The impact of live broadcasting on stadium attendance reconsidered: some evidence from 3rd division football in Germany

ABSTRACT

Research Question

If a sports competition is broadcasted live, consumers may opt for substituting gate attendance with watching that game live on TV (or online). This might be worrisome for teams, particularly those in lower divisions, whose game day revenues typically exceed broadcasting revenues. So far, however, the literature testing this claim empirically is inconclusive. We examine whether (at least parts of) this confusion might be traced back to shortcomings in the econometric modelling process.

Research Methods

We use attendance data for 1,138 games in German third division football from the 2015/16 to 2017/18 seasons and compare results for our demand equations between ordinary least squares (OLS) and endogenous treatment regressions (ETR). ETRs explicitly account for any selection bias, that is, the broadcasters’ preference to select the most attractive games for live broadcasting (which are expected to also attract comparably larger gate attendances).

Results and Findings

While OLS models reveal a significant positive impact of live broadcasts on gate attendance, this effect reverses when estimating ETRs. Even though there is suggestive evidence for postponing ticket demand to some extent to later games, the overall negative effect remains robust and large.

Implications

Our findings highlight the relevance of controlling for the selection bias when analysing the impact of live broadcasting on stadium attendance. From a managerial point of view, our findings suggest that increasing the number of games broadcasted live in German third division football might not be advisable, since additional broadcasting revenues may not exceed predicted losses in ticket revenues.

KEYWORDS:

• Substitution
• sports demand
• selection bias
• lower division football
• broadcasting revenues

Acknowledgements

The authors would like to thank the participants of the 11th Conference of the European Sport Economics Association in Gijón (Spain) for valuable comments on an earlier draft of this article. In addition to this, they would like to thank two anonymous referees who significantly helped to improve this article.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Notes

1 For instance, for third division football in Germany covered in this research, game day earnings comprise about 22 percent of total revenues, while revenues from media rights cover merely 14 percent (DFB, 2018).

2 Cross tabulations for each season show that high-ranked teams are broadcasted comparably more often throughout the seasons. Moreover, we do not find any evidence for a time trend in this selection (such as that broadcasters explicitly select less attractive teams / games early in the season or else).

3 The gross sample includes 1,140 games. Two game are removed from the analysis. One game was played in absence of spectators as a sanction measure imposed by the Deutsche Fußball Bund. For another game we are missing information on weather conditions. Accordingly, 1,138 observations remain in the sample.

4 Note that teams in our data always play in the same stadium and do not relocate home games to other venues.

5 Note that for the 2017/2018 season all games were additionally also available on pay TV. To avoid any confounding effect in our setting, we control for season dummies as explained further below.

6 Since very few clashes with UEFA Champions League or UEFA Europa League games involving Bundesliga clubs occurred during our observation window, we could not test rivalry from international club competitions.

7 Since regular football games include two halves with 45 min and a halftime break of 15 min, this variable takes the value of ‘1’ for games that were played within 105 min before or after the kickoff time of Bundesliga games.

8 As a robustness check, we also estimated a specification not complying with the exclusion restriction. Moreover, while estimation by maximum likelihood is the most efficient estimation procedure (Tucker, 2010; Wooldridge, 2010), we re-estimated these models using the Two-Step estimator. Our main findings remain with regard to both robustness checks (see Table A1 in the Appendix A).

9 Since we interacted the treatment variable with a control variable, the Stata command etregress does not directly estimate the average treatment effect (ATE). We use margins to estimate ATEs from the results of Table 5.

10 As a robustness check, we re-estimated all models using games broadcasted live on TV only (instead of TV or online stream). Moreover, estimating ETR we tested for interactions between Broadcast and Bundesliga as well as between Broadcast and the weather variables. While we find significant interaction effects for Bundesliga and Temperature, the results suggest that our main findings remain (see Table A2 and A3 in the Appendix A).

11 We also tested effects for the subsequent first and second home game after a broadcasted home game and still find increased demand. Results are available upon request.

12 In Germany, for instance, soccerwatch.tv and sporttotal.tv even broadcast German soccer games of low amateur leagues.