Print

Research Note: A regression that probably never should have been performed – the case of Norwegian top-league football attendance

Written by Giuseppe Alamia. Posted in 2(2)

Kjetil K. Haugen*, Arild Hervik and Hallgeir Gammelsæter
doi: 10.12863/ejssax2x2-2014x1

*Faculty of Economics, Informatics and Social Sciences Molde University College, Specialized University in Logistics Molde, Norway

DownloadAdobe-PDF

Abstract

In this research note, we discuss a peculiar development in the Norwegian top football league - ‘Tippeliagen’. Since 2007, a significant drop in attendance numbers has been observed, and we investigate possible causes through regression analysis.

Our main findings indicate that we can explain the complete attendance figures from 1993 to 2013 by a surprisingly simple regression model, including only (binary) new stadium variables. In addition, we estimate the duration of stadium attendance effects that turns out to be 5 years in Norway. In latter parts of the note, we discuss the sense in this model and conclude that a ’game-theoretic’ signaling model may prove worthwhile as an underlying explanatory model.

Keywords: Demand, Regression Analysis, Football Attendance, ‘Tippeligaen’


1 Introduction

Figura1 

Figure 1 Average match attencance in 'Tippeliagen': 1993 - 2013.

Figure 1 [Wikipedia, 2014] shows average attendance per match in the Norwegian top-league, ‘Tippeligaen’ between 1993 and 2013. As the figure indicates, a regular growth in attendance numbers is observed in the first part of the time period, while the latter part (from 2007 up to 2013) shows a regular decreasing pattern. Similar patterns are observed in other Scandinavian countries like Sweden and Denmark, although this is not a key point here. Notably however, as is observed from Figure 2 [IPAM, 2014], a similar attendance development is not observed in other and “bigger” European football leagues.

The observed decrease[1] from the top year in 2007 up to today, amounts to around 35%. Such a negative attendance development is bound to create attention. This has indeed been the case in

 Figura2

Figure 2 Attendance in main European football leagues: 2007 - 2013.

Norway the latter years, leading to extensive media coverage [NTB, 2014], [Dagbladet, 2014], [Aftenposten, 2014]. The main focus in the debate has been circling around causes and possible means to revert such an unpleasant development. Apart from obvious proposed causes related to an ongoing decrease in Norwegian international competitiveness both at national as well as at club level, the increase in TV matches, internet availability, match-day spread as well as decrease in local talent has been discussed. However, few of these discussions have targeted a peculiarity in Norwegian football in this period - the big amount of new stadiums. Our main hypothesis is to a great extent related to the stadium development and may be formulated as follows:

To what extent can football stadium development explain recent observed attendance demand variations in Norwegian football.

2 Literature overview

Attendance demand in sport has been extensively studied in research literature, see for instance the review by Borland and MacDonald [Borland and MacDonald, 2003]) as well as Noll and Zimbalist [Noll and Zimbalist, 1997] and Baade [Baade, 2008]. Borland and MacDonald point at different determinants of attendance, where ‘Quality of viewing’ is the most relevant for our case. Within this category, stadium quality or age is an important determinant. Borland and MacDonald identify strong inverse correlation between stadium age and attendance. Although a logical possibility of stadium age being a proxy for stadium capacity, McDonald and Rashcer [McDonald and Rascher, 2000] as well as Depken [Depken, 2001] report otherwise, and Borland and MacDonald’s conclusion seems sound

More recent and specific literature on this matter often refers to the effect on attendance by new stadiums as ‘novelty effects’. Not surprisingly, this research comes to a large extent from US sport economists, as shifting teams between various locations, creates needs for new stadiums to a larger extent in the US, than for instance in Europe. Especially, work by Coates and Humphreys [Coates and Humphreys, 2005] as well as Howard and Crompton [Howard and Crompton, 2003] use corresponding analytic grips as we do, also concluding with strong attendance effects from new stadiums. Coats and Humphreys also identify sport dependent (different) ‘novelty effects’ where the effect in NFL is smaller (has shorter duration) than for instance in MLB or NBA. The fact that such ‘novelty effects’ may be sport dependent seems reasonable. Feddersen et. al. [Feddersen et. al., 2006] performs similar analyses for the case of European football.

More recently, also US football (or soccer as it is named in the US) is examined by Love et. al. [Love et. al., 2013] An example on a more micro oriented approach, where the quality of viewing related to stadiums is studied for selected stadium factors are studied by Wakefield and Sloan [Wakefield and Sloan, 1993].

3. Data for the regression model

Appendix A contains all data for the regression model described in section 4. We have chosen to use aggregated data from available internet sources[2] . A possibly better alternative might perhaps have been to apply disaggregated club (and hence stadium related) data, but due to the nature of this project (research note) such a task seemed to formidable. Availability of such data and their quality is also questionable. As a consequence of these decisions, the amount of data is reasonably limited, as can be readily observed in appendix A.

4. A regression model

4.1 Model

In the regression model, we use aggregated data with attendance (ATT) as the dependent variable. The two independent numeric variables UEFA and FIFA contain Norway’s rankings at the club and national level respectively. The final STADIUM column in table 3 contains years for new stadiums. Hence, 10 new stadiums have been built or rebuilt in the 20-year time period from 1993 to 2013, a huge investment considering that ‘Tippeligaen’ contains 16 clubs.

We apply the following standard multiple linear regression model:

                              YtATT01 XtUEFA2 XtFIFA+∑(j=3)10 βj XjtSTADIUMt        (1)

The binary  variables are constructed with one variable per active stadium[3] as follows:

                                  XjtSTADIUM=(0,0,…,0,1,1,…,1,0,0,…,0)                          (2)

The first (number) 1 in (2) defines the opening year of the stadium, and the last (number) 1 defines how long the effect lasts. This definition opens up for individual stadium effect durations, but it assumes a yearly constant effect. Obviously, such a definition opens up for not only estimating stadium effects, but also the duration of such effects. It will however involve repeated regressions with different choices of stadium effect duration.

4.2 Results

As mentioned in our main hypothesis, there is a whole pile of variables that could be expected to explain attendance demand. We have included a limited set, performance (national and club) and infrastructure (stadiums). This is a deliberate choice, due to the results we actually achieved[4] . Our final relevant regression results are shown in tables 1 and 2.

 

Coeff.

Std. Error

t-Stat.

P-value

Lower 95%

Upper 95%

Intercept

5047.765

163.9171

20.79462

 

4698.384

5397.147

Colorline

1495.948

406.4261

3.680737

0.002225

629.6708

2362.671

Sør

1005.000

314.5252

3.195292

0.006022

334.6054

1675.395

AKA

1831.645

303.1560

6.041922

 

1185.483

2477.807

Viking

2581.198

374.7675

6.887464

 

1782.400

3379.995

Sarpsborg

797.3951

259.1757

3.076658

0.007674

244.9751

1349.815

Table 1 Regression output - Estimates

Multiple R

0.979352

 

0.959130

Adjusted

0.945506

Std. Error

444.8058

N

21

Table 2 Regression output – Summary statistics

There are two interesting points to note from tables 1 and 2. Firstly, no other variables than the stadium variables are included in the final regression model. The reason is straightforward, the performance variables turned out to be insignificant. This was somewhat surprising, as Norway did have a performance development (both at national as well as at club level) seemingly correlated to the development in attendance. Observe for instance the FIFA rank from table 3, and note that the average FIFA rank for Norway between 1993 and 2007 is around 20, while the average rank in the remaining time period is around 35. We even tested some lagged versions of these variables, still with no significant results.

Secondly, the value of was perhaps even more surprising. As table 2 shows a value of around 96%, it means that merely around 4% is unexplained variation. This is of course the main reason why we chose not to test alternative explanatory variables. The fact that all estimated regression coefficients are significant at least at the 99% level also indicates a surprisingly well-fit model.

Finally, in order to reach the final model, we did some tricks with the duration of the stadium variables. We chose not to open for individual stadium duration lengths, also to some extent as a consequence of the very high. Then, we tested different durations; reaching from 2 up to and including 8 and picked the model with the highest. The results of this process can be observed in figure 3.

 Figura3

Figure 3  Ras a function of stadium duration length.

As figure 3 indicates, the “regression-optimal” stadium effect duration is 5 years.

5. Discussion and conclusions

Our regression models have provided information supporting our initial hypothesis. The problem (if any) may be that this confirmation is a bit too strong. The fact that stadium effects alone can explain almost all variation (R2≈0.96) in the data was a surprise for the authors. The question we to some extent will address in this paragraph is - what does this mean?

We offer two tales to illustrate this question. These two tales offer different views on the actual underlying causality here. To some extent, Tale 2 is inspired by the somewhat surprising regression results, but we do believe that the implicit hypothesis of Tale 2 in itself is interesting and worthwhile to discuss.

Tale 1: Norway rose to unexpected and historically never before seen heights as a football nation in the period from 1992 to 1998. The country qualified to both World Cups in the period, 1994 in USA as well as 1998 in France. After beating Brazil at the group stage in 1998, Norway advanced to the knock out stages, losing to Italy for a place in the quarter finals. Simultaneously, although slightly lagged in time, the Norwegian club Rosenborg BK managed to enter the group stage in Champions League where they stayed consecutively, with the only exception in 2003, until 2005. This previously unseen development made an obvious economic boost in Norwegian football, both related to actual cash flows from UEFA and FIFA due to the described performance enhancement, but also from local sponsoring and sugar daddies. As a consequence, Norwegian football economy boosted, and big amounts of increased revenues were invested in stadiums. However, after 2005, this positive performance development reversed, and combined with increased competition from TV and other media platforms, much like the actual development in figure 1, a decrease in attendance demand in recent years should be expected.

Tale 2: The potential transactions between a football club and its potential customers (fans) can be viewed as a kind of signaling game. The supporters signal (most strongly) by showing up or not, but could also signal through media, and match behavior. The club signals back to its audience through team related decisions like buying (or selling) players, shifting coaches or managers or (ultimately) building new stadiums. If we consider these signals from clubs, it seems obvious that the strongest signal a club could send to the fans is building a new stadium Whether this investment is realized through revenues from increased performance, sugar daddies, general investors or nice ‘contracts’ with local politicians, a stadium is a strong and lasting signal. From the supporters’ point of view, such a signal demonstrates credibility of future performance enhancements, and could by itself be strong enough to boost demand. But, these effects decay, and after some time, new signals (stadiums) must sent (built) to avoid attendance demand decrease.

The problem with Tale 1 is the limited scientific support from our regression models. The fact that only stadium variables are included in the final model, and this models’ strong explanatory power indicates no support for many of the central arguments in this tale. On the other hand, the decision support derived from Tale 2 is that if Norway simply repeats a strategy of building new and modernizing old stadiums, the audience will return to the stadiums like the golden years 2005-2009. We leave this (complex explanatory decision) problem to the informed reader, but would obviously point out that our analysis, at least to some extent should be included in future decision support not only for Norwegian football managers, but perhaps also internationally. Stadium effects are surely important, perhaps more important than most experts and professionals realize.

Finally, the title of this research note may indicate that the authors, after allhave greater faith in Tale 1 than Tale 2. We might suspect that our regression really is a spurious. That is, Tale 1 tells the real story.

Still, a signaling game perspective is definitely interesting by itself, and a more formal game modeling approach along such lines may prove a relevant continuation of this work.

5. Discussion and conclusions

Whether one believes Tale 1 more than Tale 2 or vice versa, it is well documented in existing literature (see section 2) that quality of stadiums and ‘novelty effects’ provide strong and significant positive effects on demand. Our analysis documents such effects in the Norwegian case. It should come as no surprise that modern football customers value stadium quality highly; especially, today, when security issues are more important than ever. New stadiums may act both as a signal that clubs are serious performance-wise, but likewise guarantee that important security issues are well taken care of. In a world with steadily growing competing entertainment offers both within competing sports as well as other leisure possibilities, it is perhaps more important than ever to keep football stadium infrastructure at the correct quality level. As demonstrated here, new stadiums are important to draw fans, but keeping them new may be just as important, as ‘novelty effects’ decay. It ought to be evident that football officials, politicians as well as other relevant football stakeholders should be aware of these results when planning future stadium infrastructure.


Appendix A Regression data.

YEAR

ATT

UEFA

FIFA

STADIUM

1993

5542

29

4

-

1994

5216

22

8

-

1995

4624

19

10

-

1996

4622

18

14

-

1997

4242

15

13

-

1998

5270

10

14

Aker

1999

5404

17

7

-

2000

5639

13

14

Sarpsborg

2001

5567

15

26

Lerkendal

2002

6002

17

26

-

2003

6587

20

42

-

2004

8012

15

35

Viking

2005

9490

20

38

Colorline

2006

9101

19

50

Skagerak & Fosshaugane

2007

10521

18

29

Sør & Fredrikstad

2008

9812

18

59

-

2009

8966

19

32

AKA

2010

8113

22

12

-

2011

7997

26

25

-

2012

7003

26

24

-

2013

6829

27

54

-


6. References

Aftenposten. Hele skandinavia rammet av tilskuersvikt. http://www.aftenposten.no/, February 2014.

R. A. Baade. Professional sports as catalysts for metropolitan economic development. Journal of Urban Affairs, 18(1): 1–17, 2008.

J. Borland and R. MacDonald. Demand for sport. Oxford review of economic policy, 19(4): 478–502, 2003.

D. Coates and B. R.  Humphreys. Novelty effects of new facilities on attendance at professional sporting events. Contemporary Economic Policy, 23(3): 436–455, 2005.

Dagbladet. Tilskuersvikt i tippeligaen. http://www.dagbladet.no/, February 2014.

C. A. Depken. Fan loyalty in professional sports: An extension to the national football league. Journal of Sports Economics, 2(3): 124–138, 2001.

A. Feddersen, W. Maennig and M. Borcherding. The novelty effect of new soccer stadia: The case of Germany. International Journal of Sport Finance, 1(3): 174–188, 2006

D. R. Howard and J. L. Crompton. An empirical review of the stadium novelty effect. Sport Marketing Quarterly, 12(2), 111–117, 2003.

IPAM. European football attendances report 2011. http://www.ipam.pt/, February 2014.

A. Love, A. N. Kavazis, A. Morse and K. C. Mayer Jr. Soccer-Specific Stadiums and Attendance in Major League Soccer: Investigating the Novelty Effect. Journal of Applied Sport Management. 5(2): 32-41, 2013

M. McDonald and D. Rascher. Does bat day make cents? The effect of promotions on the demand for major league baseball. Journal of Sport Management, 14(1): 8–27, 2000.

R. Noll and A. Zimbalist. Sports, Jobs, and Taxes - The Economic Impact of Sports Teams and Stadiums. Brookings Institution Press, Wahington D. C., 1997.

Nettavisen NTB. Tilskuersvikt i tippeligaen. http://www.nettavisen.no/, February 2014.

H. J. Wakefield and H. J. Sloan. The Effects of Team Loyalty and Selected Stadium Factors on Spectator Attendance.  Journal of Sport Management, 9: 153-172, 1993.

Wikipedia. Tilskuertall i tippeligaen. http://no.wikipedia.org/wiki/, February 2014.



[1]The fact that an even more significant increase is observed from 1998 up to 2007 is of course also (scientifically) interesting, but has created limited media interest.

[2]All data is gathered from a multitude of internet sources including uefa.com, fifa.com as well as home pages for clubs in ‘Tippeligaen’.

[3]As can be observed from table 3, two stadiums were launched in the same year (Skagerak and Fosshaugane in 2006 as well as Sør and Fredrikstad in 2007). As a consequence, we remove one of the stadiums in each of these years. There is no point in applying perfectly correlated variables in a regression.

[4]In reality, we started out with a limited set of variables, and felt no need to include additional one’s after observing the results of our regressions.