You are on page 1of 11

NHL Attendance: Does

winning really matter?

Credit where credit is due


Team Statistics come from: http://www.opensourcesports.com/hockey/
Attendence data comes from:
Rodney Forts Sports Economics
https://umich.app.box.com/s/41707f0b2619c0107b8b/1/320029199/2560929897/1
https://sites.google.com/site/rodswebpages/codes

Hypothesis :Sports Attendance is correlated with team


performance
Sporting event attendance is affected by many factors
For individual teams factors such as current perceptions, may be significant but
not homogeneous across all observations.
No one likes to see their team lose
Fans are less likely to go to a game where they believe their team will lose

Murky data
Attendance at games is a factor of both teams competing
Unable to obtain attendance of just one team
When the data was adjusted initially, to only include data where the playoff ranking
was available, the number of usable observations remaining was low(274).
When the data is balanced the number of usable observations shrinks even
further(100)
Small sample size(~1000)

Data Concatenation
Data was collected from two sources.
The first source (open source sports) had data on hockey
teams from 1906 to 2010
The second source (Rodney Forts Sports Economics) had
data on attendence from 1952-2012
Initially all observations occurring outside the interval of
1952-2010 were dropped. This was done so that a 1:1
merge could be performed.
Then the command xtset name year was performed to
convert the data into panel data.

Interesting tabulations of variables & their interpretations


Note: all of these tabulations were taken before balancing on name and year.
tab rank
1.
2.

rank |

Freq.

Percent

Cum.

------------+-----------------------------------

3.

1|

177

18.40

18.40

4.

2|

180

18.71

37.11

5.

3|

180

18.71

55.82

6.

4|

181

18.81

74.64

7.

5|

168

17.46

92.10

8.

6|

54

5.61

97.71

The Regression
xtreg teamaverageattendancepergame pctwl L1.(rank) L(1/3).(playoffc) ,r

Regression Results (unbalanced data)


Unbalanced Data:
Random-effects GLS regression
Group variable: nameB
R-sq:
within = 0.3232
between = 0.2739
overall = 0.2420
corr(u_i, X) = 0 (assumed)

Number of obs =
Number of groups =

Obs per group:


min =
avg =
max =

264
29

1
9.1
28

Wald chi2(5)
= 119.48
Prob > chi2
=

0.0000

-----------------------------------------------------------------------------teamaverag~e |
Coef. Std. Err.
z P>|z| [95% Conf. Interval]
-------------+---------------------------------------------------------------pctwl | 689.3315 134.8733 5.11 0.000 424.9848 953.6783
|
rank |
L1. | -192.1163 116.5086 -1.65 0.099 -420.469 36.23642
|
playoffc |
L1. | -168.2337 47.78129 -3.52 0.000 -261.8833 -74.58409
L2. | -85.85899 51.21385 -1.68 0.094 -186.2363 14.51832
L3. | -226.2054 49.08902 -4.61 0.000 -322.4182 -129.9927
|
_cons | 17553.48 531.8035 33.01 0.000 16511.17 18595.8
-------------+---------------------------------------------------------------sigma_u | 1474.9191
sigma_e | 1595.7286
rho | .46071747 (fraction of variance due to u_i)

Conclusion
Team average attendance is correlated with current win/ loss percentage, the rank
for the last three years and the teams playoff ranking for the last three years.
Average attendance is affected by many variables so there is a strong possibility
of omitted variable bias
More data is needed for a more balanced analysis

You might also like