NCAA football television viewership: Product quality and consumer preference relative to market expectations

Highlights

•	No support for preference for anticipated outcome uncertainty.
•	Consumer preferences exist for contests which are closer than market expectations.
•	Preferences appear to fluctuate based on contest characteristics.
•	Preference for anticipated absolute quality dominates preference for relative quality.
•	Findings suggest current scheduling practices may be sub-optimal.

Abstract

The authors estimate the determinants of college football television viewership across the full quality spectrum of contests and test whether consumer preferences vary based on changes in the attributes of the core product. They utilize national television viewership data at the individual game level over a three season period and estimate numerous consumer demand models using zero-truncated negative binomial regression. The results indicate a lack of support for anticipated outcome uncertainty, but support for contests where actual outcomes are closer than market expectations. Consumer preferences are not consistent across game qualities, which may indicate that game type is linked to variation in the consumer base and reference-dependent preferences. The findings may also explain why the uncertainty of outcome hypothesis is supported in some contexts, but not others. Preference for absolute quality also dominates preference for relative quality. This finding has important implications for contest scheduling. Given the common practice of advance scheduling creates sub-optimal conference and network television schedules, stakeholders could be leaving television revenues on the table.

Keywords:

• Television viewership
• College football
• Consumer preference
• Product quality
• Outcome uncertainty
• Behavioral economics

Acknowledgement

We would like to thank two anonymous reviewers and Associate Editor Pamela Wicker for their helpful comments.

Notes

1 The BCS National Championship game and College Football Playoff games are not included in the sample, as these contests have direct national title implications. Additionally, Nielsen does not generate viewership numbers for games on all cable channels – for example, conference and regional sports networks. Therefore, contests broadcast on these particular channels are not included in our sample.

2 Bruce Payne’s power ratings can be found at http://knology.net/~brucepayne. Payne’s ratings are a weighted composite of two individual sets of ratings – a pure win/loss rating and a predictive margin of victory rating. Ratings update after the conclusion of every week of play and higher composite values denote higher quality teams and vice versa.

3 The correlation between AvgWeekPayne and AvgPowerRtg is 0.6001 and both variables have variance inflation factors below 2.50; a value which is within accepted limits with respect to collinearity (Kutner, Nachtsheim, & Neter, 2004).

4 The correlation between ClosingSpread and ScoreSpread is 0.0136, which eliminates concern of collinearity.

5 We also tested a linear time term and jointly tested a linear and square time term. The indicator approach explained more variation in the relationship between viewership and start time and was therefore preferred.

6 We include a second baseline model due to the fact that the inclusion of conference indicator variables render all non-conference matchups as the baseline category. Given that the quality of non-conference matchups can vary considerably, this baseline could cause noise in estimates of the conference indicators. The inclusion of the second baseline model serves to confirm the robustness of the results presented in the first baseline model.

7 For indicator variables, the marginal effect displayed is the effect of the specified variable in reference to the baseline category.

8 Bowl games naturally differ in quality, however, the estimated effect illustrates the average bowl game viewership premium above and beyond a regular season matchup with the same contest characteristics.

9 Model 7 includes only contests with point spreads of 20.5 points or greater. Model 8 includes contests with spreads between 11.5 and 20. Model 9 includes contests with spreads between 2 and 11, while Model 10 includes games with spreads of 1.5 or less.

10 Existing literature in Major League Baseball has identified that in local markets, consumer demand is maximized when the probability of a home team win is approximately 0.60 (i.e. – Knowles, Sherony, & Haupert, 1992; Whitney, 1988). Investigation of the point of optimal contest uncertainty to illicit maximum national television viewership is not well developed