Sorts Analytics Worksho Sotirios Drikos sdrikos@gmail.com www: drikos.weebly.com Performance Analysis in and Related Sorts. 1
List of Contents Performance Analysis Performance Indicators Net & Wall Games Modeling of sort contests Contents Sorts Performance Analysis Performance Indicators & related Sorts Modeling of Sorts Contests Skill imortance in Men s. Skill imortance in Men s Performance Sorts Concet Aims Procedure SPA is the investigation of actual sorts erformance or erformance in training. Main reason for SPA is to develo an understanding of sorts that can inform decision making to enhance sort erformance to imrove coaching Also non coaching uses of SPA, e.g. media or judging. Indicators 2
Performance Sorts Coaching rocess Concet Coach Observes Aims Athletes erform Coach lans ractice Coach conducts ractice Procedure Performance analyzed Indicators Past results accounted for Performance Sorts Concet Aims Procedure Technical Tactical movement Coach & layer education Modeling using match analysis dtbase Indicators 3
Performance Sorts Concet Aims Data gathering During or after a erformance data During or after a erformance Procedure Indicators Communication of information deending on: the relevant audience (athletes, coaches, judges, media) the aim of analysis (athlete s feedback, decision making, evaluation from judges) Performance Sorts Concet Aims Procedure Indicators Performance indicators are variables that: are valid measurements of imortant asects of sort erformance can be described in an objective measurement rocedure have a known scale of measurement are valid means of interretation 4
Performance Sorts Performance Indicators Technical Performance indicators exress result or uality of erformance SPA is focused on results. done on the basis of result and not of uality of movement. Tactical Indicators are exressed in ratios for easier comarison across layers, teams, chams. Profile Performance Sorts Performance Indicators Technical Tactical Profile Concerned with how well skills are erformed in sorts Positive/negative or winner/error ratios All the skills are not black or white. The degree of difficulty of situation is not taken into consideration Ordinal Scales (usually 3 to 7 levels). validity (More often face & content validity) reliability(reroducibility of the measurement) Imortant for ractitioners (to monitor small but ractically imortant changes) For researchers (to uantify such changes in controlled trials with samles of reasonable size) Pearson s r, Chi-suare, % error and Kaa test were used. 5
Performance Sorts Performance Indicators Technical Tactical of strategy and sort atterns Match indicators Skills lacing (where, from/to) Time of Skills Decision making athletes movement. Profile Performance Sorts Performance Indicators Technical Tactical Profile Maing of the laying surface (Scatter diagrams). 4W are necessary: Who/where/what/when Winners/Losers Models Problem: Interaction between cometitors Solution: Criteria like classification of the cometitors (e.g. 1-8 vs 9-16) Ranking where exists (Tennis, Beach Volley, Suash, Table tennis etc) Problem: Matches with big score differences Solution: Only Ambivalent matches Problem: Multicollinearity Solution: Princial comonents analysis 6
Performance Sorts Performance Indicators Technical Tactical 4 7 3c 3b 3d 3a 8 2 9 Profile 5 6 1 Performance Sorts Performance Indicators A5 A4 A3 A4 A2 A1 6 B5 B4 B3 B4 B2 B1 0 Technical C5 C4 C3 C2 C1 D5 D3 D1 Tactical Profile E5 E3 E1 7
Performance Sorts Performance Indicators Technical 4 3 2 K R 7 8 9 Tactical S I Profile 5 65 6 61 1 Performance Sorts Performance Indicators Technical Tactical Profile Performance indicators are not stable variables (e.g. anthroometrics). Many sources of variability in SPA. Performance rofile is collection of indicators that together characterize the tyical erformance. Multile matches How many? How recent? The greater the dtbase, the more accurate the model but less sensitive to changes of laying atterns. 8
Performance Sorts Performance Indicators Technical Interretation of result Comaring rofiles Athletes in different chamionshis Teams in different eriods Imortant skills Tactical Profile Crucial moments of a contest (momentum) Modeling of sort erformance Simulation of a contest Prediction of a result es. in Net Wall. & Related sorts Sort Contests Proer criterion for categorization. Structure of the Contests Invasion Games with similar structure Common erformance indicators Net Wall Striking Fielding Interaction of oonents during match in the same way. Similar systems of observation & satiotemoral recording 9
& Related sorts Sort Contests Time deendent Common laying area. Invasion Invasion Score deendent Different laying area Net Wall Striking Fielding Innings deendent Change offense/defense er eriod & Related sorts Sort Contests Formal Invasion Net Wall Net / Wall Invasion Games Striking/ Fielding Games Striking Fielding Score Deendent Time Deendent Innings Deendent 10
& Related sorts Sort Contests Invasion Invasion Games Net Wall Goal Throwing Games Try Scoring Games Goal Striking Games Striking Fielding Basketball Handball Waterolo Rugby/ Football Soccer Hockey & Related sorts Sort Contests Net / Wall Invasion Net Games Wall Games Net Wall No Volley Games Bounce and Volley Games No Bounce Games Bounce and Volley Games Striking Fielding Table Tennis Tennis Badminton Suash 11
& Related sorts Sort Contests Striking and Fielding Games Invasion Wicket Base Running Games Net Wall Striking Fielding Cricket Baseball Softball Modeling of sort Winner rofile Simulation Tennis resonse/ seuence rofile imortance Modeling in Net/Wall is easier because of hierarchical structure. In Invasion it is more comlex because of unexected change of ball s ossession and the unstable chronological seuential order of the tactical actions. Each action starts with a serve. No draw. Every action has a winner and a loser. Outcomes & Sort tactics can be carefully analyzed on the basis of robability. 12
Modeling of sort Winner rofile Simulation Tennis resonse/ seuence rofile imortance Skilled actions refer to the ability of the action to reach a ositive outcome. I.i.d. Assumtion A team/or a layer is not influenced: if the revious oint was won or lost (indeendent)- indeendence- If the current oint is of articular imortance, eg. Match oint (identical distribution)- stationarity-. Is i.i.d. assumtion a logical one for net/wall? Hot hand or strike Back to the wall. Modeling of sort Winner rofile Simulation Tennis resonse/ seuence rofile imortance Stochastic (Markov) rocesses:2fold investigation: Scoring Structure: Winner/error rofile Sort tactic: shot resonse/seuence rofile Use of the outcome the winning of a ointas the unit of the robability analysis. Use of Markov chain on the calculation of robabilities to win a game, a set or a match. 13
Modeling of sort Winner rofile Simulation Tennis resonse/ seuence rofile imortance Tennis SRV 1- + = layer A wins when serving 1-= layer A loses when serving - Modeling of sort Winner rofile Tennis/ one game Simulation Tennis resonse/ seuence rofile 0-0 1-1-0 1-0-1 1-2-0 1-1-1 1-0-2 1-3-0 1-2-1 1-2 1-0-3 1-1- 4-0 3-1 1-2-2 1-1-3 1-0-4 4-1 3-2 1-2-3 1-1-4 4-2 3-3 1-2-4 4-3 1-3-4 1-5-3 4-4 1-3-5 adv 1- adv 1- game deuce game imortance = layer A wins when serving 1-= layer A loses when serving 14
Modeling of sort Winner rofile Simulation Tennis resonse/ seuence rofile imortance 0-0 1-1- 1-0 0-1 1- Tennis / Set 1-4-0 3-0 1-2-0 3-1 1-1- 1-2-1 1-1 2-2 1-1- 1-2 0-2 1-1-3 1-1- 0-3 1-0-4 1- = layer A wins when serving 1-= layer A loses when serving = layer B wins when serving 1-= layer B loses when serving 1-5-0 1-4-1 1-3-2 2-3 1-1-4 1-0-5 1-6-0 5-1 1-4-2 1-3-3 1-2-4 1-1-5 1-0-6 6-1 1-5-2 1-4-3 3-4 1-2-5 1-1-6 6-2 5-3 1-4-4 1-3-5 1-2-6 6-2 1-5-4 4-5 1-3-6 6-4 5-5 4-6 Modeling of sort Winner rofile Simulation Tennis resonse/ seuence rofile imortance SRV Team A 1- + - SRV Team B 1- = team A wins when serving 1-=team A loses when serving = team B wins when serving 1-= team B loses when serving + - 15
Modeling of sort Winner rofile Simulation Tennis resonse/ seuence rofile imortance / Set 0-0 1-1-0 1-1- 0-1 = team A wins when serving 1-=team A loses when serving = team B wins when serving 1-= team B loses when serving 2-0 1-1-1 1-1 1-0-2 3-0 1-2-1 2-1 1-2 1-2 2-1 1-2 1-0-3 4-0 1-3-1 3-1 2-2 3-1 2-2 2-2 1-3 1-3 2-2 3-1 2-2 2-2 1-3 1-3 1-0-4 5-0 4-1 4-1 3-2 4-1 3-2 3-2 2-3 4-1 3-2 3-2 2-3 3-2 2-3 2-3 1-4 1-4 2-3 2-3 3-2 4-1 3-2 3-2 2-3 3-2 2-3 2-3 1-4 2-3 1-4 1-4 0-5 : 6 rotations/ 6 different robs to win when serving Rot Team A SRV Team B SRV 6 45% 42% 5 20% 28% 4 40% 34% 3 29% 32% 2 42% 35% 1 41% 39% Modeling of sort Winner rofile Simulation Tennis resonse/ seuence rofile imortance Shot selection behaviors and outcomes for racket sorts Seuence rofile and outcomes for Unit of analysis is the shot behavior / seuence of secific skills that leads to the outcome Probs of the analysis are obtained from data collected from observation. Game is described by Categorical states strictly defined by individual skills. Creation of a transition. Elements of the exress the robability of moving from one state (skill) to another state (skill) and to a +/-oint. 16
Modeling of sort Winner rofile In alternative order between the two ayers Simulation Tennis resonse/ seuence rofile Serve Game action Receive Neutral Offense Stroke osition Forehand Backhand Pivot Stroke Stroke Direction Short Forehand Long Forehand Short backhand Stroke techniue Tosin Drive Smash Outcome Point/ fault imortance Defense Control Backhand turn Long Backhand Close to body Net Ball Fli Cho Choing Short 9 Next rally Modeling of sort Winner rofile Simulation Tennis resonse/ seuence rofile imortance Stroke osition Forehand Backhand Pivot Backhand turn Point F_A F_B B_A B_B Pv_A Pv_B Bt_A Bt_B P_A P_B F_A 56% 8% 20% 16% F_B 68% 12% 16% 4% B_A 32% 5% 47% 16% B_B 50% 37% 6% 6% Pv_A 28% 50% 6% 16% Pv_B Bt_A Bt_B 17
C Pass Set/ Attack 1 + - C Modeling of sort Serve + - resonse/ seuence rofile Direct Attack Block Free Ball Dig Set/ Attack 2 C + - imortance Modeling of sort Seuence rofile Pass Set/ Attack 1 C + - Serve C + - Direct Attack Block Free Ball Dig Set/ Attack 2 C + - 18
Modeling of sort Winner rofile Simulation Tennis resonse/ seuence rofile 59X62 imortance Skills Imortance in Men s. Data Method of analysis Results Conclusions Suggestions Method of analysis Most recent world Chamion in Men (POLAND, 2014) Transition Matrix 59 X 62. Last two columns are terminal moves (oint + or oint- ) for the team under study. P i is the robability for a skill to end u in a oint after two subseuent game moves. n 1 k t i k 1, k 1 Pi P( Yt oint Yt Si ) PY ( t 2 oint Yt 1 Sk ) P( Yt 1 S Y S ) 19
Skills Imortance in Men s. Data Method of analysis Results Conclusions Method of analysis Measure Imortance score ( Ii). Measure of imact & uncertainty for a skill (Fellingham & Reese, 2004). I i E( Pi y) V( Pi y) Posterior mean Standard deviation Suggestions Skills Imortance in Men s. Data Method of analysis Results Conclusions Suggestions Method of analysis Assumtions 1 st assumtion: Scoring for each skill is i.i.d. 2 nd assumtion: Patterns are first order Markov chains. Pi P( Yt oint Yt Si ) P( Yt 2 oint Yt 1 S ) 1 i 20
Skills Imortance in Men s. Data Method of analysis Results Conclusions Suggestions Method of analysis Model Simle multinomial model to estimate transition & success robabilities ik ik P( Yt 1 Sk Yt Si ) For each skill we assume multinomial likelihood f ( yi1,... yi, n, yi, n 1, yi, n 2 1,... i, n, i, n 1, i, n 2) k M yik ik i Skills Imortance in Men s. Data Method of analysis Results Conclusions Suggestions Method of analysis Prior distribution We use a conjugate Dirichlet rior distribution of the tye f 1 ( ) k i A i k M k i Prior estimations from exert coaches. Low weight to exerts/coaches oinion. Multily 0.1X N i, (10% additional of data oints). All success robabilities & imortance scores were calculated using a Monte Carlo scheme of 10.000 iterations. 21
M Q QMR 0.05 Q 0.95 M Skills Imortance in Men s. Skill Imortance score Success Probability QMR Data Method of analysis Results Pass in Float 5 (1) 27.6 0.581 Pass in Jum 6 (2) 27.4 0.589 Pass in Float 6 (3) 27.2 0.569 Pass in Jum 5 (4) 27.0 0.593 Pass in Jum 4 (5) 24.9 0.548 Pass in Float 4 (6) 22.8 0.539 Attack 1 MF uick (7)21.9 0.704 Conclusions Srv Float 3 (8) 17.8 0.332 Attack 1 LS uick (9) 17.2 0.557.. Suggestions Attack 2 MB uick (22) 10.2 0.717 1.320 Attack 2 MF uick (31) 7.5 0.738 1.258 Attack 1 STR (41) 4.8 0.722 2.093 M Q QMR 0.05 Q 0.95 M Skills Imortance in Men s. Data Method of analysis Results Conclusions Suggestions Extreme values Variety of the data Quantile Mid range Ratio QMR= M Q Q M 0,95 0,05 Values >1.2 indicates negative skewness Samle size small and osterior variance high High imortance but not often execution 22
Skills Imortance in Men s. Data Method of analysis Results Conclusions Suggestions Introduction of a new sulementary index (QMR) to estimate skills. Serve: is a disadvantage for to level s men volleyball. Increased difficulty of serve does not connect linearly to the outcome. Pass: The enalty for overass (level 2) is higher than for the ass off the net (level 3). Pass level 6 has no advantage comared to ass level 5 Setting & attack in comlex 1: Quick temo is more imortant than high temo. Imortance of back row attack All levels of organized attack 1 have higher imortance score than attacks 2. Setting & attack in comlex 2: Set out of system is most imortant. Quick temo better than high temo sets Skills Imortance in Men s. for coaches Data Method of analysis Results Conclusions Serve absolute hard or safe. No errors without aces. Comlex 1 (side out oint) is very imortant. Comlex 2 (break oint): Better rearation of unredictable situations (attack out of system). Suggestions 23
Skills Imortance in Men s. Data Method of analysis Results Conclusions for researchers Imrovement of model Use of ast data of team s erformance as rior information. Standardized team rofile. On line use. Indications for coaches decisions during match. Suggestions Skills Imortance in Men s. Data Method of analysis Results Thank you for your attention! Conclusions Suggestions The End 24