I. INTRODUCTION

Professional sport leagues around the globe have instituted many extra-market rules, especially with regard to the sport labor market. Common measures in this regard are salary caps, entry drafts, or revenue sharing devices. (1) Such devices are commonly justified as necessary to maintain or improve the level of competitive balance (CB) within a league. In fact, the CB argument is the main "justification that sports leagues offer to defend agreements otherwise prohibited by antitrust laws" (Mehra and Zuercher 2006, 1505). This argument is based on the assumption that sport competitions need to be tight to be attractive for spectators--a relation first mentioned by Rottenberg (1956) and Neale (1964) in their seminal works six decades ago, which is now established in the literature as the uncertainty of outcome hypothesis (UOH) in sports.

Following Szymanski (2003), uncertainty in this regard refers to outcomes of games (match/ game uncertainty) or in-season sub-competitions such as the championship race or the fight against relegation (seasonal uncertainty) as well as the degree to which a league is dominated (or not) by a few teams over time (championship uncertainty). Importantly, single games might not only be characterized by game uncertainty. Rather, seasonal uncertainty might also unfold at the level of a single game, which is referred to as Match Relevance (e.g., Jennett 1984), Decisiveness of a Game (e.g., Geenens 2014), the League Standing Effect (e.g., Humphreys and Zhou 2016), or Competition Intensity (e.g., Scelles et al. 2013a, 2013b) in the literature. (2)

Over several decades, numerous papers have tested the impact of uncertainty--in particular game uncertainty--on the demand for sport. In contrast to the widespread belief in the UOH by policy makers, however, this empirical literature offers ambiguous findings. While there is some supportive evidence for the relevance of seasonal uncertainty, match-level attendance studies seldom find that the more tickets are sold the more uncertain the result of the game is anticipated to be. In contrast, most studies show that stadium attendance rises as the certainty of a home team or away team win rises. (3) Moreover, although there is more supportive evidence for the impact of game uncertainty on TV viewing in several sports, a very limited number of studies on soccer finds either clear (Buraimo and Simmons 2009; Meier and Leinwather 2012; Schreyer, Schmidt, and Torgler 2016a) or partial support (Schreyer, Schmidt, and Torgler 2016b, 2017) for the relevance of close soccer games for TV viewers. (4)

Currently, three different behavioral economic explanations for the remaining "lack of certainty about outcome uncertainty" (Leach 2006, 117) are discussed in the literature. (5) First, fans might exhibit loss aversion and derive more utility from the chance to see an upset. This idea is based on prospect theory and the concept of reference-dependent preferences (Kahneman and Tversky 1979) and was recently transferred into the context of sports demand by Coates, Humphreys, and Zhou (2014). The authors distinguish between two types of utility that a consumer receives from attending a sporting event, that is, "consumption" utility that corresponds to utility from standard consumer theory and "gain-loss" utility that is derived from differences between expected and actual game outcome. According to their theoretical model, the UOH only emerges when the marginal utility of an unexpected win exceeds the marginal utility of an unexpected loss. When, however, the marginal utility of an unexpected loss is larger than the marginal utility of an unexpected win, a consumer exhibits loss aversion and derives more utility from the chance to see an upset, which by definition requires a favorite team ex ante. (6)

Second, fans might perceive closeness of a game in a different way than how economists have tended to measure it due to the existence of behavioral anomalies such as framing effects (Tversky and Kahneman 1981), attention level effects (Bernheim and Rangel 2009), or threshold effects (Simon 1955). Though a consistent theoretical model incorporating these anomalies does not exist, some recent evidence based on data gathered in fan surveys suggests that such differences between "perceived" uncertainty by the fans and "objectively" (statistically) measureable uncertainty with regard to the effect on demand might exist indeed (Nalbantis, Pawlowski, and Coates 2015; Pawlowski 2013; Pawlowski and Budzinski 2013). While these studies offer a new and interesting line of research, it remains unclear, however, what previously developed subjective measures mean. In this regard, perceived "suspensefulness of a game"--which is the wording used in previous fan surveys--might proxy game uncertainty, seasonal uncertainty (at the level of a single game), both dimensions, or even further issues such as the quality of the contestants.

Third, given the fact that consumption depends upon the affective dispositions of viewers towards the competing teams (Raney 2006), it may be that the impact of game uncertainty is moderated by being a fan of the home team, the visiting team, or neither the home nor the visiting team (neutral spectator). Schreyer, Schmidt, and Torgler (2016c)--addressing the attendance behavior of season ticket holders (i.e., a specific type of home team fans) in the German Bundesliga--provide partial support for this assumption. While their results point towards a u-shaped relationship between home win probabilities and the season ticket holders' decision to attend a game, other game uncertainty measures suggest that season ticket holders do care about game outcome uncertainty. Though the authors only have access to a specific type of home team fans, their study provides some initial empirical evidence on the importance of accounting for fan-team relationships in further investigations.

Until now, considerable differences in the measurement of game uncertainty as well as a lack of appropriate data made any attempt to synthesize these plausible though different explanations for the lack of support for the UOH--in particular with regard to game uncertainty--impossible. While previous studies relying on secondary data were unable to detect possible differences between subjective and objective measures of game uncertainty, the major limitation of earlier studies employing subjective measures is that it remains unclear what these measures mean and how they relate to common objective measures. Moreover, though it appears relevant to control for fan status in this regard, no study previously looked at all different types of fans, that is, home team fans, visiting team fans, or neutral spectators.

Our study tries bridging between these plausible though different explanations by using data, representative for all soccer-interested individuals in Germany and gathered in repeated surveys, to develop a measure of perceived game uncertainty, which is closely related to a common measure based on betting odds, and subsequently test its impact on the consumers' intentions to watch soccer games live. The design of both data collection and analysis enables us to test several assumptions commonly thought to be problematic when employing a stated-preference approach. Moreover, the data allow a distinction with regard to fan status and therefore testing its (eventually) moderating role of the relation between (perceived) game uncertainty and the demand for sports.

II. DATA AND METHODOLOGY

A. Data Collection and Cleaning

Soccer-interested individuals (fans from now on) were randomly recruited from a German-wide representative online panel provided by a market research company. The first question served as a screen to identify those with a minimum interest in soccer. Individuals uninterested in soccer did not answer the questionnaire. The survey took place in the days prior to two German Bundesliga matchdays (i.e., the 10th and 27th of the 2014-15 season). Fans were asked about all nine games on the upcoming matchday, for a total of 18 games in the survey. Matches in the first and second matchday pair the same teams (though with home and away teams flipped) with the top game between Football Club (FC) Bayern Munchen (FCB) and Borussia Dortmund (BVB; see Table 1). This unique set-up allows gathering viewing intentions from all fans about all games regardless of whether they are a fan of one of the participating teams. Respondents who are neither a fan of the home nor the visiting team in a given game are assumed to be "neutral fans."

The link to the questionnaire of the second survey was initially distributed only among those who had already participated in the first survey to create a panel. Those respondents from the first survey who did not respond in the second survey were replaced by randomly selected new respondents. The overall objective was to have two samples with at least 3,000 participants in each round and a certain overlap of participants who filled in the questionnaire in both rounds. The total number of completed surveys is 6,332 between both rounds (3,029 in the first survey; 3,303 in the second survey). There are eight observations where the ID of a respondent appears twice in the survey. These observations concern respondents who encountered problems in their first attempt to complete the questionnaire. Therefore, it was decided to delete the (chronologically) first response of each respective observation. Moreover, two observations do not have an identification number and were consequently also deleted from the sample. Further quality and consistency checks were employed as described in detail in Appendix A. Finally, since several matches are played simultaneously and the decision to watch any individual match live in the stadium can make watching other matches impossible, we excluded the few stadium attendees from our sample, that is, 220 from the first survey and 170 from the second. (7) The final net sample used in our data analysis consists of 2,415 (2,686) observations in the first (second) survey.

To assess the generalizability of our results a useful study in this context was conducted by SPORTFIVE (2009) examining a representative sample of the total German population and providing information about the distribution of gender and age as sorted by the level of "general interest in soccer" and the "frequency of attendance to a live professional soccer match in the current season." The portion of females responding to the survey in our study (first survey: 47%; second survey: 43%) is very much like the SPORTFIVE (2009) sample (47%). Regarding age, again the respondents in our sample seem to be on average comparable to the respondents in the SPORTFIVE (2009) sample. In our study about 19% are less than 29 years old (SPORTFIVE: 22%), 29% are between 30 and 40 years old (SPORTFIVE: 34%), and the remaining 51% are more than 50 years old (SPORTFIVE: 44%). Unfortunately, no further variables are available for a comparison. However, these figures suggest that our sample is representative with regard to gender and age of the German population with a general interest in soccer.

B. Measures

The fans' stated intention to watch a game "live on TV" or "not live at all" serves as dependent variable in our demand models. As mentioned before, we needed to exclude the few stadium attendees from our sample and focus in the following on the intention to watch any of the games live on TV (or not). A common criticism of using such a stated-preference measure as proxy for demand is that it is based on what people say rather than what people do. In this regard, however, it is important to note the concreteness of (a) the products under consideration, that is, specific soccer games, (b) the choice scenario developed, that is, a few days prior to two matchdays, and (c) the question asked, that is, "Will you be watching game x live?" (with three possible answers being "no," "yes, in the stadium," or "yes, on TV"). This forces respondents to make a forecast about their decision to consume a clearly defined product in the very near future. While this does not rule out false statements in general, it limits the number of possible reasons for false statements. More precisely, a statement might be false because the respondent did not want to answer correctly (liar) or because her plans have changed in the short time period between when the survey was conducted and kick-off (switcher). Since we are not interested in forecasting the total number of TV viewers for any specific game but rather in discriminating between viewers and non-viewers such false statements are not a problem as long as they are randomly distributed among respondents. We do not see any reason to believe that this assumption is violated.

The first uncertainty measure developed in our study follows earlier studies on perceived competitive balance (PCB). The term PCB was established by Pawlowski (2013) and Pawlowski and Budzinski (2013) and their studies on perceived CB and suspense within a league. Nalbantis, Pawlowski, and Coates (2015) used the same term for their study. However, given the specific setting in Nalbantis, Pawlowski, and Coates (2015), that is, a single game, as well as the temporal order of their survey, that is, before the kick-off of a specific game, their ex-ante measure is probably better described as perceived game suspense. We made use of the latter term and accounted for the perceived "suspensefulness" of the single games by asking the fans to state on a scale of 0-10 (0 [equivalent to] not at all suspenseful ... 10 [equivalent to] very suspenseful) "How suspenseful do you think the upcoming GAME will be?" Many scholars define suspense as an experience of uncertainty. (8) In this regard, Mullet et al. (1994) note that in a gambling context suspense reaches its peak when the uncertainty as to the outcome of the gamble is at its highest (50% win probability). In terms of a soccer game's outcome, it may be that a fan perceives suspense at its maximum when the (perceived) likelihood of a loss is more or less even with that of a win. However, uncertainty over the outcome may not always be sufficient to generate feelings of suspense regarding the match (Madrigal and Dalakas 2008; Ortony, Clore, and Collins 1988). For instance, the exante uncertainty about the outcome of a soccer game may not be intrinsically suspenseful, unless the consequences of a win or loss are compelling. As such the hope of winning the championship, securing a place in the Union of European Football Associations (UEFA) club competitions, avoiding relegation or the fear of failing to achieve these milestones--because of the game's (uncertain) outcome--do also generate suspense. Therefore, the perceived suspense-fulness of a game might be related not only to game uncertainty, but also to seasonal uncertainty (at the level of a single game) or both. Moreover, a game could also be suspenseful because a coach or a player is on the verge of surpassing a milestone (anticipation of record-breaking performance, becoming the league's top goal scorer, etc.) or because the quality of the contestants is perceived as being high. In general, what is suspenseful may be highly idiosyncratic (Mullet et al. 1994).

In contrast to Nalbantis, Pawlowski, and Coates (2015) who used data from just a single game, we have data for 18 games. Therefore, we are able to compare this subjective measure with available objective data about the games in order to apprehend (at group level) the link between suspensefulness and game characteristics. As indicated by Figure 1, there is a strong negative correlation (for all games: r= -0.580; without FCB vs. BVB: r= -0.857) between perceived game suspense and the sum of the opponents' league ranks (prior to the matchday). This negative correlation is consistent with the idea that the most suspenseful games are between teams involved in the race for the championship or the qualification places for the UEFA Champions League and UEFA Europa League. Moreover, this measure seems to consist of a notion of clubs' brand strengths given the fact that the top game between FCB and BVB (with the highest average brand index, see Table 1) is perceived to be the most suspenseful despite the rather poor performance of Borussia Dortmund in the first half of the 2014-2015 season. In this regard it is indicative that the strong positive correlation (all games: r = 0.769) between perceived suspense and the average brand index of the contestants seems to be determined by this particular game (without FCB vs. BVB: r = 0.288).

This assumption is further supported by comparisons between our measure of perceived game suspense and different competition intensity (CI) measures as proposed by Scelles et al. (2013a), Andreff and Scelles (2015), and Scelles (2017). The correlations, provided in Appendix B, indicate that games involving clubs which are in championship contention are perceived as being more suspenseful than games involving clubs which are in contention for other sub-competitions. Moreover, it appears that perceived suspense may also reflect the quality of the contestants, since games involving clubs which are closer to the relegation zone are perceived as comparably less suspenseful.

Summing up, while we are unable to disentangle this further due to the limited number of games under consideration, these simple correlations suggest that (perceived) game suspense measures something different than (perceived) game uncertainty and might be related rather to seasonal uncertainty (at the level of a single game) as well as quality and/or brand strength of contestants. Therefore, to fully address game uncertainty in our setting and in order to compare our findings with those studies using objective measures, we developed a second measure, called perceived game uncertainty, with a novel approach in our study.

Respondents were asked to state on a scale of 0-10 (0 [equivalent to] away club will definitely win ... 10 [equivalent to] home club will definitely win) "How likely do you think will be a home win in the upcoming GAME?" The answers are interpreted as subjective home win probabilities, which, with their squared terms, we include in the regression models. Note that respondents' probability judgements are based on beliefs about the properties of the games such as the teams'/players' performance, etc. As such, a respondent's prediction for game A does not affect her prediction for game B. Therefore, the elicitation of home win predictions in our surveys is not affected by a "conjunction fallacy" (Tversky and Kahneman 1983). Moreover, the behavioral economics and psychology literatures find that subjective probabilities can be quite different than objective probabilities, particularly for low probability events with which people have little experience (Ungemach, Chater, and Stewart 2009). However, the subjective home win probabilities in our survey are strongly correlated (for all games: r = 0.889; without FCB vs. BVB: r = 0.879) with objective home win probabilities derived from betting odds (see Figure 1). Because our survey respondents are soccer fans it is likely that they are well acquainted with the game and with the teams and, therefore, are reasonably accurate, on average, in their perceptions of the likelihood of the home team winning specific matches. Consequently, we contend that this measure allows for a direct comparison with previous findings in the literature using home win probabilities derived from betting odds.

Since empirical evidence (e.g., Zillmann and Cantor 1977) shows that a positive (negative) outcome is enjoyed (disliked) based on the individual's disposition towards the protagonists and antagonists, it is likely that there is a dispositional mediation of (perceived) game uncertainty and suspense (Zillmann 1996; Schreyer, Schmidt, and Torgler 2016a, 2016b, 2016c). To take the potential moderating role of fanship status into account, our models include interactions between the variables of interest (i.e., perceived suspense and perceived home win probability) and fanship status (home fan and away fan). Furthermore, we control for socio-demographics (marital status, gender, age, and travel distance from the venue hosting the game) and game dummies in the models.

Table 2 reports the statistics, rounded to two decimals, of the sample used separately for the first (second) matchday surveys. Overall, 2.6% (second survey: 1.7%) of the respondents are fans of the home team, whereas 2.0% (2.5%) are fans of the visiting team. Moreover, 47% (43%) are female. On average, a survey participant is 47 (48) years old and lives about 372 (387) kilometers away from the venues hosting the games. Furthermore, 25% (23%) of the respondents are single. For both matchdays, the proportion of games that respondents state they intend to watch live on TV is between 23% and 25%. Regarding games' suspensefulness, on average, respondents rate the games as 5.6 (5.6) out of 10. At the same time, the average rating of the likelihood of a home win is about 5.9 (5.7) out of 10. As can be seen, there are hardly any differences between the two survey rounds with regard to the descriptive statistics. Finally, there are no substantial deviations with regard to the descriptive statistics between the sample with and without consistency-checked corrections (see Table A2 in Appendix A).

C. Empirical Strategy

The data gathering process as described before provides a panel data set with nine observations per respondent and information about each decision to watch (or not) any of the games live on TV. Pooled logit models with individual clustered error terms and fixed effects (FE) models were estimated for both matchdays. The major difference between these approaches is that the FE models only use those individuals who stated an intention to watch some (but not all or none) of the matches live on TV while the pooled models use all individuals. Though it is generally desirable to use all observations available, including individuals for whom there is no variation in the decision to view or not view despite variation in their perception of the home win probability will downward bias the effect of home win probability. Furthermore, while the pooled models allow controlling for available individual characteristics in the data (i.e., marital status, age and gender), the FE models wash out the influence of all of these individual traits that are constant across matches. In this regard, the FE models also purge any potential common method bias which might occur when independent and dependent variables are gathered with the same instrument as it was done here (for a discussion on this issue see Antonakis et al. 2010).

III. RESULTS

As indicated by the results in Table 3, the influence of the explanatory variables does not vary much--neither between the two matchdays under consideration nor between the different econometric specifications. In all models, perceived suspense is positively related to the intention to watch a game live on TV while home win probability is negatively and its square is positively related to the intention to watch a game live on TV. As expected the pooled model coefficients are closer to zero than the FE coefficients. In each case, the results imply that the probability of watching the game live on TV is higher when the game is perceived to be suspenseful. "Perceived game suspense," however, is different from "perceived game uncertainty" since the home win probability is also a significant predictor of TV viewing intentions. Interestingly, the probability of watching the game live on TV is higher when respondents strongly expect either a home or an away team win (Figure 2). This contradicts the UOH with regard to game uncertainty and is in line with previous studies employing objective measures of game uncertainty.

This finding comes along with control variables showing plausible signs. As expected, the home and away fans are actually more likely to watch the game of their favorite team than are neutral fans. The probability of viewing decreases with increasing age up to a certain point after which the likelihood to watch the game live on TV increases again. The same nonlinear pattern is also evident for the variable measuring distance to the game venue. Interestingly, being single is negatively associated with the likelihood of watching the game live on TV, whereas being female has no statistically significant effect. Finally, as expected there are game specific differences in viewing behavior with the top game between FCB and BVB attracting the most fans. (9)

To check the robustness of our results we re-estimated all models described in this paper (1) without quality threshold, (2) with no strict sample correction for "fan" and "age" as well as (3) with subjective home win probabilities interacted with "interest in the league" instead of fan status. All results remain similar to the results in the paper. Importantly, the u-shaped relation between subjective home win probabilities and the intention to watch a game live on TV still exists. (10)

IV. CONCLUSION AND DISCUSSION

This study tries bridging between plausible though different behavioral economic explanations for the lack of support of the well-known UOH in sports. We develop and test a measure of perceived game uncertainty that is comparable to objective measures frequently tested in the literature. Overall, the findings suggest that the probability of watching a soccer game live on TV is higher when respondents expect a certain home or away team win. This is in line with most previous studies employing objective measures of game uncertainty. We conclude that the common finding that fans do not value game uncertainty can be explained by fans exhibiting loss aversion with regard to game uncertainty rather than differences between perceptions and measurements of game uncertainty. In this regard, though home and away fans are actually more likely to watch the game of their favorite team than are neutral fans, we do not find any evidence of fanship status being a moderator of the relation between game uncertainty/suspense and the demand for sport.

Moreover, the paper finds that peoples' perception of the suspense fulness of a game is distinct from their perception of the relative strengths of the teams as the suspense variable and both the home win probability and its square are all individually statistically significant. The structure of our data allows comparing the developed game uncertainty and suspense measures with objective data on different characteristics of games and opponents. Results derived from simple correlations and the fact that the coefficients on the suspense variable are somewhat larger at the 27th matchday than at the 10th matchday are both consistent with the idea that perceived suspense measures seasonal uncertainty (unfold at the level of a single game) which is referred to as Match Relevance, Decisiveness of a Game, the League Standing Effect, or Competition Intensity in the literature. Moreover, perceived suspense seems to capture also the quality of the contestants, since games involving clubs which are closer to the relegation zone, are perceived as less suspenseful. Exploring this more in depth, however, is the subject of future research. In this regard, it also appears to be worth exploring whether and how the notion of suspense as developed by Ely, Frankel, and Kamenica (2015) is related to survey responses here and in other studies focusing on the relation between game uncertainty and the demand for sports. Moreover, it would be interesting and relevant to test the relations between perceived game uncertainty, suspense, and the demand for sport in other settings including different sports and countries.

ABBREVIATIONS

BVB: Borussia Dortmund

CB: Competitive Balance

CI: Competition Intensity

FC: Football Club

FCB: FC Bayern Munchen

FE: Fixed Effects

PCB: Perceived Competitive Balance

UEFA: Union of European Football Associations

UO: Uncertainty of Outcome

UOH: Uncertaintv of Outcome Hypothesis

APPENDIX A: DATA AND MODELS QUALITY CORRECTIONS

The quality correction program of Questback (11) is able to identify participants who simply "clicked through" based on the time to fill in the answers. Since the number of questions a participant has to answer might vary between participants, the time required by a participant to complete the survey as a whole is not a reasonable measure of "quality." Therefore, an individual quality variable is calculated based on the time taken by the participant to complete a particular page of the survey in relation to the average processing time of the entire sample for this page. This quality variable has a value of 0.5 if the corresponding user required exactly the average time for processing the questionnaire pages. A value of 0.25 signifies that the respondent needed only half as long as the average processing time (per page) and so on.

Table Al provides the distribution of the quality variable separately for each survey as well as for the total number of participants. In general, it is recommended by Questback to carefully check respondents with a quality threshold below 0.25. Since the inclusion or exclusion of participants with a quality threshold below 0.25 would be arbitrary, we decided to estimate all models only with data from respondents that passed the .25 quality threshold. The quality corrected database contains 5,370 observations with 2,548 (2,822) respondents participating in the first (second) survey.

CONSISTENCY-CHECKED CORRECTIONS

Taking advantage of the fact that a large portion of our sample took part in both surveys (i.e., 2,248 participants are panelists) we are able to perform some consistency checks in order to improve the quality of the data. Any inconsistency among panelists can be attributed either (a) to misstatements due to "slip-over" or "in-hurry" responses, (b) to the fact that eventually a different person from the same household responded, or (c) to the fact that the respondent's status indeed changed between both survey waves. The latter reason, however, seems plausible only for some characteristics such as residence.

To check the consistency of responses in our sample we focus on gender, age, nationality, residence, and the favorite club. The results from these checks as well as our treatment of observed inconsistencies are summarized in the following.

Gender. 51 out of 2,248 panelists reported a different gender in the two survey waves. Six out of these 51 respondents differ only by gender while age, nationality, residence, and the favorite club are the same in both waves. Therefore, "gender" was recoded to "missing" for these six observations. The remaining 45 panelists received a new ID for the second survey on the assumption that a different person from the same household responded. This might occur in rare cases according to Questback.

Age: 165 out of the 2,248 panelists reported a different age in the two waves. Further analysis revealed that 30 panelists are trouble free as they state different gender and already received a new ID for the second survey. A further 21 out of these 165 respondents differ only by age. In contrast to the gender question, the age question was designed as "dropdown" and therefore inconsistent age responses might be attributed to "slip-overs." Therefore, those 21 panelists who only differ by age while gender, nationality, residence, and the favorite club are the same in both waves received the mean value of both stated age values (average difference: 4.9 years) in the two survey rounds. For the remaining 114 panelists, "age" was recoded to "missing" in those models estimated with strict sample correction for "fan" and "age" (as indicated in the notes below the tables that display the logit model estimates).

Nationality: 30 out of the 2,248 panelists reported a different nationality in the different waves. Twenty of them switched between "German" and "German plus a second nationality." These 20 cases were recoded as "Germans" (i.e.. 0 [equivalent to] not German; 1 [equivalent to] German). The others switched between "German" and "other nationality." Therefore, "nationality" was recoded to "missing" for these observations.

Residence: 54 out of the 2,201 panelists with valid zip codes (some zip codes were falsely specified) stated a residence which is more than 20 km away than the previous stated residence, whereas 2,037 stated exactly the same zip code as before. We do not see any reason for further corrections here as it may be that these 54 panelists were moving houses or have intentionally stated a slightly modified zip instead of the truthful one to protect privacy. Whatever the reason behind this inconsistency is, the difference is either plausible (since approximately 14% of the German population are moving houses, which doubles the number of potential "movers" in our sample) or negligibly small.

Favorite club: 436 out of the 2,248 panelists "changed their club preference" between survey round one and two. At a first glance, this sounds dramatic and worrisome. However, the following explanation as well as the treatment for "switchers" as chosen here might probably relax this issue. In general, 17 panelists are trouble free as they state different gender and already received a new ID for the second survey. One hundred and fifteen out of the remaining 419 panelists had no favorite Bundesliga club in survey round one and switched to a favorite club in survey round two. Another 153 had a favorite club in survey round one and switched to "no favorite club" in survey round two. Both changes (from "no favorite club" to "fan of a club" and from "fan of a club" to "no favorite club") are generally plausible. Therefore, these 268 out of the 2,248 panelists remain in the sample with the differently stated club preference in each wave. For the remaining 151 panelists, "favorite team" was recoded to "missing" in those models estimated with strict sample correction for "fan" and "age" (as indicated in the notes below the tables that display the logit model estimates).

TABLE A1
Descriptive Statistics: Quality Variable

First Survey (10th matchday)

Quality             Freq.   Percent   Cum.

0.03                  4      0.13     0.13
0.04                  5      0.17      0.3
0.05                  2      0.07     0.36
0.06                 12       0.4     0.76
0.07                  6       0.2     0.96
0.08                 13      0.43     1.39
0.09                 13      0.43     1.82
0.1                  18       0.6     2.41
0.11                 19      0.63     3.04
0.12                 21      0.69     3.74
0.13                 23      0.76      4.5
0.14                 12       0.4     4.89
0.15                 26      0.86     5.75
0.16                 15       0.5     6.25
0.17                 31      1.02     7.27
0.18                 36      1.19     8.46
0.19                 40      1.32     9.79
0.2                  39      1.29     11.07
0.21                 38      1.26     12.33
0.22                 30      0.99     13.32
0.23                 39      1.29     14.61
0.24                 35      1.16     15.77
[greater than or    2,548    84.24     100
  equal to] 0.25
Total               3,025     100

Second Survey (27th matchday)

Quality             Freq.   Percent   Cum.

0.03                  0        0        0
0.04                  3      0.09     0.09
0.05                  6      0.18     0.27
0.06                  7      0.21     0.49
0.07                  8      0.24     0.73
0.08                 10       0.3     1.03
0.09                 17      0.52     1.55
0.1                   8      0.24     1.79
0.11                 21      0.64     2.43
0.12                 16      0.49     2.91
0.13                 18      0.55     3.46
0.14                 24      0.73     4.19
0.15                 21      0.64     4.82
0.16                 23       0.7     5.52
0.17                 34      1.03     6.55
0.18                 28      0.85      7.4
0.19                 38      1.15     8.55
0.2                  38      1.15     9.71
0.21                 30      0.91     10.62
0.22                 37      1.12     11.74
0.23                 44      1.33     13.07
0.24                 44      1.33     14.41
[greater than or    2,822    85.62     100
  equal to] 0.25
Total               3,297     100

Total

Quality             Freq.   Percent   Cum.

0.03                  4      0.06     0.06
0.04                  8      0.13     0.19
0.05                  8      0.13     0.32
0.06                 19       0.3     0.62
0.07                 14      0.22     0.84
0.08                 23      0.36      1.2
0.09                 30      0.47     1.68
0.1                  26      0.41     2.09
0.11                 40      0.63     2.72
0.12                 37      0.59     3.31
0.13                 41      0.65     3.95
0.14                 36      0.57     4.52
0.15                 47      0.74     5.27
0.16                 38       0.6     5.87
0.17                 65      1.03      6.9
0.18                 64      1.01     7.91
0.19                 78      1.23     9.14
0.2                  77      1.22     10.36
0.21                 68      1.08     11.44
0.22                 67      1.06     12.5
0.23                 83      1.31     13.81
0.24                 79      1.25     15.06
[greater than or    5,370    84.95     100
  equal to] 0.25
Total               6,322     100

TABLE A2
Sample Characteristics without Consistency
Checks and Quality Control

                               First Survey (10th matchday)

                                  Sample Used in the
                                     Pooled Models

                                M      SD     Min    Max

Intention to watch a game      0.26   0.44     0      1
  live on TV
Perceived suspense             5.57   2.71     0     10
Home team fan                  0.04   0.21     0      1
Away team fan                  0.03   0.18     0      1
Subj. home win probability     5.81   2.57     0     10
Single                         0.26   0.44     0      1
Female                         0.45   0.50     0      1
Age in years                    45    15.29   17     78
Distance to the venue of the   373     195    0.5    944
  home team by car (in km)

                               Second Survey (27th matchday)

                                  Sample Used in the
                                     Pooled Models

                                M      SD     Min    Max

Intention to watch a game      0.25   0.43     0      1
  live on TV
Perceived suspense             5.52   2.69     0     10
Home team fan                  0.02   0.15     0      1
Away team fan                  0.03   0.18     0      1
Subj. home win probability     5.59   2.57     0     10
Single                         0.25   0.43     0      1
Female                         0.43   0.50     0      1
Age in years                    46    14.84   17     91
Distance to the venue of the   388     198    2.4   1,036
  home team by car (in km)

                               First Survey (10th matchday)

                               Sample Used in the FE Models

                                M      SD    Min     Max

Intention to watch a game      0.25   0.43    0       1
  live on TV
Perceived suspense             5.60   2.67    0      10
Home team fan                  0.05   0.22    0       1
Away team fan                  0.04   0.19    0       1
Subj. home win probability     5.74   2.55    0      10
Single
Female
Age in years
Distance to the venue of the   376    196    1.4     930
  home team by car (in km)

                               Second Survey (27th matchday)

                               Sample Used in the FE Models

                                M      SD    Min     Max

Intention to watch a game      0.23   0.42    0       1
  live on TV
Perceived suspense             5.61   2.62    0      10
Home team fan                  0.03   0.16    0       1
Away team fan                  0.04   0.19    0       1
Subj. home win probability     5.52   2.60    0      10
Single
Female
Age in years
Distance to the venue of the   388    198    2.42   1,023
  home team by car (in km)

TABLE A3
Logit Model Estimates Including Game Dummies

                                           First Survey
Dependent Variable: 1 If the              (10th matchday)
Respondent Intends to Watch That
Game Live on TV, 0 Otherwise            Pooled          FE

Perceived suspense                     0.195 ***    0.518 ***
                                        (0.015)      (0.026)
Home team fan x perceived suspense    -0.132 ***      -0.374
                                        (0.046)      (0.244)
Away team fan x perceived suspense      -0.077        -0.354
                                        (0.051)      (0.466)
Home team fan                          1.574 ***     6.260 **
                                        (0.598)       (2753)
Away team fan                          1.316 **       5.645
                                        (0.513)      (4.294)
Subj. home win probability            -0.239 ***    -0.438 ***
                                        (0.039)      (0.072)
Subj. home win probability squared     0.022 ***    0.043 ***
                                        (0.004)      (0.006)
Home team fan x subj. home win           0.139        0.421
  probability                           (0.149)      (0.572)
Home team fan x subj. home win          -0.012        -0.040
  probability squared                   (0.012)      (0.047)
Away team fan x subj. home               0.010        14.917
  win probability                       (0.111)     (733.158)
Away team fan x subj. home win          -0.003        -1.356
  probability squared                   (0.011)      (75.888)
Single                                -0.342 ***
                                        (0.115)
Female                                  -0.079
                                        (0.084)
Age in years                            -0.032
                                        (0.020)
Age squared                              0.000
                                        (0.000)
Distance to the venue of the home     -0.001 ***    -0.006 ***
  team by car (in km)                   (0.000)      (0.001)
Distance squared                       0.000 **     0.000 ***
                                        (0.000)      (0.000)
FC Bayern Munchen vs.                  0.531 ***    2.623 ***
  Borussia Dortmund                     (0.067)      (0.196)
FC Schalke 04 vs. FC Augsburg          0.131 ***    0.647 ***
                                        (0.040)      (0.165)
Borussia M'gladbach vs.                 -0.002      0.570 ***
  TSG 1899 Hoffenheim                   (0.043)      (0.164)
1. FSV Mainz 05 vs.                     -0.036        -0.156
  SV Werder Bremen                      (0.039)      (0.179)
Hannover 96 vs. Eintracht Frankfurt    0.079 **       0.080
                                        (0.036)      (0.175)
VfB Stuttgart vs. VfL Wolfsburg          0.048        0.137
                                        (0.040)      (0.173)
Hamburger SV vs.                         0.043      0.606 ***
  Bayer 04 Leverkusen                   (0.049)      (0.171)
1. FC Koln vs. SC Freiburg              0.074 *       0.121
                                        (0.038)      (0.178)
Panel member

Constant                                -0.748
                                        (0.463)
Observations                            21,560        8,462
Number of clusters/ID                    2,415         947
Log-likelihood                        -11,188.768   -1,318.799

                                          Second Survey
Dependent Variable: 1 If the              (27th matchday)
Respondent Intends to Watch That
Game Live on TV, 0 Otherwise            Pooled          FE

Perceived suspense                     0.233 ***    0.586 ***
                                        (0.013)      (0.027)
Home team fan x perceived suspense      -0.087        0.515
                                        (0.068)      (0.352)
Away team fan x perceived suspense     -0.079 *       -0.144
                                        (0.044)      (0.139)
Home team fan                            1.295      35.464 **
                                        (0.894)      (17.728)
Away team fan                          1.423 ***    3 945 ***
                                        (0.432)      (1.247)
Subj. home win probability            -0.167 ***    -0.453 ***
                                        (0.035)      (0.067)
Subj. home win probability squared     0.015 ***    0.042 ***
                                        (0.003)      (0.006)
Home team fan x subj. home win           0.084       -9.760 *
  probability                           (0.194)      (5.070)
Home team fan x subj. home win          -0.008       0.654 **
  probability squared                   (0.015)      (0.330)
Away team fan x subj. home               0.120       0.580 *
  win probability                       (0.092)      (0.321)
Away team fan x subj. home win          -0.014        -0.034
  probability squared                   (0.010)      (0.038)
Single                                -0.475 ***
                                        (0.111)
Female                                   0.007
                                        (0.081)
Age in years                          -0.062 ***
                                        (0.020)
Age squared                            0.001 ***
                                        (0.000)
Distance to the venue of the home       -0.000      -0.002 ***
  team by car (in km)                   (0.000)      (0.001)
Distance squared                         0.000      0.000 ***
                                        (0.000)      (0.000)
FC Bayern Munchen vs.                  0.314 ***    2.093 ***
  Borussia Dortmund                     (0.061)      (0.173)
FC Schalke 04 vs. FC Augsburg           -0.050       0.373 **
                                        (0.041)      (0.168)
Borussia M'gladbach vs.                -0.100 **      0.171
  TSG 1899 Hoffenheim                   (0.041)      (0.172)
1. FSV Mainz 05 vs.                      0.000        0.178
  SV Werder Bremen                      (0.036)      (0.174)
Hannover 96 vs. Eintracht Frankfurt      0.017        0.015
                                        (0.036)      (0.178)
VfB Stuttgart vs. VfL Wolfsburg          0.009      0.457 ***
                                        (0.045)      (0.169)
Hamburger SV vs.                        -0.010      0.464 ***
  Bayer 04 Leverkusen                   (0.044)      (0.167)
1. FC Koln vs. SC Freiburg              -0.047        -0.012
                                        (0.036)      (0.179)
Panel member                           -0.217 **
                                        (0.085)
Constant                                -0.585
                                        (0.457)
Observations                            24,035        9,300
Number of clusters/ID                    2,686        1,038
Log-likelihood                        -12,115.727   -1,451.468

Notes: Matches in the second survey pair the same
teams with home and away teams flipped (reference
category: SC Paderborn 07 vs. Hertha Berlin). Models
are calculated with .25 quality threshold and strict
sample correction for "fan" and "age" (see Appendix A
for more information on this). Pooled models have been
estimated with clustered errors by individuals. FE, fixed
effects. Standard errors are given in parentheses.

Significance levels are: * p [less than or equal to] 10%,
** p [less than or equal to] 5%,
*** p [less than or equal to] 1%.

APPENDIX B: COMPETITION INTENSITY AND PERCEIVED SUSPENSE

Following Scelles et al. (2013a), Andreff and Scelles (2015), and Scelles (2017) we calculated CI measures for all relevant sub-competitions of the German Bundesliga, that is, the championship race (first place), securing a place for UEFA Champions League (second-fourth place) and UEFA Europa League (fifth-seventh place), reaching a place for the relegation play-offs (16th place) and being on the relegation zone (17th-18th place). The aforementioned CI studies have so far only focused on the points needed to reach different sporting prizes for the club which is the closest to a specific sporting prize. Since we deal with TV audience, we modified this index by including the sum of points needed to secure a sporting prize for both clubs.

Since sporting prizes differ with regard to their attractiveness and significance for the audience (Scelles et al. 2016), we implemented weights as introduced by Kringstad and Gerrard (2005) measuring 1 for the championship, 1/[1.5.sup.2] ([1/2.sup.2]) for qualifying for the UEFA Champions League (UEFA Europa League), and [1/3.sup.2] for relegation play-offs and direct relegation. Following Scelles et al. (2013b) we constrained the temporal horizon for these calculations andjust looked at whether (or not) a club is able to achieve a particular sporting prize within the next three matchdays.

In Table B1, correlations with our perceived suspense measure are reported for both unweighted and weighted CI measures as well as two different versions: Version 1 considers the points' difference of a club already achieving a sporting prize with the club closest to it. Version 2 awards 0 point to all clubs which are already in a position that secures them a sporting prize (i.e., they are awarded the highest CI value). Importantly, a higher unweighted CI score denotes a lower level of CI, whereas a higher weighted CI score denotes a higher level of CI for a single game.

A first thing to notice is that the correlations between the CI measures and perceived suspense are in general higher for the 27th matchday than for the 10th matchday, which implies that suspense is higher when more is at stake. All in all, the findings show that perceived suspense is positively (negatively) correlated with the weighted (unweighted) CI measures for the championship race. However, there is a negative (positive) correlation with the weighted (unweighted) CI measures for relegation (both play-off and direct relegation) and only a weak correlation with the CI measures for the European club competitions. These findings indicate that games involving clubs which are in championship contention are perceived to be more suspenseful, than games involving clubs which are in contention for all other sub-competitions. Moreover, it seems that perceived suspense may also reflect the quality of the contestants, since games involving clubs which are closer to the relegation zone are perceived as being less suspenseful.

TABLE B1
Correlation between Perceived Suspense and
Competition Intensity (CI) Measures

                       Unweighted CIs

Matchday    CHAMP    UCL     UEL    RPL    REL

Version 1
Both        -0.28    0.01    0.37   0.61   0.62
10th        -0.32   -0.01    0.64   0.55   0.55
27th        -0.75    0.10    0.52   0.84   0.85
Version 2
Both        -0.37   -0.05    0.35   0.61   0.63
10 th       -0.55   -0.34    0.70   0.54   0.55
27th        -0.86    0.12    0.43   0.84   0.86

                              Weighted CIs

Matchday    CHAMP    UCL      UEL      RPL     REL     SUM

Version 1
Both        0.28     0.01    -0.17    -0.42   -0.41    0.05
10th        0.48    -0.10    -0.64    -0.21   -0.17    0.17
27th         --      0.03     0.16    -0.57   -0.58   -0.44
Version 2
Both        0.75     0.12    -0.15    -0.41   -0.42    0.49
10 th       0.73     0.30    -0.70    -0.21   -0.20    0.59
27th        0.94    -0.03     0.12    -0.57   -0.58    0.90

Notes: Reported are correlation coefficients. A higher
unweighted CI score denotes a lower level of CI, whereas
a higher weighted CI score denotes a higher level of CI for
a single game. CHAMP, champion; RPL, relegation play-offs;
SUM, sum of the weighted CI measures of all sub-competitions;
UCL, UEFA Champions League; UEL, UEFA Europa League; --, not
calculated due to the small number of clubs in contention.

APPENDIX C: APPROXIMATING MARGINAL EFFECTS IN THE FIXED EFFECTS LOGIT MODEL

Let [y.sub.it] be a binary variable measuring the intention for i = l, ..., N individuals to watch any of the t = 1, ..., T games live on TV (or not). [x.sub.it] is a vector of explanatory variables (including subjective home win probability [hwin.sub.it] and its squared term [hwin.sup.2.sub.it]) and [[alpha].sub.i] is a term measuring individual heterogeneity, that is, a fixed effect, such that

(A1) [mathematical expression not reproducible].

Then the marginal effect of interest can be calculated by computing the derivative of the index function with respect to [hwin.sub.it] and [hwin.sup.2.sub.it] for each observation, that is,

(A2) [mathematical expression not reproducible]

This can be re-written as

(A3) [mathematical expression not reproducible]

(A4) [mathematical expression not reproducible]

Equation (A4) cannot be estimated since the individual fixed effects [[alpha].sub.i] are not consistently estimated in logit models. However, by substituting for Pr([y.sub.it] = 1|[x.sub.it], [[alpha].sub.i]) and Pr([y.sub.it] = 0|[x.sub.it], [[alpha].sub.i]) using the relative frequencies of intentions to watch (or not) any of the games live on TV, we are able to approximate the discrete change in the probability of TV viewership at each level of hwin, that is, the predictive marginal probability. The predicted probability is then calculated by subtracting the predictive marginal probability from each hwin-specific portion of TV viewership.

APPENDIX D: ROBUSTNESS CHECKS

TABLE D1
Logit Model Estimates without Quality and
Consistency-Checked Corrections

                                           First Survey
Dependent Variable: 1 If the              (10th matchday)
Respondent Intends to Watch
That Game Live on TV, 0 Otherwise       Pooled         FE

Perceived suspense                    0.202 ***    0.458 ***
                                       (0.014)      (0.025)
Home team fan X perceived suspense    -0.130 ***    -0.219 *
                                       (0.028)      (0.120)
Away team fan X perceived suspense    -0.091 ***     0.018
                                       (0.033)      (0.140)
Home team fan                         1.531 ***    4.669 ***
                                       (0.361)      (1.432)
Away team fan                         1.639 ***    4 212 ***
                                       (0.345)      (1.493)
Subj. home win probability            -0.264 ***   -0.370 ***
                                       (0.038)      (0.069)
Subj. home win probability squared    0.023 ***    0.037 ***
                                       (0.003)      (0.006)
Home team fan x subj. home             0.183 *       0.291
  win probability                      (0.098)      (0.359)
Home team fan x subj. home             -0.015 *      -0.024
  win probability squared              (0.008)      (0.030)
Away team fan x subj. home              0.032        -0.100
  win probability                      (0.081)      (0.429)
Away team fan x subj. home              -0.004       0.023
  win probability squared              (0.008)      (0.048)
Single                                -0.315 ***
                                       (0.100)
Female                                  -0.080
                                       (0.075)
Age in years                            -0.027
                                       (0.018)
Age squared                             0.000
                                       (0.000)
Distance to the venue of the          -0.001 ***   -0.005 ***
  home team by car (in km)             (0.000)      (0.001)
Distance squared                       0.000 **    0.000 ***
                                       (0.000)      (0.000)
FC Bayern Munchen vs.                 0.271 ***    2.403 ***
  Borussia Dortmund                    (0.058)      (0.190)
FC Schalke 04 vs. FC Augsburg         0.103 ***    0.643 ***
                                       (0.036)      (0.153)
Borussia M'gladbach vs.                 -0.036     0.641 ***
  TSG 1899 Hoffenheim                  (0.038)      (0.152)
1. FSV Mainz 05 vs.                     -0.048       -0.115
  SV Werder Bremen                     (0.034)      (0.165)
Hannover 96 vs. Eintracht Frankfurt    0.053 *       0.094
                                       (0.032)      (0.161)
VfB Stuttgart vs. VfL Wolfsburg         0.009        0.182
                                       (0.035)      (0.161)
Hamburger SV vs.                        -0.029     0.653 ***
  Bayer 04 Leverkusen                  (0.045)      (0.158)
1. FC Koln vs. SC Freiburg             0.069 **      0.171
                                       (0.033)      (0.163)
Panel member

Constant                               -0.791 *
                                       (0.418)
Observations                            26,587       10,341
Number of clusters/ID                   2,983        1,158
Log-likelihood                        -13,839.75   -1528.725

                                          Second Survey
Dependent Variable: 1 If the              (27th matchday)
Respondent Intends to Watch
That Game Live on TV, 0 Otherwise       Pooled          FE

Perceived suspense                     0.238 ***    0.553 ***
                                        (0.013)      (0.024)
Home team fan X perceived suspense     -0.090 **      -0.066
                                        (0.045)      (0.122)
Away team fan X perceived suspense      -0.037        0.061
                                        (0.033)      (0.087)
Home team fan                          1.457 **       2.689
                                        (0.582)      (1.705)
Away team fan                          0.774 **      1.778 **
                                        (0.323)      (0.774)
Subj. home win probability            -0.241 ***    -0.444 ***
                                        (0.033)      (0.061)
Subj. home win probability squared     0.020 ***    0.041 ***
                                        (0.003)      (0.006)
Home team fan x subj. home               0.008        -0.450
  win probability                       (0.139)      (0.514)
Home team fan x subj. home              -0.001        0.066
  win probability squared               (0.011)      (0.042)
Away team fan x subj. home             0.164 **       0.302
  win probability                       (0.075)      (0.223)
Away team fan x subj. home             -0.017 **      -0.021
  win probability squared               (0.008)      (0.026)
Single                                -0.466 ***
                                        (0.097)
Female                                  -0.032
                                        (0.073)
Age in years                          -0.058 ***
                                        (0.017)
Age squared                            0.001 ***
                                        (0.000)
Distance to the venue of the            -0.000      -0.002 ***
  home team by car (in km)              (0.000)      (0.001)
Distance squared                         0.000      0.000 ***
                                        (0.000)      (0.000)
FC Bayern Munchen vs.                  0.220 ***    1.922 ***
  Borussia Dortmund                     (0.056)      (0.154)
FC Schalke 04 vs. FC Augsburg          -0.066 *     0.417 ***
                                        (0.036)      (0.151)
Borussia M'gladbach vs.               -0.126 ***      0.196
  TSG 1899 Hoffenheim                   (0.037)      (0.154)
1. FSV Mainz 05 vs.                      0.004        0.172
  SV Werder Bremen                      (0.032)      (0.156)
Hannover 96 vs. Eintracht Frankfurt      0.043        0.133
                                        (0.033)      (0.158)
VfB Stuttgart vs. VfL Wolfsburg         -0.011      0.476 ***
                                        (0.040)      (0.152)
Hamburger SV vs.                        -0.015      0.475 ***
  Bayer 04 Leverkusen                   (0.039)      (0.151)
1. FC Koln vs. SC Freiburg             -0.054 *       -0.001
                                        (0.033)      (0.160)
Panel member                            -0.053
                                        (0.080)
Constant                                -0.439
                                        (0.398)
Observations                            29,040        11,073
Number of clusters/ID                    3,253        1,238
Log-likelihood                        -14,835.874   -1,835.524

Notes: Matches in the second survey pair the same
teams with home and away teams flipped (reference
category: SC Paderborn 07 vs. Hertha Berlin). Pooled
models have been estimated with clustered errors by
individuals. FE, fixed effects. Standard errors are
given in parentheses.

Significance levels are: * p [less than or equal to]  10%,
** p [less than or equal to]  5%, *** p [less than or equal to] 1%.

TABLE D2
Logit Model Estimates with Bundesliga Interest Interactions

                                          First Survey
Dependent Variable: 1 If the             (10th matchday)
Respondent Intends to Watch
That Game Live on TV, 0 Otherwise       Pooled          FE

Perceived suspense                     0.258 ***    0.636 ***
                                        (0.027)      (0.045)
High Bundesliga interest x            -0.110 ***     -0.095 *
  perceived suspense                    (0.030)      (0.053)
High Bundesliga interest               1.462 ***
                                        (0.337)
Subj. home win probability            -0.286 ***    -0.576 ***
                                        (0.081)      (0.132)
Subj. home win probability squared     0.027 ***    0.057 ***
                                        (0.007)      (0.012)
High Bundesliga interest x subj.         0.115        0.183
  home win probability                  (0.090)      (0.151)
High Bundesliga interest x subj.        -0.012        -0.020
  home win probability squared          (0.008)      (0.014)
Single                                 -0.277 **
                                        (0.116)
Female                                   0.129
                                        (0.086)
Age in years                           -0.038 *
                                        (0.020)
Age squared                             0.000 *
                                        (0.000)
Distance to the venue of the home     -0.002 ***    -0.007 ***
  team by car (in km)                   (0.000)      (0.001)
Distance squared                       0.000 ***    0.000 ***
                                        (0.000)      (0.000)
FC Bayern Munchen vs.                  0.778 ***    2.528 ***
  Borussia Dortmund                     (0.070)      (0.177)
FC Schalke 04 vs. FC Augsburg          0.164 ***    0.670 ***
                                        (0.042)      (0.159)
Borussia M'gladbach vs. TSG 1899         0.052      0.502 ***
  Hoffenheim
                                        (0.045)      (0.158)
1. FSV Mainz 05 vs.                     -0.007        -0.053
  SV Werder Bremen                      (0.040)      (0.172)
Hannover 96 vs. Eintracht Frankfurt    0.095 **       0.217
                                        (0.037)      (0.168)
VfB Stuttgart vs. VfL Wolfsburg          0.056        0.135
                                        (0.041)      (0.170)
Hamburger SV vs. Bayer 04               0.095 *     0.660 ***
  Leverkusen                            (0.051)      (0.164)
1. FC Koln vs. SC Freiburg             0.091 **      0.306 *
                                        (0.039)      (0.169)
Panel member

Constant                              -1.700 ***
                                        (0.515)
Observations                            21,560        8,462
Number of clusters/ID                    2,415         947
Log-likelihood                        -10,887.436   -1,507.088

                                          Second Survey
Dependent Variable: 1 If the             (27th matchday)
Respondent Intends to Watch
That Game Live on TV, 0 Otherwise       Pooled         FE

Perceived suspense                    0.321 ***    0.702 ***
                                       (0.024)      (0.045)
High Bundesliga interest x            -0.140 ***     -0.071
  perceived suspense                   (0.026)      (0.053)
High Bundesliga interest              1.652 ***
                                       (0.285)
Subj. home win probability            -0.194 ***   -0.761 ***
                                       (0.071)      (0.117)
Subj. home win probability squared     0.015 **    0.075 ***
                                       (0.007)      (0.011)
High Bundesliga interest x subj.        0.082       0.309 **
  home win probability                 (0.080)      (0.136)
High Bundesliga interest x subj.        -0.007     -0.034 ***
  home win probability squared         (0.007)      (0.013)
Single                                -0.456 ***
                                       (0.112)
Female                                0.230 ***
                                       (0.085)
Age in years                          -0.076 ***
                                       (0.020)
Age squared                           0.001 ***
                                       (0.000)
Distance to the venue of the home      -0.001 *    -0.002 ***
  team by car (in km)                  (0.000)      (0.001)
Distance squared                       0.000 *     0.000 ***
                                       (0.000)      (0.000)
FC Bayern Munchen vs.                 0.540 ***    1.888 ***
  Borussia Dortmund                    (0.065)      (0.159)
FC Schalke 04 vs. FC Augsburg           -0.029      0.412 **
                                       (0.042)      (0.162)
Borussia M'gladbach vs. TSG 1899      -0.090 **      0.207
  Hoffenheim
                                       (0.043)      (0.166)
1. FSV Mainz 05 vs.                     0.024       0.285 *
  SV Werder Bremen                     (0.037)      (0.167)
Hannover 96 vs. Eintracht Frankfurt     0.035        0.103
                                       (0.036)      (0.172)
VfB Stuttgart vs. VfL Wolfsburg        0.082 *      0.386 **
                                       (0.047)      (0.164)
Hamburger SV vs. Bayer 04              0.078 *     0.504 ***
  Leverkusen                           (0.045)      (0.161)
1. FC Koln vs. SC Freiburg              -0.034       0.151
                                       (0.037)      (0.171)
Panel member                           -0.141 *
                                       (0.084)
Constant                              -1.508 ***
                                       (0.504)
Observations                            24,035       9,300
Number of clusters/ID                   2,686        1,038
Log-likelihood                        -11822.567   -1,671.269

Notes: Matches in the second survey pair the same
teams with home and away teams flipped (reference
category: SC Paderborn 07 vs. Hertha Berlin); Bundesliga
interest is measured on 4-point scale. High interest = l(else 0)
if Bundesliga interest = 4; Models are calculated with .25 quality
threshold and strict sample correction for "fan" and "age"
(see Appendix A for more information on this). Pooled models
have been estimated with clustered errors by individuals.
FE. fixed effects. Standard errors are given in parentheses.

Significance levels are: * p [less than or equal to] 10%,
** p [less than or equal to] 5%, *** p [less than or equal to] 1%.

REFERENCES

Adler, M. "Stardom and Talent." American Economic Review, 75(1), 1985, 208-12.

Andreff, W., and N. Scelles. "Walter C. Neale 50 Years After: Beyond Competitive Balance, the League Standing Effect Tested with French Football Data." Journal of Sports Economics, 16(8), 2015, 819-34.

Antonakis, J., S. Bendahan, P. Jacquart, and R. Lalive. "On Making Causal Claims: A Review and Recommendations." The Leadership Quarterly, 21(6), 2010, 1086-120.

Bernheim, B. D., and A. Rangel. "Beyond Revealed Preference: Choice-Theoretic Foundations for Behavioral Welfare Economics." Quarterly Journal of Economics, 124(1), 2009, 51-104.

Budzinski, O., and T. Pawlowski. "The Behavioural Economics of Competitive Balance--Theories, Findings and Implications." International Journal of Sport Finance, 2017.

Buraimo, B., and R. Simmons. "A Tale of Two Audiences: Spectators, Television Viewers and Outcome Uncertainty in Spanish Football." Journal of Economics and Business, 61(4), 2009, 326-38.

Coates, D., B. Humphreys, and L. Zhou. "Reference-Dependent Preferences, Loss Aversion, and Live Game Attendance." Economic Inquiry, 52(3), 2014, 959-73.

Ely, J., A. Frankel, and E. Kamenica. "Suspense and Surprise." Journal of Political Economy, 123(1), 2015, 215-60.

Fort, R., and J. Quirk. "Cross-Subsidization, Incentives, and Outcomes in Professional Team Sports Leagues." Journal of Economic Literature, 33(3), 1995, 1265-99.

Geenens, G. "On the Decisiveness of a Game in a Tournament." European Journal of Operational Research, 232(1), 2014, 156-68.

Humphreys, B. R., and L. Zhou. "The Louis-Schmeling Paradox and the League Standing Effect Reconsidered." Journal of Sports Economics, 16(8), 2016, 835-52.

Jennett, N. "Attendances, Uncertainty of Outcome and Policy in Scottish League Football." Scottish Journal of Political Economy, 31(2), 1984, 176-98.

Kahneman, D., and A. Tversky. "Prospect Theory: An Analysis of Decision under Risk." Econometrica, 47(2), 1979, 263-92.

Kringstad. M., and B. Gerrard. "Theory and Evidence on Competitive Intensity in European Soccer." International Association of Sports Economists (IASE) Conference Papers (No. 0508), 2005.

Leach, S. A. "Financial Viability and Competitive Balance in English Football." Ph.D. dissertation. Imperial College London, 2006.

MacDonald, G. "The Economics of Rising Stars." American Economic Review, 78(1), 1988, 155-66.

Madrigal, R., and V. Dalakas "Consumer Psychology of Sport--More Than Just a Game," in Handbook of Consumer Psychology, edited by C. P. Haugtvedt. P. M. Herr, and F. R. Kardes. New York/London: Psychology Press, 2008, 857-76.

Mehra, S. K., and T. J. Zuercher. "Striking Out Competitive Balance in Sports, Antitrust, and Intellectual Property." Berkeley Technology Law Journal, 21(4), 2006, 1499-545.

Meier, H. E., and M. Leinwather. "Women as "Armchair Audience'? Evidence from German National Team Football." Sociology of Sport Journal, 29(3), 2012, 365-84.

Mullet, E., D. Hermand, M. T. M. Sastre, A. Nisot, and S. Rusineck. "Probability, Value, and ... Suspense." Journal of Economic Psychology, 15(3), 1994, 537-57.

Nalbantis, G., and T. Pawlowski. The Demand for International Football Telecasts in the United States. Houndmills, UK: Palgrave, 2016.

Nalbantis, G., T. Pawlowski, and D. Coates. "The Fans' Perception of Competitive Balance and Its Impact on Willingness-to-Pay for a Single Game." Journal of Sports Economics, 2015. DOI: 10.1177/ 1527002515588137.

Neale, W. C. "The Peculiar Economics of Professional Sports: A Contribution to the Theory of the Firm in Sporting Competition and in Market Competition." Quarterly Journal of Economics, 78(1), 1964, 1-14.

Ortony, A., C. L. Clore, and A. Collins. The Cognitive Structure of Emotions. New York: Cambridge University Press, 1988.

Pawlowski, T. "Testing the Uncertainty of Outcome Hypothesis in European Professional Football: A Stated Preference Approach." Journal of Sports Economics, 14(4), 2013, 341-67.

Pawlowski, T., and C. Anders. "Stadium Attendance in German Professional Football--The (un)importance of Uncertainty of Outcome Reconsidered." Applied Economics Letters, 19(16), 2012, 1553-6.

Pawlowski. T., and O. Budzinski. "The (Monetary) Value of Competitive Balance for Sport Consumers--A Stated Preference Approach to European Professional Football." International Journal of Sport Finance, 8(2), 2013, 112-23.

Raney, A. A. "Why We Watch and Enjoy Mediated Sports," in Handbook of Sports and Media, edited by A. A. Raney and J. Bryant. Mahwah, NJ: Lawrence Erlbaum Associates Publishers, 2006, 313-29.

Rosen, S. "The Economics of Superstars." American Economic Review, 71(5), 1981, 845-58.

Rottenberg, S. "The Baseball Player's Labour Market." Journal of Political Economy, 64(3), 1956, 242-58.

Scelles, N. "Star Quality and Competitive Balance? Television Audience Demand for English Premier League Football Reconsidered." Applied Economics Letters, 2017.

Scelles, N., C. Durand, L. Bonnal, D. Goyeau, and W. Andreff. "Competitive Balance versus Competitive Intensity Before a Match: Is One of These Two Concepts More Relevant in Explaining Attendance? The Case of the French Football Ligue 1 Over the Period 2008-2011." Applied Economics, 45(29), 2013a, 4184-92.

--. "My Club Is in Contention? Nice, I Go to the Stadium! Competitive Intensity in the French Football Ligue 1." Economics Bulletin, 33(3), 2013b, 2365-78.

--. "Do All Sporting Prizes Have a Significant Positive Impact on Attendance in a European National Football League? Competitive Intensity in the French Ligue 1." Ekonomicheskaya Politika/Economic Policy, 11(3), 2016, 82-107.

Schreyer, D., S. L. Schmidt, and B. Torgler. "Game Outcome Uncertainty in the English Premier League: Do German Fans Care?" Journal of Sports Economics, 2016a. DOI: 10.1177/1527002516673406.

--. "Game Outcome Uncertainty and Television Audience Demand: New Evidence from German Football." German Economic Review, 2016b. DOI: https://doi.org/ 10.1111/geer. 12120.

--. "Against All Odds? Exploring the Role of Game Outcome Uncertainty in Season Ticket Holders' Stadium Attendance Demand." Journal of Economic Psychology, 56, 2016c, 192-217.

--. "Game Outcome Uncertainty and the Demand for International Football Games: Evidence from the German TV Market." Journal of Media Economics, 30(1), 2017, 31-45.

Simon, H. A. "A Behavioral Model of Rational Choice." Quarterly Journal of Economics, 69(1), 1955, 99-118.

Szymanski, S. "The Economic Design of Sporting Contests." Journal of Economic Literature, 41(4), 2003, 1137-87.

Tversky, A., and D. Kahneman. "The Framing of Decisions and the Psychology of Choice." Science, 211(4481), 1981, 453-58.

--. "Extensional versus Intuitive Reasoning: The Conjunction Fallacy in Probability Judgment." Psychological Review, 90(4), 1983, 293-315."

Ungemach, C., N. Chater, and N. Stewart. "Are Probabilities Overweighted or Underweighted When Rare Outcomes Are Experienced (Rarely)?" Psychological Science, 20(4), 2009, 473-79.

Woisetschlager, D. M., C. Backhaus. J. Dreisbach, and M. Schnoring. "Fussballstudie 2014 - Die Markenlandschaft der Fussball-Bundesliga No. 96940." [Soccer study 2014 - The brand landscape of the soccer Bundesliga]. ZBW-German National Library of Economics; 2014.

Zillmann, D. "The Psychology of Suspense in Dramatic Exposition," in Suspense: Conceptualizations, Theoretical Analyses, and Empirical Explorations, edited by P. Vorderer, H. J. Wulff, and M. Friedrichsen. Mahwah, NJ: Erlbaum, 1996, 199-231.

Zillmann, D., and J. R. Cantor. "Affective Responses to the Emotions of a Protagonist." Journal of Experimental Social Psychology, 13(2), 1977, 155-65.

(1.) See Fort and Quirk (1995) for a detailed introduction into this topic.

(2.) In Section II, we will come back to the different conceptualizations of seasonal uncertainty (at the level of a single game) when discussing the uncertainty of outcome (UO) measures used in this study.

(3.) A detailed overview on the in-stadium attendance literature dealing with the relevance of the UOH in European professional football, that is, soccer, is provided by Pawlowski (2013).

(4.) A detailed overview on the literature about the demand for televised sports events and the relevance of the UOH is provided by Nalbantis and Pawlowski (2016).

(5.) A comprehensive review of this literature is provided by Budzinski and Pawlowski (2017).

(6.) Considering the economic theory of superstars (Adler 1985; MacDonald 1988; Rosen 1981), it might well be that a favorite away team attracts fans due to its strong brand and the opportunity to see star players. In this regard, increasing attendance with decreasing home win probability might be explained by fans either exhibiting loss aversion (Coates, Humphreys, and Zhou 2014) or having a preference for strong brands and superstars (Pawlowski and Anders 2012). However, in more recent studies, a u-shaped relation between home win probabilities and stadium attendance was found even when quality adjusted home win probabilities are used (Humphreys and Zhou 2016) and/or visiting team fixed effects are included (Coates, Humphreys, and Zhou 2014. Humphreys and Zhou 2016).

(7.) On a regular matchday with games on Friday, Saturday, and Sunday only four matches do not have competing matches occurring at the same time: Friday there is only one game starting at 8.30 p.m., Saturday there is a single game broadcast at 6.30 p.m. and Sunday there are two match broadcast, one starting at 3.30 p.m. and done one starting at 5.30 p.m. The other five matches are regularly played on Saturday afternoon at 3.30 p.m.

(8.) For an elaborated discussion about the definition of suspense see Zillmann (1996). More recently, Ely, Frankel, and Kamenica (2015) developed models distinguishing between suspense and surprise. Their context is one in which information is revealed over time, and individuals use that information to adjust their beliefs about the future. The authors define suspense as "induced by variance in the next period's beliefs" and surprise as "change from the previous belief to the current one."

(9.) Results on game dummies are available in Appendix A Table A3.

(10.) Results are available in Appendix D Tables D1 and D2 as well as Figures D1 and D2.

(11.) Questback. Enterprise Feedback Suite. EES Survey, version 9.1/1.2. Koln-Hurth: Questback GmbH, 2013.

TIM PAWLOWSKI, GEORGIOS NALBANTIS and DENNIS COATES *

* Earlier versions of this manuscript were presented at the XVII April International Academic Conference on Economic and Social Development in Moscow (Russia), at the 7th Conference of the European Sports Economics Association (ESEA) in Zurich (Switzerland), at the 90th Conference of the Western Economic Association International (WEAI) in Honolulu (USA), and at 22nd University Day of the German Society of Sport Science (DVS) in Mainz (Germany). We would like to thank conference participants for their valuable comments. Special credit is due to Jeff Borland for his valuable feedback on a preceding related research project, Bernd Frick as discussant of an earlier version of this paper, Nicolas Scelles for his feedback on how to calculate the competition intensity indices, and Arne Feddersen for his help in synthesizing the information on travel distances. We are grateful to two anonymous referees and the Co-Editor Rob Simmons for their thorough reviews and insightful suggestions. Any remaining errors and omissions are ours alone.

Pawlowski: Professor, Institute of Sports Science, University of Tubingen, 72074, Tubingen, Germany. Phone 49(0)7071-29-76544, Fax 49(0)7071-29-5031, E-mail tim.pawlowski@uni-tuebingen.de

Nalbantis: Research Assistant, Institute of Sports Science, University of Tubingen, 72074, Tubingen, Germany. Phone +49(0)7071-29-76545, Fax 49(0)7071-29-5031, E-mail georgios.nalbantis@uni-tuebingen.de

Coates: Professor, Department of Economics, University of Maryland, Baltimore County, Baltimore, MD 21250, Phone 1 410-455-2160, Fax 1 410-455-1054, E-mail coates@umbc.edu

doi: 10.1111/ecin.12462

Caption: FIGURE 1 Correlation between Subjective and Objective Measures of "Suspense" and "Game Uncertainty"

Caption: FIGURE 2 Predicted (Marginal) Probability of Watching Live for Subjective Home Win Probabilities

Caption: FIGURE D1 Predicted (Marginal) Probability of Watching Live Based on Estimates in Table D1

Caption: FIGURE D2 Predicted (Marginal) Probability of Watching Live Based on Estimates in Table D2

TABLE 1
Games Characteristics

#     Home team (#rank) (a)     Away Team (#rank) (a)

First survey (10th matchday)

1     FC Bayern Munchen (1)     Borussia Dortmund (15)
2     FC Schalke 04 (12)        FC Augsburg (9)
3     Borussia M'gladbach (2)   TSG 1899 Hoffenheim (4)
4     1. FSV Mainz 05 (6)       SV Werder Bremen (18)
5     Hannover 96 (7)           Eintracht Frankfurt (11)
6     VfB Stuttgart (14)        VfL Wolfsburg (3)
7     Hamburger SV (16)         Bayer 04 Leverkusen (5)
8     1. FC Koln (10)           SC Freiburg (17)
9     SC Paderborn 07 (8)       Hertha Berlin (13)

Second survey (27th matchday)

10    Borussia Dortmund (10)    FC Bayern Munchen (1)
11    FC Augsburg (6)           FC Schalke 04 (5)
12    TSG 1899 Hoffenheim (7)   Borussia M'gladbach (3)
13    SV Werder Bremen (9)      1. FSV Mainz 05 (11)
14    Eintracht Frankfurt (8)   Hannover 96 (14)
15    VfL Wolfsburg (2)         VfB Stuttgart (18)
16    Bayer 04 Leverkusen (4)   Hamburger SV (16)
17    SC Freiburg (15)          1. FC Koln (12)
18    Hertha Berlin (13)        SC Paderborn 07 (17)

                                   Kick-off
#     Home team (#rank) (a)       (time/day)     Obj. (b)

First survey (10th matchday)

1     FC Bayern Munchen (1)     18:30 Saturday    64.1%
2     FC Schalke 04 (12)         20:30 Friday      49%
3     Borussia M'gladbach (2)    15:30 Sunday     47.8%
4     1. FSV Mainz 05 (6)        15:30 Sunday     53.1%
5     Hannover 96 (7)            15:30 Sunday     41.3%
6     VfB Stuttgart (14)         15:30 Sunday     29.8%
7     Hamburger SV (16)          15:30 Sunday     26.3%
8     1. FC Koln (10)           17:30 Saturday     50%
9     SC Paderborn 07 (8)       17:30 Saturday    36.4%

Second survey (27th matchday)

10    Borussia Dortmund (10)    18:30 Saturday     28%
11    FC Augsburg (6)            15:30 Sunday      39%
12    TSG 1899 Hoffenheim (7)   15:30 Saturday    35.8%
13    SV Werder Bremen (9)      15:30 Saturday    39.3%
14    Eintracht Frankfurt (8)   15:30 Saturday    47.7%
15    VfL Wolfsburg (2)         15:30 Saturday    65.4%
16    Bayer 04 Leverkusen (4)   15:30 Saturday    70.3%
17    SC Freiburg (15)          15:30 Saturday    36.2%
18    Hertha Berlin (13)         17:30 Sunday     48.6%

#     Home team (#rank) (a)     Subj. (c)   Suspense (c)

First survey (10th matchday)

1     FC Bayern Munchen (1)        6.7          8.3
2     FC Schalke 04 (12)           6.6          5.4
3     Borussia M'gladbach (2)      6.7          6.1
4     1. FSV Mainz 05 (6)          6.5          5.1
5     Hannover 96 (7)              5.7          5.1
6     VfB Stuttgart (14)           4.9          5.4
7     Hamburger SV (16)            3.7          5.7
8     1. FC Koln (10)              6.3          4.8
9     SC Paderborn 07 (8)          5.6          5.1

Second survey (27th matchday)

10    Borussia Dortmund (10)       4.4          8.3
11    FC Augsburg (6)              4.6          5.7
12    TSG 1899 Hoffenheim (7)      4.2          5.6
13    SV Werder Bremen (9)         6.1          5.2
14    Eintracht Frankfurt (8)      5.9          4.9
15    VfL Wolfsburg (2)            7.3          5.4
16    Bayer 04 Leverkusen (4)      7.2          5.5
17    SC Freiburg (15)             5.2          5.0
18    Hertha Berlin (13)           6.1          4.6

                                [phi] Brand
#     Home team (#rank) (a)      Index (d)    Result

First survey (10th matchday)

1     FC Bayern Munchen (1)        62.3        2-1
2     FC Schalke 04 (12)           53.4        1-0
3     Borussia M'gladbach (2)      46.8        3-1
4     1. FSV Mainz 05 (6)          51.6        1-2
5     Hannover 96 (7)              47.6        1-0
6     VfB Stuttgart (14)           46.2        0-4
7     Hamburger SV (16)            49.7        1-0
8     1. FC Koln (10)              49.9        0-1
9     SC Paderborn 07 (8)          38.7        3-1

Second survey (27th matchday)

10    Borussia Dortmund (10)       62.3        0-1
11    FC Augsburg (6)              53.4        0-0
12    TSG 1899 Hoffenheim (7)      46.8        1-4
13    SV Werder Bremen (9)         51.6        0-0
14    Eintracht Frankfurt (8)      47.6        2-2
15    VfL Wolfsburg (2)            46.2        3-1
16    Bayer 04 Leverkusen (4)      49.7        4-0
17    SC Freiburg (15)             49.9        1-0
18    Hertha Berlin (13)           38.7        2-0

                                                   %
#     Home team (#rank) (a)     Tickets (e)   Capacity (f)

First survey (10th matchday)

1     FC Bayern Munchen (1)       71,000          99.8
2     FC Schalke 04 (12)          60,954          98.4
3     Borussia M'gladbach (2)     52,409           97
4     1. FSV Mainz 05 (6)         31,017          91.2
5     Hannover 96 (7)             42,200          86.1
6     VfB Stuttgart (14)          50,000          82.7
7     Hamburger SV (16)           52,990          92.3
8     1. FC Koln (10)             49,500           99
9     SC Paderborn 07 (8)         14,630          97.5

Second survey (27th matchday)

10    Borussia Dortmund (10)      80,667          100
11    FC Augsburg (6)             30,660          100
12    TSG 1899 Hoffenheim (7)     30,150          100
13    SV Werder Bremen (9)        41,000          97.4
14    Eintracht Frankfurt (8)     49,600          96.3
15    VfL Wolfsburg (2)           30,000          100
16    Bayer 04 Leverkusen (4)     30,210          100
17    SC Freiburg (15)            23,800          99.2
18    Hertha Berlin (13)          44,031          59.3

(a) Rank in the league table prior to
the matchday under consideration.

(b) Objective home win probabilities derived from
average margin-corrected betting odds
(source-football-data co uk)

(c) Subjective home win probabilities calculated
as sample mean values of responses to the question
"How likely do you think will there be a home win
in the upcoming GAME?" (0 [equivalent to] away club
will definitely win ... 10 [equivalent to] home club
will definitely win) with the sample applying the
.25 quality threshold and strict sample correction
for "fan" and "age" (see Appendix A for more
information on this).

(d) Average brand index of the opponents in the game
(source: Woisetschlager et al. 2014).

(e) Sold tickets for the game under consideration
(source: weltfussball.de).

(f) Percentage of stadium capacity utilization
(source: bundesliga.com).

TABLE 2
Sample Characteristics

                                    First Survey (10th matchday)

                                         Sample Used in the
                                            Pooled Models

                                     M        SD     Min    Max

Intention to watch a game           0.25     0.43     0      1
live on TV
Perceived suspense                  5.64     2.67     0     10
Home team fan                       0.03     0.16     0      1
Away team fan                       0.02     0.14     0      1
Subj. home win probability          5.88     2.56     0     10
Single                              0.25     0.43     0      1
Female                              0.47     0.50     0      1
Age in years                         47     15.38    18     78
Distance to the venue of the        373      195     0.5    938
home team by car (in km)

                                           Second Survey
                                          (27th matchday)

                                         Sample Used in the
                                           Pooled Models

                                   M        SD       Min   Max

Intention to watch a game           0.24     0.43     0      1
live on TV
Perceived suspense                  5.58     2.67     0     10
Home team fan                       0.02     0.13     0      1
Away team fan                       0.03     0.16     0      1
Subj. home win probability          5.67     2.58     0     10
Single                              0.23     0.42     0      1
Female                              0.43     0.50     0      1
Age in years                         48     14.65    17     71
Distance to the venue of the        387      198     2.4   1,036
home team by car (in km)

                                   First Survey (10th matchday)

                                   Sample Used in the Fixed
                                      Effects (FE) Models

                                     M      SD    Min     Max

Intention to watch a game          0.24    0.43    0       1
live on TV
Perceived suspense                 5.63    2.70    0      10
Home team fan                      0.03    0.17    0       1
Away team fan                      0.02    0.14    0       1
Subj. home win probability         5.80    2.56    0      10
Single
Female
Age in years
Distance to the venue of the        375    196    1.4     923
home team by car (in km)

                                        Second Survey
                                       (27th matchday)

                                   Sample Used in the Fixed
                                      Effects (FE) Models

                                   M       SD     Min    Max

Intention to watch a game          0.23    0.42    0       1
live on TV
Perceived suspense                 5.65    2.62    0      10
Home team fan                      0.02    0.14    0       1
Away team fan                      0.03    0.17    0       1
Subj. home win probability         5.58    2.62    0      10
Single
Female
Age in years
Distance to the venue of the        389    197    2.42   1,023
home team by car (in km)

TABLE 3
Logit Model Estimates

                                           First Survey
Dependent Variable: 1 If the              (10th matchday)
Respondent Intends to Watch That
Game Live on TV, 0 Otherwise            Pooled          FE

Perceived suspense                     0.195 ***    0.518 ***
                                        (0.015)      (0.026)
Home team fan x perceived suspense    -0.132 ***      -0.374
                                        (0.046)      (0.244)
Away team fan x perceived suspense      -0.077        -0.354
                                        (0.051)      (0.466)
Home team fan                          1.574 ***     6.260 **
                                        (0.598)       (2753)
Away team fan                          1.316 **       5.645
                                        (0.513)      (4.294)
Subj. home win probability            -0.239 ***    -0.438 ***
                                        (0.039)      (0.072)
Subj. home win probability squared     0.022 ***    0.043 ***
                                        (0.004)      (0.006)
Home team fan x subj.                    0.139        0.421
  home win probability                  (0.149)      (0.572)
Home team fan x subj. home              -0.012        -0.040
  win probability squared               (0.012)      (0.047)
Away team fan x subj.                    0.010        14.917
  home win probability                  (0.111)     (733.158)
Away team fan x subj. home              -0.003        -1.356
  win probability squared               (0.011)      (75.888)
Single                                -0.342 ***
                                        (0.115)
Female                                  -0.079
                                        (0.084)
Age in years                            -0.032
                                        (0.020)
Age squared                              0.000
                                        (0.000)
Distance to the venue of the          -0.001 ***    -0.006 ***
  home team by car (in km)              (0.000)      (0.001)
Distance squared                       0.000 **     0.000 ***
                                        (0.000)      (0.000)
Game dummies                           Included      Included
Panel member

Constant                                -0.748
                                        (0.463)
Observations                            21,560        8,462
Number of clusters/ID                    2,415         947
Log-likelihood                        -11,188.768   -1,318.799

                                           Second Survey
Dependent Variable: 1 If the              (27th matchday)
Respondent Intends to Watch That
Game Live on TV, 0 Otherwise            Pooled          FE

Perceived suspense                     0.233 ***    0.586 ***
                                        (0.013)      (0.027)
Home team fan x perceived suspense      -0.087        0.515
                                        (0.068)      (0.352)
Away team fan x perceived suspense     -0.079 *       -0.144
                                        (0.044)      (0.139)
Home team fan                            1.295      35.464 **
                                        (0.894)      (17.728)
Away team fan                          1 423 ***    3.945 ***
                                        (0.432)      (1.247)
Subj. home win probability            -0.167 ***    -0.453 ***
                                        (0.035)      (0.067)
Subj. home win probability squared     0.015 ***    0.042 ***
                                        (0.003)      (0.006)
Home team fan x subj.                    0.084       -9.760 *
  home win probability                  (0.194)      (5.070)
Home team fan x subj. home              -0.008       0.654 **
  win probability squared               (0.015)      (0.330)
Away team fan x subj.                    0.120       0.580 *
  home win probability                  (0.092)      (0.321)
Away team fan x subj. home              -0.014        -0.034
  win probability squared               (0.010)      (0.038)
Single                                -0.475 ***
                                        (0.111)
Female                                   0.007
                                        (0.081)
Age in years                          -0.062 ***
                                        (0.020)
Age squared                            0.001 ***
                                        (0.000)
Distance to the venue of the            -0.000      -0.002 ***
  home team by car (in km)              (0.000)      (0.001)
Distance squared                         0.000      0.000 ***
                                        (0.000)      (0.000)
Game dummies                           Included      Included
Panel member                           -0.217 **
                                        (0.085)
Constant                                -0.585
                                        (0.457)
Observations                            24.035        9,300
Number of clusters/ID                    2,686        1,038
Log-likelihood                        -12,115.727   -1,451.468

Notes: Models are calculated with .25 quality threshold
and strict sample correction for "fan" and "age" (see
Appendix A for more information on this). Pooled models
have been estimated with clustered errors by individuals.
FE, fixed effects. Standard errors are given in parentheses.

Significance levels are: * p [less than or equal to] 10%,
** p [less than or equal to] 5%, *** p [less than or equal to] 1%.