Sei sulla pagina 1di 7

Canadian Public Policy

Performance and Salary Differentials in the National Hockey League


Author(s): Robert C. McLean and Michael R. Veall
Reviewed work(s):
Source: Canadian Public Policy / Analyse de Politiques, Vol. 18, No. 4 (Dec., 1992), pp. 470-475
Published by: University of Toronto Press on behalf of Canadian Public Policy
Stable URL: http://www.jstor.org/stable/3551660 .
Accessed: 09/12/2011 12:40
Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at .
http://www.jstor.org/page/info/about/policies/terms.jsp
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of
content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms
of scholarship. For more information about JSTOR, please contact support@jstor.org.

University of Toronto Press and Canadian Public Policy are collaborating with JSTOR to digitize, preserve
and extend access to Canadian Public Policy / Analyse de Politiques.

http://www.jstor.org

Performanceand
Differentials in

Hockey

Salary
the

National

League*

ROBERT C. MCLEAN and MICHAEL R. VEALL


McMaster University

Lavoie, Grenieret Coulombe(1987)utilisent des donnbesconcernantun certain nombre de saisons


de la LNHjusqu'd 1983-84 et concluentque la performancedesjoueurs francophonesest supbrieure
A celle des joueurs anglophones, ce qu'ils interprbtent comme une preuve de discrimination A
l'embauche. Ce texte r6plique leurs calculs pour la saison 1989-90 et trouve des diff6rentiels de
performancedu mame signe, mais plus petits et n'6tant plus significatifs au niveau de 5 pour cent.
Nous examinons 6galement les donn6es salariales pour la saison 1990-91, en utilisant des outils
empiriques similaires et les mbmes standards de performancestatistique, et ne trouvons aucune
trace de discriminationsalariale.
Lavoie, Grenier and Coulombe (1987) use data for a number of NHL seasons up to 1983-84 and
conclude that FrancophoneCanadianplayers outperformedAnglophoneCanadian players, which
they interpret as evidence of hiring discrimination.This paper replicates their calculations for the
1989-90 season and finds performancedifferentials of the same sign, but smaller and no longer
statistically significant at the 5 per cent level. We also examine salary data for the 1990-91 season
and find no evidence of salary discrimination,using similar empiricaltools and the same standard
of statistical significance.
I Introduction
iscrimination in labour markets can be
salary discrimination or hiring (entry)
discrimination. One place to study both
types is professional sports, where there are
publicly-available statistics which attempt
to measure individual performance. For example, Marc Lavoie, Gilles Grenier and
Serge Coulombe (1987: henceforth LGC)
study entry discrimination in the National
Hockey League using data for a number of
seasons up to 1983-84. They find that Francophone Canadian players outperform Anglophone Canadian players, which they
interpret as evidence of hiring discrimina-

tion against Francophones. Using data


from the 1977-78 season, J.C.H. Jones and
William Walsh (1988) found possible evidence of salary discrimination against
Francophones, but only at the position of
defense.
This paper continues the study of discrimination against Francophone Canadians in the NHL. Section II uses data for the
1989-90 season and again finds evidence of
a performance differential, although a
smaller one than that found by LGC. Section III examines newly-released salary
data but finds no evidence of salary discrimination.

Canadian Public Policy - Analyse de Politiques, XVIII:4:470-475 1992 Printed in Canada/lmprim6 au Canada

Table 1
Averageattributes and lifetime performanceby position and origin, National HockeyLeagueplayers, 1989-90
Position
Forwards
Number of players
Percentage
Age (years)
Height (feet)
Weight (pounds)
Draft choice*

Anglophone
Canadian

Francophone
Canadian

American

181
60.1
26.4
6.03
194.2

36
12.0
26.1
6.01
193.6

53
17.6
26.4
6.01
190.0^ ^

31
10.3
27.7^ ^
6.00
190.4^

55.1

45.0

72.8 ^ ^

117.0^ ^

Seasons
Penalty min./game
Points/game

6.17
1.43
0.598

6.17
1.33
0.706 ^ ^

Defensemen
Number of players
Percentage
Age (years)
Height (feet)
Weight (pounds)
Draft choice*
Seasons
Penalty min./game
Points/game

93
58.5
26.7
6.12
200.3
54.6
6.62
1.61
0.352

19
11.9
26.1
6.04 ^ ^
198.4
33.0 ^
6.26
1.16^ ^
0.364

26
16.4
26.4
6.10
199.1
63.3
5.58 ^
1.54
0.452 ^ ^

Goaltenders
Number of players
Percentage
Age (years)
Height (feet)
Weight (pounds)
Draft choice*
Games played
Goals against average

36
72.0
27.6
5.90
179.6
76.1
217.6
3.67

8
16.0
27.4
5.81
174.3
66.3
169.5
3.43 ^ ^

6
12.0
26.5
5.89
177.0
48.0 ^170.0
3.41^ ^

4.79^ ^
1.16^
0.556

European

4.45^^"
0.57^ ^
0.678 ^
21
13.2
26.8
6.17^ ^
202.4
91.7^ ^
4.00 ^ ^
1.10^ ^
0.396

*playerswho were not draftedwere removedfrom calculation.


^ significantlydifferent from AnglophoneCanadianat 10% level.
^ ^ significantlydifferent from AnglophoneCanadianat 5% level.
SOURCE:
NHL Official Guide and RecordBook, 1990-91edition.

II Performance

Differential

LGC calculate average performance ratios


for Francophone Canadian players to Anglophone Canadian players, measuring performance as average points per game for
defensemen and forwards and as inverse
goals against average for goalies. For 198384 they calculate these Francophone/Anglophone performance ratios as 1.41, 1.21
and .98 for defensemen, forwards and goalies respectively. Compiling the data to be
as comparable to LGC as possible,1 our
corresponding numbers for 1989-90 are

1.03, 1.18 and 1.07. LGC argue that their results suggest hiring discrimination because
the average Francophone defenseman or
forward outperforms the average Anglophone, implying that team output could be
increased by hiring a new Francophone to
replace a poorly-performing Anglophone.
The 1989-90 data are still consistent with
this hypothesis although the performance
differential is now small except for forwards.2
LGC also use a regression approach to
study hiring discrimination for defensemen
and forwards and we re-estimate their

Discrimination in the NHL 471

equations on the 1989-90 data. Table 1 provides the 1989-90 averages of the variables
used in these regressions. The values for
Anglophone and Francophone players are
typically similar. One difference is that Anglophone players are drafted later on average, which is consistent with the LGC hypothesis. The Francophone/Anglophone
differences are smaller than those reported
by LGC for 1983-84, however.
In order to determine whether there is
evidence of discrimination after the draft
(i.e. discrimination conditional on draft
position), for forwards and defensemen
LGC regress players' lifetime points per
game (POINTS) on the draft choice number (DRAFT) as well as DRAFT2 (DRAFT
squared) and dummy variables for position
(DEFENSE) and origin (FREN). Estimating the equation on the 1989-90 data yields:
POINTS = 0.733 - 0.00411 DRAFT + 0.0000167 DRAFT2
(3.16)
(4.28)
+ 0.064 FREN
- 0.270 DEFENSE
(7.78)
(1.43)
N = 288
R2 = 0.234

In this and all subsequent reported regressions, absolute t-statistics are in parentheses.
This regression yielded results generally
similar to those obtained by LGC. However,

controlling for draft choice, it is now estimated from the FREN coefficient that
Francophones score about .064 points per
game or five points per season more than
Anglophones compared to the nine point
per season differential estimated by LGC.
Also the LGC estimate was statistically significant at the 5 per cent level while our
estimate is not.
While the above regression and the one
run by LGC used only Canadian players, we
added non-Canadian players plus origin
dummies AMER for American and EURO
for European and obtained:
POINTS = - 0.698 0.00298 DRAFT + 0.0000102 DRAFT2
(2.95)
(4.21)
+ 0.066 FREN
- 0.239 DEFENSE
(1.52)
(8.39)
+ 0.050 AMER
+ 0.094 EURO
(2.09)
(1.29)
N = 401
R2 = 0.199

472 RobertC. McLeanand Michael R. Veall

One can see that the results are very similar to those found using Canadian players
exclusively. It is interesting to note that the
coefficients on AMER and EURO suggest
that both the Americans and foreign players may outperform Anglophone Canadians, and in the case of EURO, the difference
is significant at the 5 per cent level.
Table 1 also indicates that Francophone
players average fewer lifetime penalty
minutes per game than Anglophone players
(although the tendency is less pronounced

than it was in the LGC1983-84 data). LGC


used this variable (called 'ROUGH' here)
as well as a player's height (HEIGHT) to
control for ability at the physical part of
hockey in a specification also including as
attributes age (AGE) and age-squared
(AGE2).
Re-estimating this equation on the 198990 data (including nondrafted players) we
obtained:
POINTS = 1.43 - 0.198 HEIGHT- 0.099 ROUGH
(2.32)
(8.10)
- 0.000150 AGE2
+ 0.0231 AGE
(0.21)
(0.58)
- 0.198 DEFENSE + 0.0502 FREN
(1.29)
(7.30)
- 0.0191 AMER
- 0.0183 EURO

(0.57)
N = 460

(0.45)
R2 = 0.305

Again these results are qualitatively


similar to those of LGC. Controlling for
height, roughness and age, we estimate
that Francophones outperform Anglophones by .05 points per game or about four
points per season compared again to an
estimate of nine points a season from the
corresponding LGC regression. Again the
LGC estimate is statistically significant at
the 5 per cent level while our estimated
coefficient is not (nor are our estimated
coefficients of the other origin dummies).
Summing up the results so far, the
empirical results for the 1989-90 data are
generally similar to those of LGC from earlier data. However while the estimated performance differentials are generally of the
same sign as in LGC, they are typically
smaller and not statistically significant at
the 5 per cent level. Following LGC in in-

Table 2
Salary averages (standarddeviations)by position and origin, National Hockey League, 1990-91
Anglophone
Canadians

Francophone
Canadians

Americans

Europeans

Total

257940
(262950)
230770
(138970)

301560
(321660)
243950
(215090)

198510
(79910)
262230
(137800)

232940
(111120)
234520
(115350)

250120
(238230)
237990
(146080)

Goalies

251670
(106370)

201250
(134580)

247500
(119320)

Total

249060
(217850)

271440
(274370)

221460
(106520)

Forwards
Defense

(---)
(---)

242550
(110590)

233580
(111720)

245650
(195460)

SOURCE:
The HockeyNews (1991).

terpreting these differentials as indicating


hiring discrimination, the evidence would
suggest that hiring discrimination against
Francophone Canadians has become less
severe.
III Testing For Salary
Discrimination
Studies of salary discrimination in baseball
and other team sports have concluded that
little or none existed (see Kahn, 1991 for a
survey). There are many different methods
of testing for fair compensation. To begin,
Table 2 presents salary averages and standard deviations for those players from our
1989-90 sample who were still active in
1990-91.3 While none of the differences is
statistically significant at the 5 per cent
level, the overall average of Francophone
Canadian salaries is the highest.

ready defined as well as POINTS2


(POINTS squared) and a relative plusminus
variable
RPLUSMINUS.
RPLUSMINUS takes the player's individual plus-minus statistic and subtracts
the team's average plus-minus value.4 This
is done because a talented defenseman on a
poor team may have a negative plus-minus.
The results are:
LNSAL= 0.374 + 0.262 POINTS + 0.982 POINTS2
(0.78)
(3.15)
[0.81]
+ 0.151 AGE

Again it seems appropriateto use a regression methodology to attempt to control


for other variables which may influence
salaries. In all cases, the dependent variable is the national logarithm of the player's
salary in thousands of dollars (LNSAL). We
have tried a number of such specifications
for each of defensemen, forwards and goaltenders and in no case found a coefficient
on the Francophone Canadian variables
that was statistically significant at the 5 per
cent level. For defensemen, a reasonably
simple specification with acceptable goodness-of-fit includes several variables al-

[2.98]
- 0.00265 AGE2

(1.94)
(1.96)
[1.53]
[1.46]
+ 0.515HEIGHT + 0.000353 ROUGH
(3.03)
(2.62)
[2.95]
[4.43]
+ 0.00591RPLUSMINUS
(4.99)
[3.00]
+ 0.0694 FREN
+ 0.0274 AMER
(0.94)
(0.42)
[1.39]
[0.38]
+ 0.0390 EURO

(0.53)
N = 159

[0.40]

R2 = 0.6125 RESET (2,3,4) = (.35, .61, .79)

Because the null hypothesis of hois rejected using the


moskedasticity
Breusch-Pagan-Godfrey test (see e.g., Maddala, 1992:207-9), we also report heteroskedastic-consistent
t-statistics (White,
1980) in square brackets. Because the specifications were new, we also report RESET
specification tests (Ramsey, 1969) of the
second, third and fourth order. These are
F-tests and the null hypothesis of no-specification is not rejected at the 5 per cent

Discrimination in the NHL 473

level.
Point production and age both tend to increase salary. As expected, height also
benefits the defenseman financially. The
ROUGH coefficient is positive, presumably
because penalty minutes is a proxy for
physical play, and the RPLUSMINUS
coefficient is positive. The coefficients of
the Francophone Canadian dummy variable and the other origin dummies are not
significant at the 5 per cent level.
Some of the regressors in the defensemen equation seem potentially less important for forwards, and when the forward
data are used, their coefficients have either

wrong signs or small t-statistics. As it has


no effect on our empirical conclusions with
respect to the Francophone Canada wage
differential, we exclude those variables and
present the simpler regression equation:5
LNSAL = 0.848 + 0.473 POINTS + 0.417 POINTS
(4.59)
(3.13)
[4.29]
12.94]
- 0.0047 AGE2
+ 0.279 AGE
(4.49)
(4.88)
[6.18]
[5.671]
- 0.107 AMER
-.00353 FREN
(0.06)

[0.06]

(2.21)

[2.50]

- 0.124 EURO

(2.05)
11.74]
N = 301 R2 = 0.6355 RESET (2,3,4) = (4.10, 2.23, 1.50)

Again higher point production tends to


yield a higher salary and an older player
generally gets compensated for his years.
The coefficient of the dummy variable for
Francophone is extremely small and not
significiant, suggesting there is no salary
discrimination against Francophone forwards. However the coefficients on AMER
and EURO are negative and significant,
which could be interpreted as discrimination.6
The value of the goaltender may possibly
depend on many things but a simple model
using either wins per season or games
played per season seems adequate. Presenting results using the latter variable:
LNSAL = 4.654 + 0.0254 GPS -0.162 FREN
(1.30)
(6.23)
[1.19]
[6.62]

474 RobertC. McLeanand Michael R. Veall

- 0.0051 AMER

(0.04)
[0.04]
N = 50 R2 = 0.4938 RESET (2,3,4) = (.18, .39, .92)

Notice that EURO is omitted as only one


European goalie played in the NHL in
1989-90; that observation is also omitted.
As expected, more games played tends to
raise salaries. In fact, one more game
played per season than the average predicts
a raise of about $6,000. While the coeffi-

cient of the Francophonedummyis not significant at the 5 per cent level, this may be
lack of statistical power because of the
smaller sample of goaltenders; the coefficient itself is large and negative.
As discussed, in all of the above cases we
attempted a wide variety of specifications.
The above were chosen for presentation because they were reasonably parsimonious
yet we judged the goodness-of-fit to be satisfactory. In no case did we find a statistically
significant negative coefficient on the
FREN dummy. Lavoie and Grenier (1991)
use 1989-90 data with more complex specifications yet find the same conclusion.
There appears to be little evidence of salary
discrimination
against Francophone
Canadians in the NHL.
To summarize, no evidence has been
found of discrimination against Franco-

phone Canadianswhen it comes to compensation for performance. As mentioned earlier, this result is consistent with those
found in previous studies of salary discrimination in team sports.
IV Conclusion
Considering the National Hockey League
as a labour market, one can test for the existence of discrimination. The empirical
findings of Lavoie, Grenier, and Coulombe
(1987), using data up to 1983-84 which may
be interpreted as suggesting hiring discrimination, are confirmed for the 1989-90
data. However, the differentials appear to
have fallen and are seldom statistically significant at the 5 per cent level. This change
may be due to market pressure given the

return to winning and/or it may be associated with an increased number of Francophone coaches and general managers, as
suggested by Lavoie and Grenier (1991). In
addition, a similar method and newly-available data are used to test for salary discrimination, but little evidence of this is
found.
Notes
*

4
5

This researchwas begunin spring1991by McLean


as a projectfor Economics767 at McMasterUniversity. Valuable comments were provided by
LonnieMagee(who co-taught767 with Veall) and
participants in the McMaster 'Long Labour
Lunch' seminar in May 1991.In a workingpaper
published at the same time, Lavoie and Grenier
(1991)independentlyobtainedvery similarresults
using very similardata.Wethank the refereesand
the Editor of this journal for their valuable help
and comments.
To be includedin the data set, playersmust have
played 15 games during the 1989-90 season (for
goalies,this restrictiondroppedto four if they had
played20 careergames). In most cases, the origin
(and original language) of the player is wellknown. In the case of the few Canadianswe had
difficulty assigning, a player was counted as a
Francophoneif drafted from the Quebec Major
Junior Hockey League and Anglophone otherwise. As we have noted, Lavoieand Grenier(1991)
obtain similar results using their classification.
Our classificationis availableon request.
The change in the performancedifferentialsare
also consistent with the changes in the proportions of Francophoneplayers. Between 1983-84
and 89-90, the percentageof Francophoneplayers
rose from 7.2 to 11.9%for defensemen and from
9.9 to 12%for forwards,both of whose averageperformance ratios fell. The percentage of Francophones fell from 27 to 16% for goalies, whose
averageperformanceratio rose.
The salaries included are those for the 1990-91
season since it is assumed players' salaries are
based on previousperformance.The salaries are
base salarieswith incentivesaddedif the incentive
target was reachedthe previousyear.
A team's averageplus-minusis an averageof the
individualplus-minusvalues on a team.
The only RESETrejectionfor the salaryequation
is for RESET(2) and the forward equation. We

therefore re-ran the forward equation omitting


Wayne Gretzkyand MarioLemieux;the RESET
test statistics became very small but the origin
dummy coefficients and t-statistics were almost
unchanged.
Unlike our finding in the Francophone wage
differentialthe coefficientson AMERand EURO
are sensitive to specificationand in some specifications these coefficients are not significantly
differentfrom zero at the 5%level. In any case, as
a referee points out, our finding only indicates
that an adjustedmean salarydifferentialbetween
Canadians and these other groups is not accountedfor by the regressorsin this specification.

References
The Hockey News (1991) 'Bottom Line On Salaries,' The Hockey News, 44:21:46-47.
Jones, J.C.H. and William D. Walsh (1988)
in the National
'Salary Determination
Hockey League The Effects of Skills, Franchise Characteristics, and Discrimination,'

Industrial and Labor Relations Review,


41:4:592-604.
Kahn, Lawrence M. (1991) 'Discrimination in
Professional Sports A Survey of the Litera-

ture.' Industrial and Labor Relations Review, 44:3:395-418.


Lavoie, Marc and Gilles Grenier (1991) 'Discrimination and Salary Determination in the
National Hockey League.' Working Paper
No. 9110E, University of Ottawa.
Lavoie, Marc, Gilles Grenier and Serge
Coulombe (1987) 'Discrimination and Performance Differentials
in the National
Canadian
Public Policy Hockey League,'

Analyse de Politiques, XIII:4:407-22.


G.S. (1992) Introduction
to
Maddala,
Econometrics (New York: Macmillan).
National Hockey League, Official Guide and Re-

cord Book (1990-91).


Ramsey,J.B. (1969) 'Tests for SpecificationErrors in Classical Linear Least Squares Regression Analysis,' Journal of the Royal
Statistical Society, Ser. B, 2:350-71.
White, H. (1980) 'A Heteroskedasticity-Consistent Covariance Matrix Estimator and a

Direct Test for Heteroskedasticity,'


Econometrica, 48:817-38.

Discrimination in the NHL 475

Potrebbero piacerti anche