows for predicting respondents' preferences based on their decision-making (Train, 2009). The conjoint analysis technically consists of four stages (Chapman et al., 2009;Johnson & Orme, 2007;Sawtooth Software, 2020): (1) Build-your-own task (BYO). Respondents assemble their preferred concept by selecting the most valuable level of each attribute.
(2) The screening section. Respondents state whether each concept is considered (a possibility or not a possibility). Each screen simultaneously offers respondents three out of all possible concepts in this experiment, repeated four times. The analysis generates three pieces of information; first, it identifies required levels (must-haves) and second, those that are not acceptable (unacceptables). Additionally, it calculates the respondents' utility threshold (None-Value).
(3) The choice tasks. Respondents are asked to pick their unique preferred concept from a set of alternatives based on the previous steps, repeated five times.
Using a hierarchical Bayesian estimation technique, the individuals' part-worth utilities at each level and the resulting importance of the attribute are estimated.
By adding up the part-worth utilities, the total utility value of a concept was calculated. Thus, the respondents' binary support rate for each of the 216 Olympic concepts was calculated. If the total utility value of a feasible Olympic concept exceeded the individuals' None-Value threshold, the respondent was assumed to support the concept. Otherwise, it was rejected.

Socio-demographics
Overall, 687 respondents completed the survey, whereby 58.7% indicated their support for the Games. Table 2 provides the socio-demographic profile of the participants. Similar to other surveys in the field, the sample is relatively young (Lu et al., 2019;Ritchie et al., 2020).

Conjoint analysis findings
The reliability of the conjoint analysis can be evaluated using the root likelihood value (RLH), which ranges from 0 to 1 (Sawtooth Software, 2020). The mean RLH was 0.68 with a standard error of 0.003, indicating the high quality of the estimated model (Deal, 2005) and high consistency in the responses throughout the survey. Respondents indicated infrastructure construction strategy (21.79%) and Olympic Villages' subsequent use (21.32%) as the two most important attributes (Table 3). However, both attributes showed considerable variability, providing evidence of heterogeneous preferences, which were not yet considered in previous studies on mega-sports events (Supplementary Material Table  A.2.). While the mean importance of the attribute's destination promotion (13.18%) and host cities (15.48%) was only slightly lower than the mean of 16.67%, the importance of the construction strategy of the Olympic Village was far below at 11.36%.
Contrary to the indicated heterogeneity in the attributes' importance, the results of the respondents' first-best choices (BYO) displayed a high level of consent. About three-quarters of the respondents preferred a sustainable construction of the Olympic Village, its subsequent use as public housing, and eco-friendly planning of the Games. Almost two-thirds required sustainable infrastructure renovation. Only the attribute host cities showed a more uniform distribution.
These results demonstrate the existence of dominant preferences. 10% of the respondents stated a preference for low destination promotion, the construction of new infrastructure, standard environmental planning, or the sale of the Olympic Village as unacceptable. Up to 14% of the respondents indicate eco-friendly planning and the subsequent use of Olympic Villages as public housing as must-haves. Overall, about 60% expressed at least one level of an attribute (up to five levels) as unacceptable or a must-have, violating the frequently used assumption of substitutability between different attributes.

Simulated support for potential Olympic concepts
If the total utility value of an Olympic concept exceeded the respondents' support threshold (None-Value), it was counted as supported. Otherwise, the concept was considered rejected. Thus, the approach estimated the respondents' support for the 216 feasible concepts. By aggregating the results, hypothetical Olympic referendum support rates were simulated. More than half (124 of 216) of the concepts did not reach the 50% majority (Supplementary Material Figure A.1). Nevertheless, 16 Olympic concepts achieved the support of at least 80% (Table 4). The concept with the highest support rate showed that the respondents demanded the following: Tyrol as host, high destination promotion, a sustainable renovation of the infrastructure, and a sustainable renovation of the Olympic Village. The Games should be eco-friendly, and the Olympic Village should be subsequently used as public housing. This corresponds precisely to the first-best choice of the respondents; the BYO results (Table 3). However, completely different concepts can still achieve high support. Each level occurs at least once in any Olympic concept with a support rate of above 50% in the rows of Table 4, except for the selling of the Olympic Village.

Robustness checks
The internal coherence of the surveys was investigated to check the robustness of the results. The Olympic concept with the highest support of 92% corresponded precisely to the respondents' preferred levels, stated in the respondents' BYO, which set the scope for any concept that could reach a majority.

Relative importance
concept in mind, based on the information provided by the government before the referendum: a medium destination promotion, Tyrol as host, sustainable renovation of the infrastructure and the Olympic Village, standard environmental planning, and the selling of the Olympic Village. This led to a support rate of 51.7% (standard error ±3.8%) and a 95% confidence interval of 44.1% and 59.3%, respectively. The confidence interval includes the actual Olympic referendum's outcome with a support rate of 46.7%.

Discussion
This study aims to gain deeper insights into the estimation of residents' preference structure on mega-sports events using ACBC, a particular form of choice experiments, with more flexible decision-theoretical assumptions (Chapman et al., 2009;Johnson & Orme, 2007;Meginnis et al., 2018). Discrete choice experiments have already been used in tourism and event research (Fotiadis et al., 2016;Lyu & Han, 2017), but studies have not yet examined whether important common assumptions made by previous studies in this area are justified. In particular, this study investigates the assumption of homogeneous preferences (Johnston, Naylor, & Dickson, 2021;Kurscheidt & Prüschenk, 2020; and the absence of dominant, so-called lexicographic preferences (Scott, 2002) are accurate. The results indicate that the method offers significant potential to elucidate residents' preferences and organize the Olympic Games according to the residents' demands. Thus, the probability of achieving a supported referendum can be increased, and if no concept is viable, a futile effort can be avoided.

Empirical findings
While this study uses a different empirical approach than previous research on mega-sports events (Gursoy et al., 2017;Ritchie et al., 2009), the ACBC results are qualitatively compatible with previous studies' findings but can identify the preferences of the respondents more precisely. Of all six attributes, the infrastructure construction strategy and the subsequent use of the Olympic Village as part of the necessary infrastructure (Flyvbjerg et al., 2016) reveal by far the highest importance. This corresponds to previous findings indicating that infrastructure, a proxy for the financial burden of hosting mega-sports events (Schnitzer & Haizinger, 2019), is one of the essential aspects (Khraiche & Alakshendra, 2020;Pereira, 2018). The findings of the ACBC reveal that the construction strategy does not have to be as economical as possible. The part-worth utilities suggest that respondents are not primarily concerned about the financial burden but interested in a sustainable renovation of existing infrastructure, including the Olympic Village. Since the method allows for dominant preferences, it reveals that it is unacceptable for about 14% of the respondents to sell the Olympic Village instead of reusing it as public housing. Several studies found that residents demand the sustainable use of public funds and taxpayers' money, the prime source for infrastructural investments for mega-sports events (Deccio & Baloglu, 2002;Kim et al., 2006). Others (Ritchie et al., 2020;Scheu & Preuss, 2018) highlighted the importance of developing sustainable long-term event legacies to encourage residents' support. Thus, stakeholders should focus on a sustainable renovation of sports facilities and the necessary infrastructure (e.g. Olympic Village) as subsequent reasonable use of space is of vital interest to the respondents to support the organization of the Games. This is precisely one of the critical points anchored in the Olympic Agenda. The recommendations aim to reduce financial expenses by reusing existing infrastructure (IOC, 2014(IOC, , 2018. Environmental planning seems less important than the financial burden, proxied by infrastructure construction strategy. In contrast to the survey findings, ecological developments following megasports events, such as environmental legacies, are from a particular interest in other results (Gold & Gold, 2009;Karadakis & Kaplanidou, 2012). About 13% of the respondents indicated that standard environmental instead of eco-friendly planning of the event is unacceptable.
Destination promotion, used as a proxy for economic benefits, is surprisingly the least essential aspect of the survey. Several studies corroborate an increase in tourism arrival while hosting mega-sports events (Fourie & Santana-Gallego, 2011) and short-term destination image boots (Kassens-Noor et al., 2019;Lai, 2018). However, research is not entirely assured whether megasports events generate long-term destination promotion effects (Hahm et al., 2019;Vierhaus, 2019).

Method
Methodologically, the existence of unacceptables and must-haves confirms the assumption of dominant preferences (Scott, 2002). Sixty percent of the respondents state at least one dominant level (must-have) or a level that led them to reject hosting the Olympic Games (unacceptable) instantly. This finding strongly indicates that the dominant preferences are not negligible. An attribute of the Olympic Games may be that important or intolerable that none of the others can compensate for this must-have or unacceptable attribute. Only a few perceived unacceptable attributes or levels can damage general support for tourism events. This implies that survey results that do not consider dominant preferences can lead to invalid results.
The broad range in the importances, the part-worth utilities, and the dominant preferences provide evidence for heterogeneous preferences. (A cluster analysis confirms this finding, further details can be found in the Supplementary Material). Thus, the assumption of homogeneous preferences, supported by previous studies Wicker et al., 2017), is demonstrated to be, at least, incorrect for this sample. If groups with different preferences exist, optimizing the average preferred Olympic concept is inappropriate. By fulfilling one's needs, others are dissatisfied, and therefore, none of these concepts can reach a majority in an Olympic referendum. However, implementing a corresponding concept of the Olympic Games adapted to the residents' demands could reach an agreement using individualized communication for the different interest groups. The support or rejection of feasible Olympic concepts has to be estimated considering individual disparities, demonstrated for the first time in this study.
The results indicate that the proposed method (ACBC), with its broad spectrum in preference indications, is an improvement compared to the commonly applied approaches from a theoretical and an empirical point of view. Comparing the estimated part-worth utilities with the estimated None-Value offers a very intuitive interpretation. It allows calculating the respondents' support share for any feasible Olympic concept. In our sample, many of them (82 out of 216) could achieve a majority share in a hypothetical Olympic referendum. Since the standard errors in the findings are relatively large, support of at least 60% should be the target to ensure a favourable Olympic referendumalmost 30% of the 216 concepts achieved this threshold.

Limitations
An essential part of the discrete choice experiment is the selection of attributes and levels. This study is the first attempt to apply the conjoint analysis on preference analyses to mega-sports events. Indeed, there could be scope for improvement in choosing the attributes and levels, even though they were reasonably selected based on the literature review and related to the recommendations of the Olympic Agenda 2020. Moreover, the empirical analysis is based on the theory of random utility maximization with rational, transitive and temporarily stable preferences. This study does not consider decisionmaking processes with the possibility of priming or intervention effects throughout the experiment (Erb et al., 2002;Lewis et al., 2015), although this is an interesting question for further research.
As a first plausibility test of the simulation results, a specific Olympic concept was assumed that respondents could have expected. However, it is necessary to explicitly ask the interviewees how they expect the Olympic Games. Thus, a more appropriate application for a robustness check can be provided. Nevertheless, the applied procedure is sufficient to give the first idea of the robustness of the approach.
Finally, the study was conducted in a specific geographical context with certain peculiarities. The region (Tyrol) was already the host site of the previous Olympic Games and thus has an Olympic history, which may have influenced the survey participants. Therefore, the integration of different attributes into other contexts should be considered.

Conclusion and implications
Preference analyses on mega-sports events can benefit from using discrete choice experiment analyses. Studies are currently somewhat limited to the question of whether residents support or reject an event instead of asking how the event should be designed to get residents' support. The proposed method enhances the current research by allowing dominant, so-called lexicographic preferences (must-haves and unacceptables) and provides an added value by considering preference heterogeneity.
The use of the ACBC at an early stage of the Olympic bidding process can provide information on whether or not feasible concepts for the Olympic Games could find a majority in a public referendum. Even if no viable concept can be found, further expensive and useless efforts can be avoided. However, the method more likely provides a set of Olympic concepts capable of a majority in a referendum. Thus, the optimized concept can be selected, and the application process can be designed efficiently.
Moreover, the results of the ACBC can be used for market segmentation. If necessary, the identified groups with similar preferences can then be targeted with group-specific information. First, the method may appear complex and costly. However, with about 15$ per survey and 1000 contacted respondents, the costs of the discrete choice experiment remain marginal compared to the total financial expenses in the application process.