Sports analytics and performance data

In short

Sports analytics is the collection, measurement and statistical interpretation of data about athletes, teams, matches and fans to support selection, training, tactics and commercial decisions. It has two halves: performance analytics (what happened on the field and inside the athlete's body) and business analytics (tickets, sponsorship, fans, broadcast). In India it drives the IPL and other franchise leagues, SAI/Khelo India talent programmes, the BCCI's National Cricket Academy, and vendors such as Sportz Interactive, Stats Perform/Opta, Sportradar, CricViz and Catapult/STATSports resellers. University papers on "Athletes and analytics", sports science and sports marketing examine metric definitions, computation and interpretation; recruiters test formula fluency (strike rate, economy, NRR, xG, ACWR), tool skill (Excel, Power BI, SQL, Python/R, SportsCode/Nacsport) and the ethics of athlete data under the DPDP Act, 2023.

Explained

1. What sports analytics is, and its levels

• Definition: a systematic process - define a question, collect data, clean it, analyse it, visualise it, and convert it into a coaching or commercial decision. Data without a decision is scorekeeping, not analytics.
• Descriptive (what happened): averages, strike rates, possession %, distance covered.
• Diagnostic (why it happened): phase-of-play video coding, shot maps, pitch maps, pressure situations.
• Predictive (what will happen): win probability, injury risk, projected auction value, opponent's likely tactic.
• Prescriptive (what to do): optimal batting order, bowling match-ups, training load plan, ticket price.
• Data types: *event data* (discrete actions - a pass, a shot, a ball bowled, a raid), *tracking data* (x, y, z coordinates of every player and the ball, typically 10-25 frames per second), *physiological data* (heart rate, HRV, sleep, blood markers), *subjective data* (RPE, wellness questionnaires), *outcome data* (results, standings), *business data* (CRM, ticketing, OTT viewing).
• Moneyball / sabermetrics origin: Oakland Athletics (baseball) used on-base percentage instead of scouts' impressions to buy undervalued players - the founding case study that analytics finds *mispriced skill*, not just describes it.

2. How performance data is captured

• GPS/GNSS vests (Catapult, STATSports, Polar) - 10-18 Hz outdoor tracking: distance, speed zones, accelerations, decelerations; accuracy falls for very short, sharp movements.
• LPS (local positioning systems) using UWB radio - used indoors (basketball, kabaddi, badminton halls) where satellites fail.
• Optical/computer-vision tracking - multiple fixed cameras plus AI track all players and the ball without any device on the athlete; used in broadcast-grade leagues.
• IMU (inertial measurement unit) - accelerometer + gyroscope + magnetometer inside the vest pod; gives jumps, collisions, change of direction and the vendor-specific composite PlayerLoad.
• Ball and projectile tracking - Hawk-Eye (multi-camera triangulation) for LBW/DRS, line calls in tennis and badminton; radar/Doppler systems (TrackMan, PitchVision-type) for speed, spin and swing.
• Physiological and medical - chest-strap HR, HRV on waking, GPS-derived HR zones, sleep bands, force plates for jump height and asymmetry, isokinetic dynamometry, blood lactate, DEXA/skinfolds for body composition.
• Video coding software - SportsCode, Nacsport, Dartfish, Hudl: an analyst tags events live into a timeline, producing clip packages plus countable event data.
• Manual/scorer data - ball-by-ball scoring (CricHQ-type apps, BCCI scoring), still the backbone of Indian domestic cricket analytics.

3. Sport-by-sport performance metrics

• Cricket - batting: average, strike rate, balls per boundary, control %, false-shot %, dot-ball %, runs by phase (powerplay/middle/death), runs against spin vs pace, wagon-wheel zones, MVPI/Impact Index-style composite ratings.
• Cricket - bowling: economy rate, bowling average, bowling strike rate, dot-ball %, average speed, revolutions per minute (spin), release point, pitch map (line and length), beaten-outside-off %, phase-wise economy.
• Cricket - match: net run rate, required run rate, DLS (Duckworth-Lewis-Stern) par score using resource percentages from a published table, win probability, pitch-wise par scores. Note the ICC uses DLS for interrupted limited-overs matches; the older Duckworth-Lewis method was superseded by the Stern-updated version.
• Football: expected goals (xG), expected assists (xA), progressive passes, PPDA (passes allowed per defensive action - a pressing measure), possession %, pass completion %, field tilt, pass networks, shot maps, pressures, expected points.
• Kabaddi (Pro Kabaddi): successful raid %, raid points, tackle success %, do-or-die raid conversion, super raids, defensive strike rate - heavily used in PKL auctions.
• Basketball: points per possession, offensive/defensive rating, true shooting % , effective FG%, usage rate, PER, shot-location heat maps, plus/minus.
• Hockey: circle entries, penalty-corner conversion %, press success, turnover locations.
• Athletics/swimming: split times, stroke rate, velocity curves, reaction time, personal-best progression, season plan against peaking target.
• Load metrics (any sport): total distance, high-speed running (HSR) distance, sprint count, acceleration/deceleration counts, metres per minute (intensity), PlayerLoad per minute.

4. Training load, injury risk and return to play

• External load = what the athlete did (distance, sprints, bowling overs, jumps, serves). Internal load = how the body responded (HR, RPE, HRV).
• Session RPE (sRPE) = duration in minutes × RPE on Borg CR-10; summed into weekly arbitrary units (AU) - cheap, validated and the standard method in Indian academies with no GPS budget.
• Training monotony = mean daily load ÷ SD of daily load; strain = weekly load × monotony. High monotony with high load is the classic illness/injury pattern.
• ACWR (acute:chronic workload ratio) = 7-day load ÷ rolling 4-week average weekly load; a band of roughly 0.8-1.3 is treated as the "sweet spot" and above ~1.5 as a spike. Important exam caveat: the ACWR's predictive validity is contested in recent literature (coupling of numerator and denominator, small samples), so it is used as a conversation-starter, not a diagnosis.
• Fitness testing: Yo-Yo Intermittent Recovery Test (level and distance), 2 km time trial, 40 m sprint, countermovement jump, sit-and-reach, bronco/beep test. The BCCI has used a Yo-Yo pass mark (raised over time, reported around 17.1) and a 2 km time trial; such benchmarks are revised by circular, so quote the current board standard rather than a remembered number.
• Monitoring wellness: daily questionnaire (sleep, soreness, mood, stress, fatigue), morning HRV, body mass change for hydration.
• Return-to-play analytics: graded load progression against pre-injury baselines, limb symmetry index from force plates, GPS-matched "did he reach his normal top speed?" checks before clearance. The NCA (Bengaluru) and SAI's National Centre of Sports Science and Research (NCSSR) run this work in India.

5. The analytics workflow and methods

1. Question framing with the coach ("why do we lose the death overs?") - not "here is a dashboard".
2. Data collection - vests fitted, cameras calibrated, taggers briefed on a fixed code window/definitions.
3. Cleaning - drop GPS dropouts and satellite-loss segments, reconcile tagger disagreement, standardise player IDs.
4. Analysis - descriptive statistics (mean, median, SD, percentiles), z-scores to compare a player with squad norms, correlation and regression, clustering to find player types, logistic regression or gradient boosting for win/injury probability, Markov models for state-to-state play.
5. Visualisation and delivery - one-page pre-match sheet, clip package, live bench tablet, post-match report within 24 hours.
6. Evaluation - did the recommendation change behaviour and did the outcome move? Keep a decision log.
7. Tool stack commonly asked for in Indian job ads: Excel (pivot tables, Power Query), SQL, Python (pandas, scikit-learn) or R, Power BI/Tableau, SportsCode/Nacsport, Catapult/STATSports cloud, APIs from Opta/Sportradar/CricViz.
8. Reliability and validity: a metric must be *reliable* (same event tagged the same way twice) and *valid* (actually measures what it claims). Inter-rater reliability checks and vendor comparability (two GPS brands are not interchangeable) are standard exam points.

6. Business and commercial analytics

• Auction and recruitment analytics: role-based shortlists with cost per unit of expected contribution; IPL/PKL/ISL teams model purse allocation, overseas-slot constraints and retention value. Analytics prices *roles*, not reputations.
• Fan and CRM analytics: ticket yield and dynamic pricing, membership churn, merchandise basket analysis, stadium footfall, OTT watch-time and drop-off, sponsorship exposure valuation from broadcast logo-visibility data.
• Fantasy sports and prediction products were a large consumer of Indian player data. The Promotion and Regulation of Online Gaming Act, 2025 prohibits online money games and their advertising while promoting e-sports and online social games; major real-money fantasy platforms moved to free-to-play formats, and challenges to the Act have been litigated - so treat real-money fantasy as not permitted and check the latest judicial position before building anything on it. Betting and gambling fall in the 40% GST demerit slab introduced on 22 September 2025.
• Data rights: match data and tracking feeds are commercially licensed by leagues/boards; collecting and reselling a league's data without rights invites contractual and IP action.

7. Governance, ethics and the Indian institutional setting

• Athlete data is personal data. The Digital Personal Data Protection Act, 2023 with the DPDP Rules, 2025 governs it; the operative Data Fiduciary duties (notice, consent, breach intimation, erasure) commence on 13 May 2027, and until then the IT Act section 43A route and the SPDI Rules, 2011 apply to sensitive data such as health records. Health and biometric readings are the most sensitive category in any athlete programme - consent forms, retention limits and access control are the analyst's responsibility.
• Employment and contracts: wearable use, data ownership on transfer, and publication rights are contract clauses; the four Labour Codes in force from 21 November 2025 govern employment conditions of salaried staff at academies and franchises.
• Anti-doping: NADA India under the National Anti-Doping Act, 2022, applying the WADA Code; whereabouts and testing data are separate from performance data and must not be merged casually.
• Indian institutions: Sports Authority of India (SAI), NCSSR, Khelo India and the Target Olympic Podium Scheme (TOPS) fund sports science support; NSNIS Patiala trains coaches; BCCI's NCA handles cricket testing and rehab; National Sports Governance Act, 2025 reshaped federation governance - mention it as the governing statute without inventing clause detail.
• Ethical lines: no selection decision from a single metric; de-identify data shared with sponsors or media; never use injury data to suppress a player's market value; be explicit about model uncertainty.

Key points

• Sports analytics - turning match, athlete and fan data into decisions; four levels: descriptive, diagnostic, predictive, prescriptive.
• Event vs tracking data - discrete tagged actions vs continuous x/y coordinates of all players and the ball.
• External load - what the athlete did; internal load - how the body responded. Both are needed.
• sRPE - duration (min) × RPE (0-10) = arbitrary units; the cheapest valid load measure.
• ACWR - 7-day load ÷ 4-week average weekly load; 0.8-1.3 sweet spot, >1.5 a spike; evidence is contested.
• Monotony and strain - sameness of training and its consequence; high strain precedes illness.
• Strike rate (batting) - runs per 100 balls; economy rate - runs conceded per over.
• Net run rate - runs scored per over minus runs conceded per over, full quota counted when a side is bowled out.
• DLS - resource-percentage method for rain-affected limited-overs targets; replaced plain Duckworth-Lewis.
• xG - probability a shot becomes a goal, from distance, angle, body part and situation; sums to an expected score.
• PPDA - passes allowed per defensive action; low value = intense pressing.
• PlayerLoad - vendor-specific accelerometer composite; not comparable across brands.
• Z-score - how many standard deviations a player is from squad mean; the standard way to normalise mixed metrics.
• Elo rating - rating updated by actual minus expected result; basis of many team-strength tables.
• Pythagorean expectation - win share predicted from points/runs scored and conceded.
• Reliability vs validity - repeatability of measurement vs measuring the right thing; test both before trusting a metric.
• DPDP Act, 2023 / Rules, 2025 - athlete health and biometric data is personal data; Data Fiduciary duties operative from 13 May 2027, s.43A and SPDI Rules apply until then.
• Online money gaming - prohibited by the Promotion and Regulation of Online Gaming Act, 2025; real-money fantasy products moved to free-to-play.

Formulas and examples

Example 1 - A T20 batter's and bowler's season card. Batter: 1,247 runs in 38 innings with 5 not-outs off 1,026 balls.

• Average = 1,247 ÷ (38 − 5) = 1,247 ÷ 33 = 37.79 runs per dismissal
• Strike rate = (1,247 ÷ 1,026) × 100 = 121.54 runs per 100 balls

Bowler: 42 wickets, 1,560 runs conceded off 2,010 balls.

• Overs = 2,010 ÷ 6 = 335.0 → Economy = 1,560 ÷ 335 = 4.66 runs per over
• Bowling average = 1,560 ÷ 42 = 37.14 runs per wicket; Bowling strike rate = 2,010 ÷ 42 = 47.86 balls per wicket

Reading: an average of 37.8 is high for T20 while a strike rate of 121.5 is only modest - this is an anchor's profile, not a finisher's; the bowler's economy is good while his average and strike rate are poor - he contains rather than takes wickets.

Example 2 - Net run rate in a league table. A team scored 1,120 runs in 196.0 overs and conceded 1,045 runs in 198.3 overs (198 overs and 3 balls = 198.5 overs).

• Runs per over scored = 1,120 ÷ 196.0 = 5.7143
• Runs per over conceded = 1,045 ÷ 198.5 = 5.2645
• NRR = 5.7143 − 5.2645 = +0.450

Rule to remember: if a side is bowled out, its full quota of overs is used in the denominator, not the overs actually faced - which is why collapses hurt NRR badly.

Example 3 - Weekly load, monotony, strain and ACWR. Daily sRPE loads for one week (Mon-Sun): 90 min × 6 = 540; 75 × 8 = 600; rest = 0; 60 × 4 = 240; 105 × 7 = 735; 45 × 9 = 405; rest = 0.

• Weekly load = 540 + 600 + 0 + 240 + 735 + 405 + 0 = 2,520 AU
• Mean daily load = 2,520 ÷ 7 = 360 AU; SD of the seven values = 268.9 AU
• Monotony = 360 ÷ 268.9 = 1.34; Strain = 2,520 × 1.34 = 3,377 AU
• Chronic load: last 28 days total 8,960 AU → mean weekly = 8,960 ÷ 4 = 2,240 AU → ACWR = 2,520 ÷ 2,240 = 1.13 (inside the 0.8-1.3 band)
• If the next week jumps to 3,500 AU against a chronic average of 2,400 AU, ACWR = 3,500 ÷ 2,400 = 1.46 - a spike; the practical response is to cap HSR volume and add a recovery day, not to stop training.

Example 4 - xG versus goals. Shots in a match carried xG of 0.76 (penalty), 0.09, 0.04, 0.33, 0.12 and 0.05.

• Team xG = 0.76 + 0.09 + 0.04 + 0.33 + 0.12 + 0.05 = 1.39
• The team scored 3, so it over-performed xG by +1.61 in this match. Interpretation: finishing variance or exceptional quality - over a season, teams converge towards their xG, so persistent over-performance is a flag, not proof.

Example 5 - Elo update and a z-score. Team A rating 1,720, Team B 1,640, K = 30, A wins.

• E_A = 1 ÷ (1 + 10^{(1640 − 1720)/400}) = 1 ÷ (1 + 10^−0.2) = 1 ÷ (1 + 0.6310) = 0.6131
• A's new rating = 1,720 + 30 × (1 − 0.6131) = 1,720 + 11.6 = 1,731.6; B's = 1,640 − 11.6 = 1,628.4

Z-score: a winger covers 10,850 m against a squad mean of 9,600 m with SD 750 m → z = (10,850 − 9,600) ÷ 750 = +1.67, i.e. in roughly the top 5% of the squad for volume.

Where it is used

• Franchise leagues: IPL, Women's Premier League, ISL, Pro Kabaddi and Prime Volleyball teams employ performance analysts, strength-and-conditioning staff and data scientists for auction modelling, match-ups and load management.
• National programmes: SAI's NCSSR, Khelo India centres, TOPS-supported athletes and NSNIS Patiala use testing batteries, GPS and video analysis; the BCCI National Cricket Academy, Bengaluru, runs fitness benchmarking and rehab analytics.
• Data and media companies: Sportz Interactive, Stats Perform/Opta, Sportradar, CricViz, Hawk-Eye Innovations and broadcasters producing live graphics, win-probability overlays and OTT statistics.
• Hardware and software vendors and their Indian channel partners: Catapult, STATSports, Polar, TrackMan, SportsCode/Nacsport - roles in deployment, training and support.
• Job roles: performance analyst, video analyst, sports data analyst/scientist, S&C coach, athlete-monitoring coordinator, scouting analyst, fan-insights and ticketing analyst, e-sports analyst.
• Beyond elite sport: school and university sports science labs, physiotherapy and rehab clinics, fitness tech and wearables start-ups, sports-marketing agencies valuing sponsorship exposure.

Interview questions

• Q: What is the difference between external and internal load? A: External load is the physical work performed (distance, sprints, overs bowled); internal load is the athlete's physiological and perceptual response to it (heart rate, HRV, RPE). Two athletes can do the same session and carry very different internal loads.
• Q: How do you calculate session RPE and why is it popular in India? A: Duration in minutes multiplied by the Borg CR-10 RPE, giving arbitrary units. It needs only a stopwatch and an honest athlete, so academies without GPS budgets can still monitor load validly.
• Q: Explain ACWR and its criticism. A: It is the last 7 days' load divided by the rolling 4-week average weekly load, with roughly 0.8-1.3 treated as safe. It is criticised because the numerator is part of the denominator, thresholds were derived from small samples, and it has not replicated consistently - so it informs conversation rather than decides selection.
• Q: What does expected goals actually tell a coach? A: It converts each shot into a scoring probability from distance, angle, body part and situation, so a team can see whether it is creating good chances independent of whether they went in. It says nothing about defensive organisation or the scoreline context.
• Q: A player's batting average is high but his strike rate is low in T20. How do you present that? A: Show phase-wise data: he may be valuable in the powerplay or on slow pitches and a liability at the death. Role-specific metrics, not a single season figure, decide team balance.
• Q: Why do you not compare PlayerLoad across two GPS brands? A: It is a proprietary accelerometer composite with different filtering and weighting per vendor, so the units are not equivalent. Comparisons are valid only within one system and firmware.
• Q: How is NRR affected when a team is bowled out? A: Its full quota of overs is used in the denominator rather than the overs actually faced, so a collapse damages NRR far more than a narrow chase does.
• Q: What legal framework governs athlete health data in India? A: The Digital Personal Data Protection Act, 2023 with the DPDP Rules, 2025; the main Data Fiduciary duties - notice, consent, breach intimation, erasure - commence on 13 May 2027, and until then IT Act section 43A and the SPDI Rules, 2011 cover sensitive personal data.
• Q: Is real-money fantasy sport a legitimate analytics market today? A: The Promotion and Regulation of Online Gaming Act, 2025 prohibits online money games and their advertising, and major platforms shifted to free-to-play; betting and gambling also fall in the 40% GST slab. Confirm the current judicial position before building a product on it.
• Q: You have one week before a match and 20 metrics. What do you give the head coach? A: One page: three or four opponent weaknesses with clips, our own trend on the two KPIs the coach has prioritised, and availability/load flags. Everything else goes in the appendix.
• Q: How do you check whether a metric is trustworthy? A: Test reliability (the same event coded twice by the same and different analysts gives the same value) and validity (it correlates with the outcome it claims to predict), and state the sample size behind any claim.
• Q: Give an example of analytics finding mispriced value. A: The Oakland Athletics' use of on-base percentage over scouting impressions; the franchise-league equivalent is buying a death-overs bowler on phase-wise economy rather than on headline wickets.

Common mistakes

• Dividing batting runs by innings instead of by dismissals, or forgetting to convert balls to overs (÷6) before computing economy rate.
• Using overs actually faced instead of the full quota in NRR when a side is bowled out.
• Quoting ACWR as an injury predictor without stating the 0.8-1.3 band's weak and contested evidence base.
• Mixing load units - comparing sRPE arbitrary units with GPS metres, or one vendor's PlayerLoad with another's.
• Treating correlation as causation: "players who run more win more" ignores that winning teams chase the game differently.
• Reporting 25 metrics instead of the two or three a coach will act on, and delivering them after the next session has been planned.
• Handling athlete injury, menstrual-cycle or blood data without a consent trail, retention limit or access control - a data-protection breach as well as an ethical one.

Related topics

Athlete management and player contracts · Sports psychology · Sports marketing and sponsorship · Sports law and governance in India · Business statistics and data visualisation · Exercise physiology and strength and conditioning