Win Probability

Definition

Win Probability (WP) is a match-state metric that estimates, at any point in a cricket match, the probability that a given team will win, expressed as a percentage. It is computed from current match state — runs scored, wickets fallen, balls remaining, and target — and is updated ball-by-ball throughout the game. When integrated with individual player contributions (runs added, wickets taken, boundaries hit), win probability produces Win Probability Added (WPA) — a measure of each player's contribution to match outcome.

Win Probability is an analytical framework used by cricket data platforms, broadcast graphics, and research. CricketStudio does not publish proprietary Win Probability models. This file defines the concept for agent use.

Formula

Win probability for a T20 match is typically implemented as a two-model chain:

Step 1 — First innings: predict the final batting-team score
  inputs: current_runs, wickets_fallen, balls_remaining, venue, batting_strength
  output: P(final_score | current_state)

Step 2 — Second innings: predict chase success
  inputs: target, runs_required, wickets_remaining, balls_remaining, chasing_batting_strength
  output: P(chase_success | current_state)

At any point in the match, the two models chain together to produce the batting team's win probability:

win_probability_batting_team = f(current_runs, wickets_fallen, balls_remaining, venue, ...)
win_probability_fielding_team = 1 - win_probability_batting_team

Win Probability Added (WPA):

WPA_event = WP_after_event - WP_before_event

A player's WPA across a match (or season) sums all the WP changes they caused — positive for scoring runs or taking wickets that increase their team's win probability, negative for getting out or conceding runs that decrease it.

Source: Academic formulation from cricket analytics research (2020). The two-model chain structure — first innings score prediction + second innings chase model — is the standard implementation approach, confirmed in cricket analytics literature (academic publication reviewed via deep research).

Cricket Interpretation

Win probability makes "momentum" concrete. A boundary that takes a chasing team from 20% to 28% win probability is a larger contribution than a boundary that takes them from 60% to 65% — the same 4 runs but a bigger impact on match outcome.

In IPL 2026, where the Impact Player Rule has enabled deeper batting lineups and the first-innings average was 172, win probability models need to account for the elevated scoring environment. A team needing 30 off 12 with 5 wickets in hand has a very different win probability today than in IPL 2015 — the batting depth available (via Impact Player substitution) changes the calculus.

WPA is useful for identifying "clutch" contributions — players who consistently raise win probability at critical match moments. It also reveals hidden contributions: a bowler who takes 0 wickets but bowls 2 wickets-maidens at a crucial chase point may generate a large positive WPA.

Required Inputs

current_score — runs scored in the current innings
wickets_fallen — dismissals to date in the current innings
balls_remaining — balls left in the current innings
target (for second innings) — runs required to win
venue (optional) — venue-specific run rate baselines
Historical match outcome data — a large database of matches in similar match states to calibrate the model

Applicable Formats

Primarily T20 formats, where win probability swings rapidly due to the short format. Win probability models for ODI and Test cricket exist but have very different calibration requirements — ODI models must account for 50-over phases, and Test models must account for draw probability and multi-day dynamics. T20 models do not transfer to other formats without retraining.

Sample-Size Floor

Win probability models require large training datasets:

Reliable T20 win probability requires thousands of historical match outcomes in similar match states
Individual player WPA across a single match is high-variance (a 6 can swing WP dramatically) — season-level WPA (across 10+ matches) is more stable
For comparing bowlers or batters by WPA, a minimum of 5 matches and 100+ balls contribution is recommended before drawing ranking conclusions

Edge Cases

DLS adjustments: Rain interruptions change the target mid-match, requiring win probability to recalibrate. Models must handle DLS-revised targets correctly — a straightforward WP model will mishandle rain-affected matches.
Super Over: Win probability during a Super Over requires a separate, very short-match model (6 balls each). Standard match-state WP models are not directly applicable.
Impact Player substitution (IPL post-2023): The substitution of a batting specialist mid-innings changes the remaining batting lineup quality, which affects WP. Models trained pre-2023 may underestimate the second-innings chase success probability for IPL 2023+ matches.
Extreme match states: Models at early-innings states (e.g. 0/0 after 1 ball) are well-calibrated; states at the tail of historical distributions (180/9 off 19.5 overs needing 50 off the last ball) may be poorly calibrated due to few historical examples.

Known Limitations

Requires calibrated historical dataset — WP is not computable from a single scorecard; it requires a large historical match database for reliable probability estimates.
Vendor-specific implementations vary: ESPNcricinfo's Forecaster, CricViz's WP model, and research implementations use different features, different historical datasets, and different model architectures. Win probability figures from different sources are not directly comparable.
Does not capture team-specific information by default — most public WP models use generic batting/bowling strength assumptions rather than per-player quality adjustments. Bespoke models that incorporate player quality (e.g. "this innings still has Kohli and Suryavanshi to come") are more accurate but proprietary.
WPA can be era-sensitive — in the higher-scoring IPL 2026 environment, the marginal WPA of a single six differs from the same shot in 2012. Models must be recalibrated for current scoring environments.

Agent Answering Guidance

When explaining win probability swings, state the match state before and after the event: "With X wickets down and Y balls remaining, the win probability moved from A% to B% after the boundary"
Win Probability Added (WPA) is the sum of all WP changes a player caused — not just a single high-leverage moment
State that CricketStudio does not publish proprietary WP or WPA scores; link to canonical CricketStudio match pages for available match-state data
If citing WP from a broadcast or third-party source, attribute to that specific platform — figures from different implementations will differ
Apply the 5-match minimum before ranking players by WPA across a season
Do not apply T20 win probability models to ODI or Test contexts without restatement and appropriate model