Upload
illa-adeng
View
237
Download
0
Embed Size (px)
Citation preview
7/31/2019 Notational Analysisa Math Perspective
1/91
ModellingModelling FeedbackFeedback
AnalysisAnalysis
ImprovementImprovement
ObservationObservation
PerformancePerformance
Notational analysisNotational analysis a mathematicala mathematical
perspectiveperspective
Mike Hughes, Benn Blackburn and Nic James
7/31/2019 Notational Analysisa Math Perspective
2/91
7th Australasian Conference on Mathematics7th Australasian Conference on Mathematicsand Computers in Sportand Computers in Sport
Linear relationships in data gathering and feedback
Research
Data
Performance
Analyst
Coach/Athlete
7/31/2019 Notational Analysisa Math Perspective
3/91
7th Australasian Conference on Mathematics7th Australasian Conference on Mathematicsand Computers in Sportand Computers in Sport
The role of the performance analyst using early analogue video
and computer systems
Performance Analyst
DATA
Coach/Athlete
DATA DATA
Motor ControlNotational AnalystBiomechanist
Motor ControlNotational AnalystBiomechanist
Gathering
systems
Gathering
systems
Gathering
systems
Processing
systems
Processing
systems
Processing
systems
7/31/2019 Notational Analysisa Math Perspective
4/91
7th Australasian Conference on Mathematics7th Australasian Conference on Mathematicsand Computers in Sportand Computers in Sport
A digital systems approach to the data sharing that the interactive commercial
systems have enabled for performance analysts working with coaches and athletes
(apologies to Popper).
Coach
Athletes
Motor
Control
Notational
Analyst
Coach
Athletes
Biomechanist
Performance
Data
Coach
Athletes
Performance Analysis Team
7/31/2019 Notational Analysisa Math Perspective
5/91
7th Australasian Conference on Mathematics7th Australasian Conference on Mathematicsand Computers in Sportand Computers in Sport
ANALYST
The answer to the mystery of the universe?
7/31/2019 Notational Analysisa Math Perspective
6/91
7th Australasian Conference on Mathematics7th Australasian Conference on Mathematicsand Computers in Sportand Computers in Sport
ANALYST
PERFORMANCE
INDICATORS
7/31/2019 Notational Analysisa Math Perspective
7/91
Performance IndicatorsPerformance Indicators
7/31/2019 Notational Analysisa Math Perspective
8/91
7th Australasian Conference on Mathematics7th Australasian Conference on Mathematicsand Computers in Sportand Computers in Sport
IntroductionIntroduction
Performance Indicators arePerformance Indicators are
a selection or combination ofa selection or combination of
action variable(s) that aim toaction variable(s) that aim todefine some aspect, or all, ofdefine some aspect, or all, of
a performance.a performance.
hat are:-
ERFORMANCE INDICATORS?
Hughes and Bartlett, 2002)
7/31/2019 Notational Analysisa Math Perspective
9/91
7th Australasian Conference on Mathematics7th Australasian Conference on Mathematicsand Computers in Sportand Computers in Sport
IntroductionIntroduction
Performance Indicators are usedPerformance Indicators are used
to assess performance eitherto assess performance either
comparatively, with previouscomparatively, with previous
performances, or absolutely.performances, or absolutely.
Why use:Why use:--
PERFORMANCE INDICATORS?PERFORMANCE INDICATORS?
7/31/2019 Notational Analysisa Math Perspective
10/91
Notational AnalysisNotational Analysis -- Performance IndicatorsPerformance Indicators
CRICKETCRICKET : Strike rate, Dismissal rate, Fielding Efficiency: Strike rate, Dismissal rate, Fielding Efficiency
Examples:Examples:--
SOCCERSOCCER : Shots, Passes, Passing Accuracy: Shots, Passes, Passing Accuracy
RUGBYRUGBY : Turnovers, Tackles, Passes/Possession: Turnovers, Tackles, Passes/Possession
BADMINTONBADMINTON : W/E ratio, shots/rally, Quality serve/return: W/E ratio, shots/rally, Quality serve/return
7/31/2019 Notational Analysisa Math Perspective
11/91
7th Australasian Conference on Mathematics7th Australasian Conference on Mathematicsand Computers in Sportand Computers in Sport
Performance Indicators?Performance Indicators?
How do we choose the performance indicators?
7/31/2019 Notational Analysisa Math Perspective
12/91
Definition of Success:Definition of Success:--
It could be defined byIt could be defined by
winning (scoring more goalswinning (scoring more goals
than the opposition) but itthan the opposition) but it
may not.may not.
Notational AnalysisNotational Analysis -- Performance IndicatorsPerformance Indicators
Or a coach may be lookingOr a coach may be looking
for a qualitative improvementfor a qualitative improvement
in performancein performance -- whichwhich
could be assessed by acould be assessed by a
performance indicator.performance indicator.
7/31/2019 Notational Analysisa Math Perspective
13/91
Definition of Success:Definition of Success:--
Success then is relativeSuccess then is relative --
either to your opposition,either to your opposition,
racket sportsracket sports --
more winners; less errors,more winners; less errors,
invasive gamesinvasive games --
more points or goals than themore points or goals than theoppositionopposition
Notational AnalysisNotational Analysis -- Performance IndicatorsPerformance Indicators
or to previousor to previous
performancesperformances
of your own team, orof your own team, or
individual player,individual player,
or to aggregated meansor to aggregated means
of peer performancesof peer performances
7/31/2019 Notational Analysisa Math Perspective
14/91
Notational AnalysisNotational Analysis -- Performance IndicatorsPerformance Indicators
POSITIVE or NEGATIVEPOSITIVE or NEGATIVE
Types of Performance Indicators:Types of Performance Indicators:--
SCORING :SCORING : Goals etc., W, E, W/E, Goals/Shots, DismissalGoals etc., W, E, W/E, Goals/Shots, Dismissal
rate, etc.rate, etc.
QUALITY :QUALITY : Turnovers, Tackles, Passes/Possession,Turnovers, Tackles, Passes/Possession,shots/rally, Strike rate, etc.shots/rally, Strike rate, etc.
7/31/2019 Notational Analysisa Math Perspective
15/91
Dangers of Performance Indicators:Dangers of Performance Indicators:--
In other areas of science,In other areas of science,
performance indicatorsperformance indicators
tend to be ratios oftend to be ratios of
variables, orvariables, or
combinations ofcombinations of
variables, that thenvariables, that then
render the final P.I.render the final P.I.dimensionless:dimensionless:--
Notational AnalysisNotational Analysis -- Performance IndicatorsPerformance Indicators
=density=density
Reynolds No. =Reynolds No. = UdUd U=velocityU=velocity
=viscosity=viscosity
d=size of objectd=size of object
E.g.E.g.
In aerodynamics,In aerodynamics,
Mach No. =Mach No. = Velocity of aircraftVelocity of aircraft
Velocity of soundVelocity of sound
In fluid dynamics,In fluid dynamics,
7/31/2019 Notational Analysisa Math Perspective
16/91
7th Australasian Conference on Mathematics7th Australasian Conference on Mathematicsand Computers in Sportand Computers in Sport
MMaattcchh
CCllaassssiiffiiccaattiioonn
TTeecchhnniiccaall TTaaccttiiccaall
Performance IndicatorsPerformance Indicators -- GENERICGENERIC
Data for bothData for both
teamsteams (T(TVV))NN/ (T/ (TVV)TOTAL)TOTAL
Means of peerMeans of peer
performancesperformances
oror
(T(TVV))NN/ (/ (POSSPOSSnn)) TOTALTOTAL
7/31/2019 Notational Analysisa Math Perspective
17/91
Dangers of Performance Indicators:Dangers of Performance Indicators:--
Performance Indicators should bePerformance Indicators should be normalisednormalised oror
standardisedstandardised in some way to the respectivein some way to the respective
performance variables, and should also be usedperformance variables, and should also be used
comparatively with either your opponentscomparatively with either your opponents data,data,
previous data of your own performances, orprevious data of your own performances, or
with aggregated data of performances of yourwith aggregated data of performances of your
own level.own level.
Notational AnalysisNotational Analysis -- Performance IndicatorsPerformance Indicators
7/31/2019 Notational Analysisa Math Perspective
18/91
To differentiate between important dataTo differentiate between important data
and DIRTY WASHING:and DIRTY WASHING:--
*Choose parameters*Choose parameters
that relate strongly tothat relate strongly to
outcome or quality ofoutcome or quality of
performance.performance.
Notational AnalysisNotational Analysis -- Performance IndicatorsPerformance Indicators
*What are the units of your*What are the units of your
Performance Indicator?Performance Indicator?
*Can these be related to*Can these be related to
other variables or previouslyother variables or previously
calculated means?calculated means?
*Are there ways of*Are there ways of
combining variables incombining variables in
a group that will saya group that will say
more about thismore about this
performance?performance?
7/31/2019 Notational Analysisa Math Perspective
19/91
Notational AnalysisNotational Analysis -- Performance IndicatorsPerformance Indicators
Dangers of Performance Indicators:-
A single action variable, taken in isolation, can givedistorted impression of a performance because of other
variables, more or less important.
E.g. TEAM A: 12 Turnovers; TEAM B: 8 Turnovers
TEAM B playing better than TEAM A?
7/31/2019 Notational Analysisa Math Perspective
20/91
TEAM B playing better than TEAM A?
Dangers of Performance Indicators:-
E.g. TEAM A: 12 Turnovers; TEAM B: 8 Turnovers
This will depend upon the possession of both the teams
- if TEAM A have had twice as many possessions (48) as
TEAM B (24) then their relative performance w.r.t.TURNOVERS/POSSESSION (T/P)
will be better than that of TEAM B
(T/P)A = 1/4; (T/P)B = 1/3
Notational Analysis - Performance Indicators
7/31/2019 Notational Analysisa Math Perspective
21/91
Dangers of Performance Indicators:-
Let us consider a more complex example
One of the most quoted
research studies in
notation is that of
Reep and Benjamin (1968).
Notational Analysis - Performance Indicators
Most frightening is the
effect that this study
has had on British soccer
and its coaching.
7/31/2019 Notational Analysisa Math Perspective
22/91
Dangers of Performance Indicators:-
They found:-
80% of goals resulted from
a sequence of three passes or
less,
60% of all goals came from
possession gained in the final
attacking third of the pitch,and
a goal is scored every 10
shots (approximately).
Notational Analysis - Performance Indicators
Hughes, C. (1985,1990)
reinforced these ideas and,
as he was the Director of
coaching for the English
Football Association (F.A.),
his influence in the game
was considerable. So much
so, that these tenets were
included in the coaching
literature produced by the F.A.
7/31/2019 Notational Analysisa Math Perspective
23/91
Dangers of Performance Indicators:-
Patterns of goal scoring with respect to the different
lengths of possessions in the 1990 and 1994 world cups
for soccer.
Notational Analysis - Performance Indicators
0
5
10
15
20
25
30
35
Goals
0 1 2 3 4 5 6 7 8 >
Touches/Possession
1990
1994
7/31/2019 Notational Analysisa Math Perspective
24/91
Dangers of Performance Indicators:-
Frequency of each possession string in the two
tournaments.
Notational Analysis - Performance Indicators
0100020003000
400050006000700080009000
Frequency
0 1 2 3 4 5 6 7 8 >
Touches/Possession
1990
1994
7/31/2019 Notational Analysisa Math Perspective
25/91
Dangers of Performance Indicators:-
Analysis of the number of goals scored
per 1000 possessions for the 2 world cups.
Notational Analysis - Performance Indicators
0
2
4
6
8
10
1214
0 1 2 3 4 5 6 7 8 >
Touches/Possession
(G/P)*1000
1990
1994
Mean
7/31/2019 Notational Analysisa Math Perspective
26/91
Dangers of Performance Indicators:-
Frequency of shots per 1000 possessions
for the 1990 and 1994 World Cups.
Notational Analysis - Performance Indicators
0
20
40
60
80
100
120140
0 1 2 3 4 5 6 7 8 >
Touches/Possession
(S/P)*1000
1990
1994
Mean
7/31/2019 Notational Analysisa Math Perspective
27/91
Dangers of Performance Indicators:-
They found:-
80% of goals resulted from
a sequence of three passes or
less,
60% of all goals came from
possession gained in the final
attacking third of the pitch,and
a goal is scored every 10
shots (approximately).
Notational Analysis - Performance Indicators
NOT
SO
SIMPLE
7/31/2019 Notational Analysisa Math Perspective
28/91
7th Australasian Conference on Mathematics7th Australasian Conference on Mathematics
and Computers in Sportand Computers in Sport
ANALYST
PERFORMANCE
INDICATORS
7/31/2019 Notational Analysisa Math Perspective
29/91
7th Australasian Conference on Mathematics7th Australasian Conference on Mathematics
and Computers in Sportand Computers in Sport
ANALYST
PERFORMANCE
INDICATORS
WHICH ARE
MOST
IMPORTANT?
7/31/2019 Notational Analysisa Math Perspective
30/91
Multivariate StatisticsMultivariate Statistics
Multiple Linear RegressionMultiple Linear Regression
Discriminant Function AnalysisDiscriminant Function Analysis
Binary Logistic RegressionBinary Logistic Regression
7/31/2019 Notational Analysisa Math Perspective
31/91
7th Australasian Conference on Mathematics7th Australasian Conference on Mathematics
and Computers in Sportand Computers in Sport
Multiple Linear RegressionMultiple Linear Regression
Linear Regression produces an equation inLinear Regression produces an equation inthe formthe form Y = a + b.XY = a + b.X
Multiple Linear Regression produces anMultiple Linear Regression produces anequation in the formequation in the form
Y = bY = b00 + b+ b11.X.X11 + b+ b22.X.X22 + b+ b33.X.X33 ++ ++ bbnn.X.Xnn By using databases from tournaments,By using databases from tournaments,
European Championships, WorldEuropean Championships, WorldChampionships, etc., we can assess theChampionships, etc., we can assess therelative importance of the PIrelative importance of the PIs selected. Thats selected. Thatis we are predicting Y (a known outcome)is we are predicting Y (a known outcome)using several PIusing several PIss -- XX11 toto XXnn
7/31/2019 Notational Analysisa Math Perspective
32/91
7th Australasian Conference on Mathematics7th Australasian Conference on Mathematics
and Computers in Sportand Computers in Sport
Assumptions of MLRAssumptions of MLR
((NtoumanisNtoumanis, 2001), 2001) Ratio of cases to independent (X) variables should be atRatio of cases to independent (X) variables should be at
least 5 : 1 and ideally 20 : 1least 5 : 1 and ideally 20 : 1
All outliers should be excluded or transformedAll outliers should be excluded or transformed Save standardised residuals and explore theseSave standardised residuals and explore these
ResidualsResiduals
Should be normally distributedShould be normally distributed There should be no relationship between anyThere should be no relationship between any
independent (X) variables and residualsindependent (X) variables and residuals
There should be no relationship between residuals andThere should be no relationship between residuals andpredicted valuespredicted values which can be savedwhich can be saved
Any relationship between the residuals and dependentAny relationship between the residuals and dependent(Y) variable should be linear(Y) variable should be linear
The residuals should be independent,The residuals should be independent, ieie no relationshipno relationshipbetween order of observation (timebetween order of observation (time--order) and residualorder) and residual
7/31/2019 Notational Analysisa Math Perspective
33/91
7th Australasian Conference on Mathematics7th Australasian Conference on Mathematics
and Computers in Sportand Computers in Sport
ANALYST
PERFORMANCE
INDICATORS
WHICH ARE
MOST
IMPORTANT?
7/31/2019 Notational Analysisa Math Perspective
34/91
7th Australasian Conference on Mathematics7th Australasian Conference on Mathematics
and Computers in Sportand Computers in Sport
ANALYST
PERFORMANCE
INDICATORS
PERFORMANCERELIABILITYWHICH ARE
MOST
IMPORTANT?
7/31/2019 Notational Analysisa Math Perspective
35/91
7th Australasian Conference on Mathematics7th Australasian Conference on Mathematics
and Computers in Sportand Computers in Sport
ReliabilityReliability
A guide to some ideas about
some of the issues and problems
associated with reliability.
Analysis procedures for non-parametric data from performance analysis
MIKE HUGHES, STEVE-MARK COOPER AND ALAN NEVILL
(2001, eIJPAS, 2)
7/31/2019 Notational Analysisa Math Perspective
36/91
7th Australasian Conference on Mathematics7th Australasian Conference on Mathematics
and Computers in Sportand Computers in Sport
7/31/2019 Notational Analysisa Math Perspective
37/91
7th Australasian Conference on Mathematics7th Australasian Conference on Mathematics
and Computers in Sportand Computers in Sport
ANALYST
PERFORMANCE
INDICATORS
PERFORMANCERELIABILITYWHICH ARE
MOST
IMPORTANT?
7/31/2019 Notational Analysisa Math Perspective
38/91
7th Australasian Conference on Mathematics7th Australasian Conference on Mathematics
and Computers in Sportand Computers in Sport
ANALYST
PERFORMANCE
INDICATORS
PERFORMANCERELIABILITYWHICH ARE
MOST
IMPORTANT?
HOW MUCH
DATA?
PERFORMANCE
PROFILE
7/31/2019 Notational Analysisa Math Perspective
39/91
7th Australasian Conference on Mathematics7th Australasian Conference on Mathematics
and Computers in Sportand Computers in Sport
ANALYST
PERFORMANCE
INDICATORS
PERFORMANCERELIABILITY
WHICH ARE
MOST
IMPORTANT?
HOW MUCH
DATA?
EMPIRICALMETHODS
PERFORMANCE
PROFILE
7/31/2019 Notational Analysisa Math Perspective
40/91
7th Australasian Conference on Mathematics7th Australasian Conference on Mathematics
and Computers in Sportand Computers in Sport
Normative Profiles
Some examples of sample sizes for profiling in sport.
Research
Reep & Benjamin (1969)
Eniseler et al., (2000)
Larsen et al., (2000)
Hughes et al., (1988)
Tyryaky et al., (2000)
Hughes (1986)
Hughes & Knight (1993)
Hughes & Williams (1987)
Smyth et al., (2001)
Blomqvist et al., (1998)
O'Donoghue (2001)
Hughes & Clarke (1995)
O'Donoghue & Ingram (2001)
SportSoccer
Soccer
Soccer
Soccer
SoccerSquash
Squash
Rugby Union
Rugby Union
Badminton
Badminton
Tennis
Tennis
N(matches for profile)
3,216
4
4
8 (16 teams)
4 and 3 (2 groups)12, 9 & 6 3 groups
400 rallies
5
5 and 5
5
16, 17, 17, 16, 15
400 rallies
1328
7/31/2019 Notational Analysisa Math Perspective
41/91
7th Australasian Conference on Mathematics7th Australasian Conference on Mathematics
and Computers in Sportand Computers in Sport
Performance ProfilesPerformance profiling of an elite male badminton player (Hughes, Evans and Wells, 2001)Search for a normative profile
0
2
4
6
8
10
12
14
16
1 2 3 4 5 6 7 8 9 10 11 12 13 14
Number of a data set
Cumulative
meanofdatavalue
Series1
7/31/2019 Notational Analysisa Math Perspective
42/91
7th Australasian Conference on Mathematics7th Australasian Conference on Mathematics
and Computers in Sportand Computers in Sport
Performance ProfilesPerformance profiling of an elite male badminton player
The cumulative means of each variable were examined over a series of
matches/games.At the first point, the number of matches, N(E), where the cumulative mean
consistently lay within set limits of error was recorded as the establishment of a
normative template of play. These limits of error are a percentage deviation (+/- 1%;
+/- 5%; +/- 10%) of the overall data mean about the overall mean.
Let n = the variable number of matchesg = the variable number of games
N(E) = value of n to reach limits of error
N(T) = total number of matches
Cumulative mean = (Sum of the frequencies of n) / n
Limits of error (10%) = Mean N(T) (Mean N(T) x 0.1)
Limits of error (5%) = Mean N(T) (Mean N(T) x 0.05)
Limits of error (1%) = Mean N(T) (Mean N(T) x 0.01)
7/31/2019 Notational Analysisa Math Perspective
43/91
7th Australasian Conference on Mathematics7th Australasian Conference on Mathematics
and Computers in Sportand Computers in Sport
Performance ProfilesStudy 2:- Performance profiling of an elite male badminton player
M e a n n u m b e r o f s h o t s p e r r a l l y b y m a t c h
4 . 0
5 . 0
6 . 0
7 . 0
8 . 0
9 . 0
1 0 . 0
1 1 . 0
1 2 3 4 5 6 7 8 9 1 1 1
N u m b e r o f m a t c h e s
C u m u l a t i v e m e a n P l u s 1 0 %
L e s s 1 0 % P l u s 5 %
l e s s 5 %
7/31/2019 Notational Analysisa Math Perspective
44/91
7th Australasian Conference on Mathematics7th Australasian Conference on Mathematics
and Computers in Sportand Computers in Sport
Performance ProfilesStudy 2:- Performance profiling of an elite male badminton playerF i g u r e 4 . 5 M e a n n u mb e r o f s h o t s b y g a me
300.0
350.0
400.0
450.0
500.0
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24
Number of games
Cumulative mean '+10% '-10% '+5% '-5% '+1% '-1%
7/31/2019 Notational Analysisa Math Perspective
45/91
7th Australasian Conference on Mathematics7th Australasian Conference on Mathematics
and Computers in Sportand Computers in Sport
European
0.00
25.00
50.00
75.00
100.00
125.00150.00
175.00
200.00
225.00
250.00
275.00
300.00
325.00
350.00375.00
400.00
425.00
450.00
475.00
500.00
525.00
0 1 2 3 4 5 6 7 8 9 10 11 12 13
Number of matches
Meanscore
Pass
Runs
Dribbles
Crosses
Header
Shots
Number of matches needed to achieve a normative profile for attacking variables for
European teams
7/31/2019 Notational Analysisa Math Perspective
46/91
7th Australasian Conference on Mathematics7th Australasian Conference on Mathematics
and Computers in Sportand Computers in Sport
ANALYST
PERFORMANCE
INDICATORS
PERFORMANCERELIABILITY WHICH ARE
MOSTIMPORTANT?
HOW MUCH
DATA?
EMPIRICALMETHODS
PERFORMANCE
PROFILE
7/31/2019 Notational Analysisa Math Perspective
47/91
7th Australasian Conference on Mathematics7th Australasian Conference on Mathematics
and Computers in Sportand Computers in Sport
ANALYST
PERFORMANCE
INDICATORS
PERFORMANCERELIABILITY WHICH ARE
MOSTIMPORTANT?
HOW MUCH
DATA?
EMPIRICALMETHODS
PREDICTION
FROM
VARIANCE
PERFORMANCE
PROFILE
7/31/2019 Notational Analysisa Math Perspective
48/91
7th Australasian Conference on Mathematics7th Australasian Conference on Mathematics
and Computers in Sportand Computers in Sport
James et al. (2004, JSS, in press)James et al. (2004, JSS, in press)
James et al. (2004) suggested an alternative approach whereby thJames et al. (2004) suggested an alternative approach whereby thee
specific estimates of population means are calculated from thespecific estimates of population means are calculated from thesample data through confidence limits (CLsample data through confidence limits (CLs).s).
CLCLs represent upper and lower values between which thes represent upper and lower values between which thetrue (population) mean is likely to fall based on the observedtrue (population) mean is likely to fall based on the observedvalues collected.values collected.
Calculated CLCalculated CLs naturally change as more data is collected,s naturally change as more data is collected,typically resulting in the confidence interval (CItypically resulting in the confidence interval (CI -- upper CLupper CL
minus lower CL) decreasing.minus lower CL) decreasing.
Confidence intervals (CIConfidence intervals (CIs) were therefore suggested to bes) were therefore suggested to bemore appropriate as performance guides compared to usingmore appropriate as performance guides compared to using
mean values.mean values. Using a fixed value appears to be too constrained due toUsing a fixed value appears to be too constrained due to
potential confounding variables that typically affectpotential confounding variables that typically affectperformance, making prescriptive targets untenable.performance, making prescriptive targets untenable.
7/31/2019 Notational Analysisa Math Perspective
49/91
7th Australasian Conference on Mathematics7th Australasian Conference on Mathematics
and Computers in Sportand Computers in Sport
James et al. (2004, JSS, in press)James et al. (2004, JSS, in press)
From a theoretical perspective, James et al. argued that theFrom a theoretical perspective, James et al. argued that theuse of CIuse of CIs can also add significance to the judgement of thes can also add significance to the judgement of thepredictive potential of a data set, i.e. whether enough data haspredictive potential of a data set, i.e. whether enough data has
been collected to allow a reasonable estimation.been collected to allow a reasonable estimation.
For their investigation a criterion was formulated to test theFor their investigation a criterion was formulated to test therate of change of the CI for stability.rate of change of the CI for stability.
Initially 95% CIInitially 95% CIs were calculated for each performances were calculated for each performanceindicator as soon as enough match data had been collected (indicator as soon as enough match data had been collected (NN= 2) and each time more data was added the new CI was= 2) and each time more data was added the new CI wascalculated.calculated.
This meant that CIThis meant that CIs could be constructed for eachs could be constructed for eachperformance indicator after 2, 3 andperformance indicator after 2, 3 and..NNmatchesmatchesrespectively.respectively.
7/31/2019 Notational Analysisa Math Perspective
50/91
7th Australasian Conference on Mathematics7th Australasian Conference on Mathematics
and Computers in Sportand Computers in Sport
ANALYST
PERFORMANCE
INDICATORS
PERFORMANCERELIABILITY WHICH ARE
MOSTIMPORTANT?
HOW MUCH
DATA?
EMPIRICALMETHODS
PREDICTION
FROM
VARIANCE
PERFORMANCE
PROFILE
7/31/2019 Notational Analysisa Math Perspective
51/91
7th Australasian Conference on Mathematics7th Australasian Conference on Mathematics
and Computers in Sportand Computers in Sport
ANALYST
PERFORMANCE
INDICATORS
PERFORMANCERELIABILITY WHICH ARE
MOSTIMPORTANT?
HOW MUCH
DATA?
EMPIRICALMETHODS
PREDICTION
FROM
VARIANCE
PERFORMANCE
PROFILE COMPARING
PROFILES
7/31/2019 Notational Analysisa Math Perspective
52/91
7th Australasian Conference on Mathematics7th Australasian Conference on Mathematics
and Computers in Sportand Computers in Sport
Comparing profilesComparing profiles
Many research papers have used parametric tests in the past thesehave been found to be slightly less sensitive than the non-parametric
tests, and they did not respond to large differences within the data.
The results of performance analysis are very often recorded asdiscrete events. Clearly, investigating categorical differences in
discrete data using traditional parametric tests of significance (e.g.
ANOVA, based on the continuous symmetric normal distribution) is
inappropriate.
More appropriate statistical methods are promoted based on twokey discrete probability distributions, the Poisson and binomial
distributions.
Nevill, A., Atkinson, G., Hughes, M. and Cooper, S-M. (2002).
Journal of Sports Science, 20, 829 - 844.
7/31/2019 Notational Analysisa Math Perspective
53/91
7th Australasian Conference on Mathematics7th Australasian Conference on Mathematics
and Computers in Sportand Computers in Sport
Comparing profilesComparing profiles
When carrying out tests of significance on continuous data using
regression and analysis of variance (ANOVA), the observed randomvariation is assumed to have a normal distribution.
Clearly, the frequency distribution of discrete events, such as thenumber of shots per rally in tennis or squash, do not follow a normal
distribution.
For example, the frequency distribution of shots per rally of an elitesquash player over a three-game match (total number of rallies = 104)
is discrete, positively skewed and not normally distributed.
Two key distributions for such discrete data are the Poisson andbinomial distributions.
7/31/2019 Notational Analysisa Math Perspective
54/91
7th Australasian Conference on Mathematics7th Australasian Conference on Mathematics
and Computers in Sportand Computers in Sport
Comparing profilesComparing profiles
7/31/2019 Notational Analysisa Math Perspective
55/91
7th Australasian Conference on Mathematics7th Australasian Conference on Mathematics
and Computers in Sportand Computers in Sport
Comparing profilesComparing profiles
Two approaches are proposed and compared using examples
from notational analysis:-
The first approach is based on the classic chi-square test of
significance (both the goodness-of-fit test and the test ofindependence).
The second approach adopts a more contemporary method
based on log-linear and logit models fitted using the statisticalsoftware GLIM.
7/31/2019 Notational Analysisa Math Perspective
56/91
7th Australasian Conference on Mathematics7th Australasian Conference on Mathematics
and Computers in Sportand Computers in Sport
Comparing profilesComparing profilesProvided relatively simple one-way and two-waycomparisons in categorical data are required, both of these
approaches result in very similar conclusions.
However, as soon as more complex models or higher-ordercomparisons are required, the approach based on log-linear and
logit models is shown to be more effective.
Indeed, when investigating those factors and categoricaldifferences associated with binomial or binary response
variables, such as the proportion of winners when attemptingdecisive shots in squash or the proportion of goals scored from
all shots in association football, logit models become the only
realistic method available.
7/31/2019 Notational Analysisa Math Perspective
57/91
7th Australasian Conference on Mathematics7th Australasian Conference on Mathematics
and Computers in Sportand Computers in Sport
FurtherFurther ?? WeWe needneed teststests ofofdifferencedifference
thatthat areare farfar moremore sensitivesensitive..
TheThe winnerwinner ofofthethe womenwomenss
400m400m OlympicOlympic GoldGold inin
SydneySydney performedperformed 11 -- 2%2%
betterbetter thanthan thethe personperson whowho
waswas 8th8th -- oneone isis aa millionairemillionaire
-- thethe otherother??
7/31/2019 Notational Analysisa Math Perspective
58/91
7th Australasian Conference on Mathematics7th Australasian Conference on Mathematics
and Computers in Sportand Computers in Sport
ANALYST
PERFORMANCE
INDICATORS
PERFORMANCE
RELIABILITY WHICH ARE
MOSTIMPORTANT?
HOW MUCH
DATA?
EMPIRICAL
METHODS
PREDICTION
FROM
VARIANCE
PERFORMANCE
PROFILE COMPARING
PROFILES
7/31/2019 Notational Analysisa Math Perspective
59/91
7th Australasian Conference on Mathematics7th Australasian Conference on Mathematics
and Computers in Sportand Computers in Sport
ANALYST
PERFORMANCE
INDICATORS
PERFORMANCERELIABILITYWHICH ARE
MOSTIMPORTANT?
HOW MUCH
DATA?
EMPIRICAL
METHODS
PREDICTION
FROM
VARIANCE
PERFORMANCE
PROFILE COMPARING
PROFILES
MODELLING
PERFORMANCE
7/31/2019 Notational Analysisa Math Perspective
60/91
7th Australasian Conference on Mathematics7th Australasian Conference on Mathematics
and Computers in Sportand Computers in Sport
Empirical ModellingEmpirical Modelling
Stochastic ModellingStochastic Modelling
PerturbationsPerturbations
Artificial IntelligenceArtificial Intelligence
Expert SystemsExpert Systems
Neural NetworksNeural Networks
Modelling
7/31/2019 Notational Analysisa Math Perspective
61/91
M d lliM d lli P b bilitP b bilit
7/31/2019 Notational Analysisa Math Perspective
62/91
7th Australasian Conference on Mathematics7th Australasian Conference on Mathematics
and Computers in Sportand Computers in Sport
ModellingModelling -- ProbabilityProbability
Sport and Chance
Reep and Benjamin (1968)
Ladany and Machol (1977)
Alexander et al (1988)
McGarry and Franks (1996)
Stochastic Modelling
Stochastic ModellingStochastic Modelling
7/31/2019 Notational Analysisa Math Perspective
63/91
7th Australasian Conference on Mathematics7th Australasian Conference on Mathematics
and Computers in Sportand Computers in Sport
Stoc ast c ode gg
Problems:-
too few data
antecedent shot is a naive predictor of the
next shot to be selected
memory-limiting nature of stochastic(Markov) processes, where the future is
predicted only from the present, might be an
insufficient descriptor of sports behaviours
Stochastic ModellingStochastic Modelling
7/31/2019 Notational Analysisa Math Perspective
64/91
7th Australasian Conference on Mathematics7th Australasian Conference on Mathematics
and Computers in Sportand Computers in Sport
gg
Problems:-
Sports analysts have tended to record all the
data from a sports contest and to search thosedata for patterns. Implicit in this method of
analysis are two assumptions.
The first assumption is that if the data areto have information value then they are
likely to be repeated under similar future
circumstances.
The second assumption is that the data
are of equal importance, at least in the long
run.
PerturbationsPerturbations
7/31/2019 Notational Analysisa Math Perspective
65/91
7th Australasian Conference on Mathematics7th Australasian Conference on Mathematics
and Computers in Sportand Computers in Sport
PerturbationsPerturbations
World Congress of Notational Analysis of Sport,Burton Manor, 1992
Downey talked of rhythms in badminton rackets, co-
operation, until there was a dislocation of therhythm a perturbation sometimes
resulting in a rally end situation ( a critical
incident), sometimes not.
PerturbationsPerturbations
7/31/2019 Notational Analysisa Math Perspective
66/91
7th Australasian Conference on Mathematics7th Australasian Conference on Mathematics
and Computers in Sportand Computers in Sport
PerturbationsPerturbations
Harmonic Motion
Fig. 3.1. A schematic example of Simple Harmonic Motion (SHM) from
Weinstein (2004).
PerturbationsPerturbations
7/31/2019 Notational Analysisa Math Perspective
67/91
7th Australasian Conference on Mathematics7th Australasian Conference on Mathematics
and Computers in Sportand Computers in Sport
PerturbationsPerturbations
Figure 1.2. Distance Time Graph for Pilot study 1
0
2
4
6
8
10
12
14
0.0 1.0 2.0 4.0 5.2 7.2 8.4 9.6 11.0 13.0 14.2 15.4 17.0
Time (s)
Server Receiver
Distance (m) Rally 21
0
1
2
3
4
5
6
00:14:13:01
00:14:21:12
00:14:29:23
00:14:38:09
00:14:46:20
00:14:55:06
00:15:03:17
00:15:12:03
00:15:20:14
00:15:29:00
00:15:37:11
00:15:45:22
00:15:54:08
00:16:02:19
00:16:11:05
Time (s)
D
istance(m)
Series1
PerturbationsPerturbations
7/31/2019 Notational Analysisa Math Perspective
68/91
7th Australasian Conference on Mathematics7th Australasian Conference on Mathematics
and Computers in Sportand Computers in Sport
PerturbationsPerturbations
If we study a system only in the linear range of its
operation where change is smooth, its difficult if not
impossible to determine which variables are essential andwhich are not.
Most scientists know about nonlinearity and usually try
to avoid it.
Here we exploit qualitative change, a nonlinear
instability, to identify collective variables, the implication
being that because these variables change abruptly, it is
likely that they are also the key variables when the systemoperates in the linear range.
Scott Kelso, 1999
PerturbationsPerturbations
7/31/2019 Notational Analysisa Math Perspective
69/91
7th Australasian Conference on Mathematics7th Australasian Conference on Mathematics
and Computers in Sportand Computers in Sport
PerturbationsPerturbations
SQUASH
McGarry, Khan & Franks, 1999
SOCCER
Hughes, Dawkins, Davids & Mills, 1998
7/31/2019 Notational Analysisa Math Perspective
70/91
7th Australasian Conference on Mathematics7th Australasian Conference on Mathematics
and Computers in Sportand Computers in Sport
Artificial IntelligenceArtificial Intelligence
- Roger Bartlett
.and Jurgen Perl!
The Decision Making Scope of AIThe Decision Making Scope of AI
7/31/2019 Notational Analysisa Math Perspective
71/91
7th Australasian Conference on Mathematics7th Australasian Conference on Mathematics
and Computers in Sportand Computers in Sport
g pg p
Expert systems :Expert systems :
RuleRule--based.based.
Fuzzy.Fuzzy.
FrameFrame--based.based.
Artificial neural networks :Artificial neural networks :
Biological and artificial neural networks.Biological and artificial neural networks.
TheThe PerceptronPerceptron..
MultiMulti--layer neural networks.layer neural networks.
Recurrent neural networks.Recurrent neural networks.
SelfSelf--organising neural networks.organising neural networks.
Expert SystemsExpert Systems
7/31/2019 Notational Analysisa Math Perspective
72/91
7th Australasian Conference on Mathematics7th Australasian Conference on Mathematics
and Computers in Sportand Computers in Sport
AdvantagesAdvantages
7/31/2019 Notational Analysisa Math Perspective
73/91
7th Australasian Conference on Mathematics7th Australasian Conference on Mathematics
and Computers in Sportand Computers in Sport
Separate knowledge from processing, unlike conventionalSeparate knowledge from processing, unlike conventional
programs.programs.
Provide an explanation facility.Provide an explanation facility.
Can deal with incomplete and vague data.Can deal with incomplete and vague data.
Can model fuzzy human decisionCan model fuzzy human decision--making.making.
Are good for diagnosis.Are good for diagnosis.
ShellsShells for development of expert systems are widelyfor development of expert systems are widely
available (e.g. addavailable (e.g. add--ons to MATLAB).ons to MATLAB).
DisadvantagesDisadvantages
7/31/2019 Notational Analysisa Math Perspective
74/91
7th Australasian Conference on Mathematics7th Australasian Conference on Mathematics
and Computers in Sportand Computers in Sport
gg
Need to acquire knowledge from experts; this is a majorNeed to acquire knowledge from experts; this is a majorproblem.problem.
Very domainVery domain--specific; fast bowling one could not be used forspecific; fast bowling one could not be used forjavelin throwing.javelin throwing.
Opaque relationships between rules.Opaque relationships between rules.
In general, do not have an ability toIn general, do not have an ability to learnlearn..
Have to manage conflicts between rules.Have to manage conflicts between rules.
Ineffective rules searchingIneffective rules searching trawl through all rules in eachtrawl through all rules in each
cycle.cycle.
Expert SystemsExpert Systems -- Performance AnalysisPerformance Analysis
7/31/2019 Notational Analysisa Math Perspective
75/91
7th Australasian Conference on Mathematics7th Australasian Conference on Mathematics
and Computers in Sportand Computers in Sport
Given that they are good diagnostic tools and thatGiven that they are good diagnostic tools and that
systemsystem shellsshells easily available, how widespread is theeasily available, how widespread is the
use of them in PA?use of them in PA?
Not very!Not very!
The reality conflicts with the positive view of theirThe reality conflicts with the positive view of their
utility byutility by LaphamLapham and Bartlett in 1995and Bartlett in 1995(3)(3)..
Artificial Neural NetworksArtificial Neural Networks
7/31/2019 Notational Analysisa Math Perspective
76/91
7th Australasian Conference on Mathematics7th Australasian Conference on Mathematics
and Computers in Sportand Computers in Sport
Allow computers to learn from experience and byAllow computers to learn from experience and by
anologyanology(1)(1)..
A computer program that tries to create a mathematicalA computer program that tries to create a mathematical
model of neurons in the brainmodel of neurons in the brain(2)(2)..
An interconnection of simple adaptable processingAn interconnection of simple adaptable processing
elements or nodeselements or nodes(2)(2)::
Nodes simplified models of brain neurons.Nodes simplified models of brain neurons.
Store experiential knowledge as pattern of connectedStore experiential knowledge as pattern of connected
nodes and synaptic weightings between them.nodes and synaptic weightings between them.
NonNon--linear programs that represent nonlinear programs that represent non--linear systems,linear systems,such as the human movement system and games.such as the human movement system and games.
Originally developed to exploit the power of parallelOriginally developed to exploit the power of parallel
processing, now mostly PC based.processing, now mostly PC based.
AdvantagesAdvantages
7/31/2019 Notational Analysisa Math Perspective
77/91
7th Australasian Conference on Mathematics7th Australasian Conference on Mathematics
and Computers in Sportand Computers in Sport
Learn by experience; in the case of selfLearn by experience; in the case of self--organisingorganising ANNsANNs,,
without a teacher!without a teacher!
Are good for classification, clustering and predictionAre good for classification, clustering and prediction
tasks.tasks.
Can be adapted for inexact or incomplete data throughCan be adapted for inexact or incomplete data through
fuzzyfuzzy ANNsANNs..
Are widely available, e.g. the MATLAB Neural NetworkAre widely available, e.g. the MATLAB Neural NetworkToolbox, and relatively simple programs.Toolbox, and relatively simple programs.
Seem to mimic brain processes.Seem to mimic brain processes.
Provide link to dynamic systems theory as nonProvide link to dynamic systems theory as non--linearlinearprogram representations of nonprogram representations of non--linear biological systems.linear biological systems.
DisadvantagesDisadvantages
7/31/2019 Notational Analysisa Math Perspective
78/91
7th Australasian Conference on Mathematics7th Australasian Conference on Mathematics
and Computers in Sportand Computers in Sport
They are opaqueThey are opaque black boxesblack boxes with no explanation of thewith no explanation of the
reasoning process.reasoning process.
TheThe rulesrules within the nonwithin the non--linear network are not welllinear network are not well
understood; the nonunderstood; the non--linear characteristics may prohibitlinear characteristics may prohibitsimple and understandable rulessimple and understandable rules(1)(1)..
To validate their output, they need test cases for whichTo validate their output, they need test cases for which
output is known.output is known.
They often do not work well for inputs outside the rangeThey often do not work well for inputs outside the range
used for learning.used for learning.
Back propagation is very slow, although widely used forBack propagation is very slow, although widely used for
pattern recognition.pattern recognition.
KohonenKohonen SOMsSOMs need lots of learning data and aren'tneed lots of learning data and aren't
dynamic.dynamic.
ANNS in Performance AnalysisANNS in Performance Analysis
7/31/2019 Notational Analysisa Math Perspective
79/91
7th Australasian Conference on Mathematics7th Australasian Conference on Mathematics
and Computers in Sportand Computers in Sport
As their main function is for classification,As their main function is for classification,
clustering and prediction, and that they are nowclustering and prediction, and that they are now
easily available (but only recently)easily available (but only recently) -- howhow
widespread is their use in PA?widespread is their use in PA?
NOT VERYNOT VERY
They have been used in PA, both in techniqueThey have been used in PA, both in technique
analysis and notational analysis, and in otheranalysis and notational analysis, and in otherbranches of sport and exercise science.branches of sport and exercise science.
ANALYST
7/31/2019 Notational Analysisa Math Perspective
80/91
7th Australasian Conference on Mathematics7th Australasian Conference on Mathematics
and Computers in Sportand Computers in Sport
PERFORMANCE
INDICATORS
PERFORMANCERELIABILITYWHICH ARE
MOSTIMPORTANT?
HOW MUCH
DATA?
EMPIRICAL
METHODS
PREDICTION
FROM
VARIANCE
PERFORMANCE
PROFILE COMPARING
PROFILES
MODELLING
PERFORMANCE
ANALYST
7/31/2019 Notational Analysisa Math Perspective
81/91
7th Australasian Conference on Mathematics7th Australasian Conference on Mathematics
and Computers in Sportand Computers in Sport
PERFORMANCE
INDICATORS
PERFORMANCERELIABILITYWHICH ARE
MOSTIMPORTANT?
HOW MUCH
DATA?
EMPIRICAL
METHODS
PREDICTION
FROM
VARIANCE
PERFORMANCE
PROFILE COMPARING
PROFILES
MODELLING
PERFORMANCE
PREDICTION
7/31/2019 Notational Analysisa Math Perspective
82/91
Performance PredictionPerformance Prediction
Multiple Linear RegressionMultiple Linear RegressionDiscriminant Function AnalysisDiscriminant Function Analysis
Binary Logistic RegressionBinary Logistic Regression
Neural NetworkNeural Network
ExampleExample 2003 Rugby World Cup2003 Rugby World CupPeterPeter O'DonoghueO'Donoghue and Jason Williamsand Jason Williams
7/31/2019 Notational Analysisa Math Perspective
83/91
7th Australasian Conference on Mathematics7th Australasian Conference on Mathematics
and Computers in Sportand Computers in Sport
Individual Human PredictionsIndividual Human Predictions
Expert focus groupExpert focus group
MLRMLR
BLRBLR
ANNANN
Simulation packageSimulation package
ExampleExample 2003 Rugby World Cup2003 Rugby World CupPeterPeter O'DonoghueO'Donoghue and Jason Williamsand Jason Williams
7/31/2019 Notational Analysisa Math Perspective
84/91
7th Australasian Conference on Mathematics7th Australasian Conference on Mathematics
and Computers in Sportand Computers in Sport
Points difference between higher and lower ranked teamsPoints difference between higher and lower ranked teamsis the Y variableis the Y variable
The X variables areThe X variables are
Difference in ranking pointsDifference in ranking points
Difference in distance travelled to tournamentDifference in distance travelled to tournament
Difference in recovery days since last matchDifference in recovery days since last match
Used 137 cases from 1987 to 1999 to do the linearUsed 137 cases from 1987 to 1999 to do the linear
regression model in SPSSregression model in SPSS
Used 40 group matches to see predictions in ExcelUsed 40 group matches to see predictions in Excel
ExampleExample 2003 Rugby World Cup2003 Rugby World CupPeterPeter O'DonoghueO'Donoghue and Jason Williamsand Jason Williams
7/31/2019 Notational Analysisa Math Perspective
85/91
7th Australasian Conference on Mathematics7th Australasian Conference on Mathematics
and Computers in Sportand Computers in Sport
Rugby union is an easier sport than soccer to predict theRugby union is an easier sport than soccer to predict theoutcome for because there is a greater amount of scoring inoutcome for because there is a greater amount of scoring inrugby union and currently less strength in depth in therugby union and currently less strength in depth in the
international game.international game. The most successful machine based method was theThe most successful machine based method was the
simulation package which produced a prediction thatsimulation package which produced a prediction thatrecognised the effect of combined conditional probabilitiesrecognised the effect of combined conditional probabilities
on the overall outcome of the tournament.on the overall outcome of the tournament. Quantitative and computerQuantitative and computer--based prediction methods werebased prediction methods were
more successful at predicting the results of the 2003 Rugbymore successful at predicting the results of the 2003 RugbyWorld Cup than most of the predictions made byWorld Cup than most of the predictions made by
individual humans which were based on qualitativeindividual humans which were based on qualitativeanalysis.analysis.
However, the expert focus group demonstrated thatHowever, the expert focus group demonstrated thathuman expertise still exceeds that of machine basedhuman expertise still exceeds that of machine based
methods.methods.
ANALYST
7/31/2019 Notational Analysisa Math Perspective
86/91
7th Australasian Conference on Mathematics7th Australasian Conference on Mathematics
and Computers in Sportand Computers in Sport
Discriminate between PIs
Performance Indicators?
Predictive Profiling Methods?Comparing data
Modelling?Prediction?
Reliability
Empirical Profiling
PERFORMANCE
INDICATORS
PERFORMANCERELIABILITYWHICH ARE
MOST
IMPORTANT?
HOW MUCH
DATA?
EMPIRICAL
METHODS
PREDICTION
FROM
VARIANCE
PERFORMANCE
PROFILE COMPARING
PROFILES
MODELLING
PERFORMANCE
PREDICTION
7/31/2019 Notational Analysisa Math Perspective
87/91
7th Australasian Conference on Mathematics7th Australasian Conference on Mathematics
and Computers in Sportand Computers in Sport
ANALYST
The answer to the mystery of the universe?
7/31/2019 Notational Analysisa Math Perspective
88/91
7th Australasian Conference on Mathematics7th Australasian Conference on Mathematics
and Computers in Sportand Computers in Sport
ANALYST
42?To get the correct answers youhave to ask the right questions.
7/31/2019 Notational Analysisa Math Perspective
89/91
7th Australasian Conference on Mathematics7th Australasian Conference on Mathematics
and Computers in Sportand Computers in Sport
International Society of Performance Analysis of Sport
Mike HughesInternational Society of Computers in Sport Science
Jurgen Perl
E-Journals-
International Journal ofPerformance Analysis of Sport
International Journal ofComputers in Sport Science
Notational analysisNotational analysis a mathematical perspective.a mathematical perspective.
7/31/2019 Notational Analysisa Math Perspective
90/91
Mike Hughes,Mike Hughes,
Centre for Performance Analysis,Centre for Performance Analysis,
University of Wales Institute Cardiff.University of Wales Institute Cardiff.
THANKS for your attention.THANKS for your attention.
PP
7/31/2019 Notational Analysisa Math Perspective
91/91
7th Australasian Conference on Mathematics7th Australasian Conference on Mathematics
and Computers in Sportand Computers in Sport
PapersPapers
Hughes, M.D. (2004). Performance AnalysisHughes, M.D. (2004). Performance Analysis
a mathematical perspective.a mathematical perspective. EIJPAS,EIJPAS,International Journal of PerformanceInternational Journal of PerformanceAnalysis Sport (Electronic)Analysis Sport (Electronic),, 44, 2, 97, 2, 97 --
139.139.
Hughes, M.D. and Bartlett, R.(2002). TheHughes, M.D. and Bartlett, R.(2002). Theuse of performance indicators inuse of performance indicators inperformance analysis.performance analysis. Journal of SportsJournal of SportsScience 20,Science 20, 739739 754.754.