Prediction of Skating Speed with Off-Ice

Testing in Professional Hockey Players

Teresa Mascaro, PT, ATC '
Bill 1. Seaver, PhD2
Lorna Swanson, MFA, PT, ATC3

I
ce hockey is rapidly develop The purpose of this study was to determine the best off-ice predictors of sprint skating speed
Downloaded from www.jospt.org at on November 8, 2015. For personal use only. No other uses without permission.

ing a reputation as the fastest
sport in professional athletics.
It is essential for the profes-
sional hockey player to pos-
sess superior speed skating ability
from goal line to goal line. Hockey is
Copyright © 1992 Journal of Orthopaedic & Sports Physical Therapy®. All rights reserved.
in professional ice hockey players. Off-ice functional tests included a 40-yd dash, standing long
jump, and vertical jump, as well as isokinetic testing of the quadriceps and hamstrings at speeds of
60 and 180". Skating speed was determined by a 54.9-m sprint skate test. Nine professional ice
hockey players served as subjects. Skating speed was significantly correlated with vertical jump
anaerobic power as determined by the lewis formula and with several isokinetic power measure-
ments. The single best predictor of skating speed was vertical jump anaerobic power. An even better

primarily considered an anaerobic
activity, which commonly consists of
40-60 sec on-ice shifts. T h e ice
hockey player is required to develop
muscle strength, power, and anaero-
bic endurance on the ice. This train-
ing can be difficult to simulate in the
off season due to the absence of ice.
Johansson et al (1 1) demon-
Journal of Orthopaedic & Sports Physical Therapy®
prediction of skating speed was possible using two isokinetic measurements, the average power of
the left quadriceps at 180" (APLQ180) and the average power of the right hamstrings at 60'
(APRH60). These measurements may be used to objectively evaluate off-ice training programs
designed to enhance speed skating performance.
Key Words: ice hockey, isokinetics, power measurements
Presented at the Sports Physical Therapy Section Team Concept Meeting, December 1991, New Orleans, LA.
' Self-employed, Clinton, TN
' Statistical consultant, QualPro, Knoxville, TN
' Staff therapist, Knoxville Sports Therapy, Knoxville, TN; doctoral student, exercise science, The University of

strated the need for individualized Tennessee

off-season training regimens; how-
ever, nowhere in the scientific litera-
ture is there a standardized measure-
ment tool with which to evaluate the
effectiveness of such training. This
study was conducted to: 1 ) identify
the best off-ice predictor(s) of sprint
speed skating time in professional
hockey players and 2) establish the
validity of a specific functional meas-
urement which will allow a more ob-
jective evaluation of the efficacy of
the off-season training program.
MATERIALS AND METHODS
Subjects
The sample initially included 16
professional hockey players from the
East Coast Conference who were
tested at the end of the competitive
season. T h e final sample was re-
duced to nine as a result of injuries
and player relocation. All nine sub-
jects were free of any acute lower
extremity injury and were not in-
volved in any other athletic activity.
Their ages were 23.3 1.6 years, +
+
height was 182.0 4.0 cm, weight
+
was 88.8 7.5 kg, and percent body
fat was 10.1 f 2.2 (predicted by
three sites with Lange skinfold cali-
pers using the Jackson and Pollock
equation) (20). T h e subjects con-
sisted of five forwards, three de-
fensemen, and one goaltender.
Materials
A Biodex (Biodex Corp., Shirley,
NY) dynamometer was used for all
isokinetic testing, and a Bodyguard
990 lower extremity cycle was used
for the warm-up phase of isokinetic
testing. T h e 40-yd dash, long jump,
and vertical jump were all tested on
astro turf in an indoor football prac-
tice facility. The speed skating test
was completed on a standard-size
hockey arena.
Methods
Isokinetic tests, field tests, and
ice skating tests were all completed
on separate days. Biodex testing fol-
lowed the protocol used in the Fier-
ing et al reliability study (5). The
Biodex was calibrated one day prior
to testing, as prompted by the com-
puter software. A 5-min cycle warm-

- -
Volume 15 Number 2 February 1992 *JOSPT
 u p used a pedal rpm of 8 8 and a
target reading of 600 kpm. Follow-
ing the bike warm-up, subjects per-
formed a SO-sec static stretch for the
hamstrings and quadriceps. Fach
subject then performed three maxi-
mal repetitions a t a speed of 6 0 ° /
sec. Fifteen maximal repetitions
were performed a t 180°/sec.
T h e field testing consisted of a
vertical jump, a standing long jump.
and a 40-yd dash. Preceding field
testing, the warm-up included a 100-
yd jog and a 30-sec stretch of all
lower extremity muscle groups. T h e
athletes were given three opportuni-
ties at each test, and the best per-
Downloaded from www.jospt.org at on November 8, 2015. For personal use only. No other uses without permission.
formance was recorded. T h e vertical
jump anaerobic power was deter-
mined by the Lewis Nomogram (3).
For the jump and reach scores on
vertical power testing, the athletes
were allowed to use an arm swing.
Copyright © 1992 Journal of Orthopaedic & Sports Physical Therapy®. All rights reserved.
Identical position and arm swing
were used in the standing long jump.
T h e 40-yd dash began in a sprinter's
starting position. Time was meas-
ured by a hand-held stopwatch.
Before attempting the sprint
skate test, all athletes participated in
a 20-min, on-ice warm-up, which in-
cluded lower extremity stretching as
Journal of Orthopaedic & Sports Physical Therapy®
a group and slow skating drills. Fach
subject performed the sprint skate
test in full hockey pads and with a
stick. A hand-held stopwatch was
used t o record the time it took t o get
from goal line t o goal line (54.9 m).
T h r e e skating trials were allowed,
with the best recorded. All athletes
were given a 30-sec rest between
each trial. Speed skating anaerobic
power (SSAP) was determined by the
formula utilized in the Watson and
Sargeant study (26):
subject's weight distance skated
(1)
time
Statistical Methods
Initially, linear regression
models were used to evaluate the
functional relationship between skate
time and nine other independent
JOSPT Volume 15 Number 2 February 1992
variables. T h e basic multiple linear
regression model follows:
whereby the authors seek t o mini-
mize the sum of the residuals
squared,
Min .21 e:
1-
= Min 5
i= 1
(Y, - f,)'
with ei an estimate of the true resid-
ual, ci ( 18). However, recent statisti-
cal literature has pointed out that
just one observation can cause the
least squares estimators t o take on
large aberrant values (23). In the
case of small samples, which fre-
The best predictor of
54.9-m skating time
for fonvards and
defensemen was the
vertical jump.
quently occur, unusual data values
can be particularly damaging to
interpretations and conclusions.
Many statistical packages (such as
SAS, SPSSX) now include diagnos-
tics for flagging singular observa-
tions as atypical in linear regression
models. T h e r e a r e two kinds of a t y p
ical data in least squares regression.
T h e first are called high leverage
points, which are outlying in the ex-
planatory o r independent variables
(9, 22). T h e second typical situation
in least squares regression is an a t y p
ical response of the dependent vari-
able to the independent variables.
Observations in this category are
called outliers (1 8). These statistical
diagnostics are not automatic keys to
immediately eliminate observations
from the analysis but are a means of
highlighting observations that d o not
fit the formulated model, that are
recording o r experimental errors, o r
that reveal another trend in the
data. These kinds of atypical obser-
vations affect the following: coeffi-
cient estimates, the standard errors,
Pearson's correlation, t-tests, the F-
test for the full model, the coeffi-
cient of determination (R2), variable
selection for a best model, interrela-
tionships among variables in multiple
(3)
linear regression, and the predicta-
bility of the model (2, 18, 23).
Recent statistical research has
show" that all of these classical diag-
nostic measures become unreliable in
the case of multiple atypical data (2,
22). When the sample is small and
there are multiple atypical data
points, as in this case, it is difficult to
discard any data.
As a result, a four-fold strategy
will be used. First, the classical diag-
nostic measures were utilized but not
reported. Secondly, since least
squares estimates break down with
just one outlier, a robust estimator
able t o tolerate up t o 50 percent of
the data being contaminated with
outliers was also used. This robust
estimator, the least median of
squares estimator (LMS), minimizes
the median of the observer residuals
squared,
Minimize median e:
b(k) i = . n -. (4)
Basically, this LMS estimator of the
parameters is calculated by repeated
sampling of elemental sets of obser-
vations until the pattern of residuals
converges to stability and the scatter
of the residuals is minimal with re-
spect t o the k-estimated coefficients
(bb) (23). T h e LMS estimator pro-
vides a good estimate of the coeffi-
cients and flags outliers. Because
multiple atypical data points can hide
their impact, a plot of standardized
robust residuals from the LMS a p
proach versus robust distances will
also be used to look for groups of
observations that may have an i n a p
propriate effect as t o outlyingness o r
 RESEARCH S T U D Y

leverage (22). Rousseeuw and van Pearson Correlation Coefficients

Zomeren suggest cutoff points of 2.5 APRQ APRH APLQ APLH APRQ APLQ LEWIS APRH APLH
and -2.5 for the standardized robust 60 60 60 60 180 180 180 180
residuals and a cutoff of [X2p,,q,5]'h SKTIME
for the robust distance. APRQ60
Finally, the clinical researcher APRH60
would like to validate the predicta- APLQ60
APLH60
bility of a selected best model. If the APRQ180
data set is too small to split for varia- APLQl80
tion purposes, there is a ~ ~ , , , , d , LEWIS
which will reflect model prediction APRHl8O
capabilities (1 8). Given n observa- Spearman Correlation Coefficients
SKTIME .72 .23 .65 .45 77 .85
tions in the data set, each observa- APRQ60 -.27 .82 .35 .92 .94
tion is individually withheld o r set APRH60 -.28 .47 -.I8 -.07
aside, and the remaining n- 1 obser- APLQ60 .27 .78 .88
vations are used to estimate the APLH60 .48 .36
omitted observation. This is done n APRQ l8O .88
APLQ180
Downloaded from www.jospt.org at on November 8, 2015. For personal use only. No other uses without permission.

times (nine times here), resulting in LEWIS

n predictions, n prediction errors, APRHl8O
and one R'lprCd.T h e best choice of a
predictive model would favor the TABLE 1. Correlation coefficients.
largest R~,,,~.
Copyright © 1992 Journal of Orthopaedic & Sports Physical Therapy®. All rights reserved.

standardized residuals and robust Two Variables

RESULTS
One Variable
T h e first key that something was
suspicious with the data occurred in
the computation of the Pearson and
Spearman correlations shown in Ta-
ble 1. O f the 4 5 pairwise correla-
Journal of Orthopaedic & Sports Physical Therapy®
distances pinpoints only skater nine
as an outlier (as shown in the upper All possible regressions were
panel of Figure 2). performed on all variable subsets;
but meaningless negative coeffi-
cients, complicated outliers, ques-
lohansson et a1 tionable nonsignificance, and poor
predictability limited focus to two-
demonstrated the variable models. T h e best two-vari-
able model does have an outlier
need for

tions, 16, o r about 36 percent, of the
Spearman correlations were differ-
ent by .20 in magnitude o r by sign.
A look at Table 1 shows these differ-
ences. T h e existence of these differ-
ences encourages the use of LMS
regression in tandem with least
squares regression.
All possible regressions were
performed. Six of the best LSR scat-
terplots are shown in Figure 1. T h e
best one-variable results a r e in Table
2. T h e best one-variable LSR model
used the Lewis score (R2 = .724).
Singular regression diagnostics
would flag observations one and nine
as potential outliers and observation
four as a leverage point (see Figure
1). T h e LMS regression (Table 2)
gives a better fit (R2 = .880), even
with these three observations identi-
fied as suspect. A plot of the LMS
individualized off-
season training
regimens.
With the omission of the one
outlier (skater nine, who was a
goalie), the LSR model is closer t o
the LMS model (Table 2). T h e one-
variable model with the goalie de-
leted yields a RVprrd= .494 for the
LEWIS variable, a better result than
the R2,,,,d = .302 for the second best
one-variable model, the average
power of the left quadriceps at
180°/sec (APLQI 80). Thus, the
one-variable model, ? = 236.30 +
3.65(LEWIS), may be used t o pre-
dict skating speed anaerobic power.
(again the goalie) as evidenced in the
robust plot in the lower panel of Fig-
ure 2. This best interpretive and
predictive model contained
APLQ180 and APRH6O (Table 3).
This model used only eight observa-
tions and excluded the goalie, giving
an R 2 = .928. T h e RZprrd= .769 for
this same model was far better than
any other two-variable model. Thus,
the predictive equation using least
squares regression but confirmed by
the least median squares regression
is as follows: ? = 354.35
+.69(APLQ180) +
1.47(APRH60).
T h e bottom line in the statistical
analysis is that LEWIS is an excellent
one variable to predict skate time o r
speed skating anaerobic power.
However, if two variables are
needed, APLQI 8 0 and APRH6O
are the better choice.

Volume 15 Number 2 Februav 1992 JOSPT


Downloaded from www.jospt.org at on November 8, 2015. For personal use only. No other uses without permission.
Copyright © 1992 Journal of Orthopaedic & Sports Physical Therapy®. All rights reserved.
Journal of Orthopaedic & Sports Physical Therapy®
FIGURE 1. Scatter plots lor different independent variables. The Y axis is always SKTIME, speed skating
anaerobic power according to equation (1).
DISCUSSION
Speed skating is one of the ma-
jor components of performance in
professional hockey. An ice hockey
shift demands short bursts of maxi-
mal effort as the forwards and de-
fensemen skate rapidly from goal t o
goal line. A primary factor in a
hockey player's success is his abiltiy
to develop great amounts of muscu-
lar tension very rapidly, ultimately
generating skating speed. Smith e t al
have suggested that more emphasis
should be placed on developing mus-
cular power at high speeds and have
recommended that improvements in
high speed power output should be a
goal of both on- and off-ice training
JOSPT Volume 15 Number 2 February 1992
programs (24). However, it has been
shown that the training regimen
should be specifically tailored for
each individual player due t o signifi-
Performance time
increases due to
increased mass of
the skates worn during
the test.
cant intraindividual variation in per-
formance (1 I). Off-ice testing meas-
ures that will accurately measure
progress toward personal goals a r e
RESEARCH STUDY
necessary to determine the propriety
of prescribed training regimens, es-
pecially during the ofT season. These
tests should be relatively easy t o ad-
minister yet highly correlated with
skating speed.
T h e muscles responsible for
knee flexion and extension are ex-
tremely important in skating per-
formance and, therefore, were the
ones examined. T h e quadriceps de-
velop the largest contractile forces
when extending the knee joint in the
skating thrust (8). Electromyo-
graphic studies of ice skating have
shown that the vastus medialis and
the vastus lateralis have the most ac-
tivity (1 2, 15). Furthermore, a rapid
depletion of glycogen was reported
to occur in the vastus lateralis during
supramaximal skating (6). T h e ham-
strings function as a knee stabilizer
during the weight shift and push off
of the skating thrust (8).
Isokinetic power tests were in-
cluded because the developriient of
maximal muscular power output
during athletic activity has previ-
ously been best simulated with isoki-
netic tests (7). Numerous studies
have also investigated anaerobic
power production in athletes using
functional tests, and these tests have
been used t o predict specific athletic
performance. For example, the verti-
cal jump has been shown t o be a val-
uable predictor of Olynipic weight
lifting performance (25). T h e pres-
ent authors found that the vertical
jump scores, as calculated by the
Lewis formula, were highly corre-
lated with knee extension power a t
180°/sec (in a nonparametric way
with a Spearman correlation of 22).
These results are similar to those of
Manning, who concluded that knee
extension at l SO0 and the vertical
jump using the Lewis formula could
be used interchangeably as a nieas-
ure of anaerobic power (10, 13, 14).
Perrine also found total leg thrust
power output on an isokinetic device
to have a high correlation with the
vertical jump (a Pearson correlation
 RESEARCH STUDY

slowly than the forwards o r the de-

fensemen, which would be expected
due to the fact that they wore heav-
ier equipment and heavier skates
with a different blade. Added mass
due to an increase in equipment
weight has been shown to result in a
significantly slower performance on
the speed component of a hockey fit-
ness test (1 7). In addition, perform-
ance time increases due t o increased
mass of the skates worn during the
test (4). Furthermore, the goalie is
not trained in speed performance as
are the other position players.
Therefore, the goalie's exclusion,
which was deemed to be valid statis-
tically, is physiologically valid as well.
Downloaded from www.jospt.org at on November 8, 2015. For personal use only. No other uses without permission.

T h e best predictor of 54.9-m

skating time for forwards and de-
fensemen was the vertical jump
score. Skating speed can be pre-
dicted by the following equation: p =
236.30 + 3.65(LEWIS). Since the
Copyright © 1992 Journal of Orthopaedic & Sports Physical Therapy®. All rights reserved.

dependent variable is really speed

skating anaerobic power (SSAP), a
10-unit increase in the Lewis score
will yield an increase of 36.5 kg-m/
sec in SSAP. A 180-lb athlete who
receives a Lewis score of !30 and
who skates 54.9 m would have a pre-
dicted SSAP of 7 10.8 kg-m/sec.
T h a t translates into a skating speed
Journal of Orthopaedic & Sports Physical Therapy®

FIGURE 2. Robust residuals versus robust distances. ROBRES are the robust residuals from the least median of 6.32 sec. T h e corresponding 9 5
squares analysis, and ROBMD are the robust distances for the independent variables. The upper panel is for percent confidence interval for this
the best one-variable model, and the lower panel is for the best two-variable model. individual observation would be
from 630.3-79 1.3 in SSAP, which
Model N Intercept Lewis Coeff. R2 of .88) (19). Both of these measures translates into 5.68-7.12 sec in skat-
LSR 9 27.52 4.95" .724 also proved t o be closely related t o ing speed. T h e vertical jump test is
LMS 9 162.80 4.16 .880 skating speed. easy to administer and requires little
LSR 8 236.30' 3.65" .810 T h e subjects in this study were equipment. It can easily be utilized
* Reflects statrctical significance at a .05 level. similar in height, weight, and body by athletes, coaches, and medical
" Reflects statistical significance at a .01 level. fat percentage t o previously de- professionals t o objectively evaluate
scribed amateur and Olympic ice the success of off-season training, es-
TABLE 2.One-variable model possibilities. hockey players (1, 16, 24). T h e pecially if ice is unavailable. It can
mean 54.9-m skating time for the also serve as a quick and economical
Model N Intercept APLQ180 APRHCO 'R forwards and defensemen was a p tool with which to estimate skating
proximately .5-1.0 sec faster than speed during preseason and postsea-
LSR 9 250.79 1.03' 1.25 .645 times reported by Reid, Watson and son physical screenings and can assist
LMS 9 296.29 .45 2.38 .650
LSR 8 354.35" .69" 1.47" .928 Sargeant, and Smith e t al (1 6, 2 1, coaches and players in setting sea-
24). This difference is probably due sonal goals.
' Reflects statislical significance at a .05 level. t o a difference in skill level as none An even better prediction of
" Reflects statistical significance at a .01 level.
of the other studies examined skating speed is possible with two iso-
TABLE 3. Two-variable modelpossibilities: partial professional hockey players. In all kinetic measurements, the average
regression coefficients. cases, the goaltenders skated more power of the left quadriceps a t 180"

Volume 15 Number 2 February 1992 JOSPT


a n d the average power o f the right
hamstrings at 60". T h e prediction
equation is: ? = 354.35
+.69(APLQl80) + 1.47(APRH60).
A n increase o f 1 0 watts in quadri-
ceps power h o l d i n g hamstring power
constant increases SSAP 6.9 kg-m/
sec, while a n increase o f 10 watts i n
hamstring power h o l d i n g quadriceps
power constant increases SSAP
1 1.47 kg-ni/sec. T h e SSAP predic-
tion f o r the same 180-lb athlete used
earlier w h o had a n isokinetic reading
o f 326.6 watts f o r APLQ180 a n d o f
106.8 watts f o r APRHGO yields a n
SSAP o f 735.9 kg-m/sec, a skating
speed o f 6.1 sec. Again, t h e corre-
Downloaded from www.jospt.org at on November 8, 2015. For personal use only. No other uses without permission.
sponding 9 5 percent confidence in-
terval f o r this individual observation
would range f r o m 679.7-792.1 kg-
m/sec f o r SSAP o r f r o m 5.67-6.6 1
sec. W h i l e i t is possible t o predict
skating time w i t h either t h e one- o r
Copyright © 1992 Journal of Orthopaedic & Sports Physical Therapy®. All rights reserved.
two-variable models, i t is obvious
f r o m this example that the individual
confidence interval is tighter w i t h
the t w o isokinetic variables. F u r t h e r
studies are clearly necessary t o vali-
date the proposed models.
T h e emergence o f specific limbs
and speed as the variables o f choice
in these equations may b e related t o
Journal of Orthopaedic & Sports Physical Therapy®
skating mechanics. First, the quadri-
ceps must develop torque very
quickly d u r i n g push-off t o obtain
maximal velocity while the knee flex-
ors function m o r e as stabilizers,
which may b e accomplished using a
slower contractile speed. Addition-
ally, most hockey players habitually
cross over w i t h their left leg while
skating, which would emphasize left
knee extension a n d right hamstring
stabilization. Future studies t o fur-
ther investigate these differences i n
large sample sizes would b e war-
ranted. I t should also be noted that
the isokinetic tests were done o n a
Riodex dynamometer a n d that cau-
tion must b e used when generalizing
the results t o values obtained o n
other isokinetic devices.
JOSPT * Volume 15 * Number 2 * February 1992
CONCLUSIONS
In the g r o u p o f professional
hockey players tested, the height
achieved o n a vertical j u m p test was
the single best predictor o f skating
speed. I f isokinetic measures are
available, average isokinetic power o f
the left quadriceps at 18 0 ° , coupled
w i t h the average power o f the right
hamstrings at 60°, can b e used t o
predict skating speed even m o r e ef-
fectively.
A second critical f i n d i n g o f this
study was the need f o r close interac-
t i o n between clinicians a n d statisti-
cians when engaging in sports medi-
cine research. Traditional statistical
methods applied t o small sample
sizes may yield inappropriate results,
a n d m o r e sophisticated techniques
may b e necessary t o validate per-
formance relationships. Additional
studies using other samples o r larger
sample sizes (if obtainable) would he
necessary t o fully validate the models
proposed i n this investigation.
JOSPT
ACKNOWLEDGMENTS
Special thanks t o the Depart-
ment o f Management I n f o r m a t i o n
Resources at U'estern Illinois Uni-
versity a n d the Department o f Statis-
tics at T h e University o f Tennessee
at Knoxville f o r p r o v i d i n g the com-
p u t e r time f o r this research.
REFERENCES
Agre IC, Casal DC, Leon AS, McNally
MC, Baxter TL, Serfass RC: Professional
ice hockey players: Physiologic, an-
thropometric and musculoskeletal
characteristics. Arch Phys Med Rehabil
69: 188- 192, 1988
Antille C, Ritschard C: Robust and
classical outlyingness indicators, a sim-
ulation study. Commun Statist - Simula
19:505-5 12, 7 990
Banister EW, Makjavic IS: Experiments
in Human Performance, p 138. Bur-
naby, BC, Canada: Holt, Rinehart and
Winston of Canada, Limited, 1986
Chamay I,Montgomery DL, Hoshizaki
TB, Brayne SP: The effect of added
weight on performance in an ice
RESEARCH S T U DY
.-- -. -
hockey fitness test (Abstr). Can 1Appl
Sports Sci 7:240, 7 982
5. Feiring DC, Ellenbecker TS, Derscheid
CL: Test retest reliability of the biodex
isokinetic dynamometer. I Orthop
Sports Phys Ther 7 1:298-300, 7990
6. Green HI: Glycogen depletion pat-
terns during continuous and intermit-
tent ice skating. Med Sci Sports
70:183-187, 1978
7. Halling AH, Dooley IN: The impor-
tance of isokinetic power and its spec-
ificity to athletic conditions. Athl Train
14:83-86, 7 979
8. Halliwell AA: Determination of mus-
cle, ligament and articular forces at the
knee during a simulated skating thrust.
Unpublished master's thesis, Univer-
sity of British Columbia, 1978
9. Hoaglin DC, Welsch RE: The hat matrix
in regression and ANOVA. Am Statist
32: 7 7-22, 1978
70. Houston ME, Creen HI: Physiological
and anthropometric characteristics of
elite Canadian ice hockey players. /
Sports Med 16: 123- 7 28, 7 976
7 I.lohansson C, Lorentzon R, Fugl-Meyer
AR: lsokinetic muscular performance
of the quadriceps in elite ice hockey
players. Am ISports Med 7 7:30-34,
7 989
72. Kumamoto M, Ito M, Yamashita N,
Nakagawa H: Electromyographic
study of the (SIC) ice skating, p p 730-
134. lnternat Cong Winter Sports Med,
Sapporo, lapan, 1972
13. Manning /M, Dooly-Manning C: An-
aerobic power tests which can be used
interchangeably. Med Sci Sports Exerc
18:S24,SZS, 1986
74. Manning /M, Dooly-Manning C, Perrin
DH: Factor analysis of various anaero-
bic power tests. ISports Med Phys Fit
28: 138- 144, 1988
15. Mashima H, Aoki I,Maeshima T, Shi-
mizu T, Sato T: Telemetric electro-
myography in the training of speed
skaters. DD 724-129. Int Cone Winter
Sports Mid, Sapporo, lapan, 1972
16. Montgomery DL: Physiology of ice
hockey. Sports Med 5:99- 7 26, 1988
17. Montgomery DL: The effect of added
weight on ice hockey performance.
Phys Sports Med 10:9 1-99, 1982
18. Meyers RH: Classical and Modern
Regression with Applications, p p 123-
274. Boston: PWS-Kent Publishing
Company, 1990
19. Perrine 11, Cregor R, Munroe R, Edger-
ton VR: Muscle power capacities and
temporal output patterns of skilled vs
non-skilled vertical jumpers (Abstr).
Med Sci Sports, 10:64, 1978
20. Pollock ML, Wilmore IH, Fox SM: Ex-
 RESEARCH STUDY
ercise in Health and Disease, pp 2 19-
227. Philadelphia: DB Saunders, 1984
21. Reid A, Hanson H, Cotton C, Cauthier
R, lette M, Thoden 1, Wenger H: De-
velopment and validation of an on ice
hockey fitness test (Abstr), Can I Appl
Sports Sci 4:245, 1979
22. Rousseeuw PI, vanzomeren BC: Un-
masking multivariate outliers and le-
verage points (with comments and re-
Downloaded from www.jospt.org at on November 8, 2015. For personal use only. No other uses without permission.
Copyright © 1992 Journal of Orthopaedic & Sports Physical Therapy®. All rights reserved.
Journal of Orthopaedic & Sports Physical Therapy®
joinder). I Am Statist Assoc 85:633-
651, 1990
23. Rousseeuw PI, Leroy AM: Robust
Regression and Outlier Detection.
New York: john Wiley & Sons (Inter-
science Publication), 1987
24. Smith Dl, Wenger HA, Quinney HA,
Sexsmith /R, Steadward RD: Physiol-
ogy profiles of the Canadian Olympic
hockey team (1980). Can I Appl Sports
Sci 7: 142- 146, 1982
25. Stone MH, Byrd R, Tew I, Wood M:
Relationship between anaerobic
power and Olympic weightlifting per-
formance. I Sports Med Phys Fit 20:99-
102, 1980
26. Watson RC, Sargeant TLC: Laboratory
and on-ice test comparisons of anaer-
obic power of ice hockey players. Can
I Appl Sports Sci 1 1:2 18-224, 1986
Volume 15 Number 2 February 1992 JOSPT