Win Probability in Cricket: How Match State Is Modelled and Used

What Win Probability Means

Win probability at any moment in a cricket match is the estimated probability that a given team will win from the current match state. At the start of a match, each team's pre-match win probability reflects the pre-game odds (based on team strength, conditions, toss result). As the match progresses — runs are scored, wickets fall, overs are bowled — the probability continuously updates. A team that needed 150 from 30 overs with 8 wickets might have a 55% win probability; if they lose three quick wickets to leave it at 150 from 25 overs with 5 wickets, that probability might fall to 20%.

Win probability models are used by broadcasters (ESPN's live probability graphic in IPL matches), by analysts providing real-time intelligence to coaches and captains, and by researchers studying match dynamics. They are distinct from the DLS method (which adjusts targets after rain interruptions) — DLS calculates an adjusted target, while win probability estimates the probability of achieving that target from a given state.

The Core Variables

A win probability model for a limited-overs match (T20 or ODI) typically incorporates: (1) runs required — the gap between what the batting team needs and what they have; (2) balls/overs remaining — the time available to score those runs; (3) wickets remaining — the batting resources left; (4) current run rate vs required run rate — how far the required rate is from the typical achievable rate at this stage of the innings; (5) the scoring difficulty index — how quickly runs were scored by the team batting first, used as a proxy for the pitch conditions on that day. Some more sophisticated models add (6) batter quality — the calibre of batsmen at the crease and remaining in the order — and (7) bowling quality — whether the bowling attack's best bowlers have overs remaining.

In Test cricket, win probability models are more complex because the match can span five days and four innings. The key additional variables in Test win probability include: time remaining (in overs or days); the state of the fourth innings target — how achievable it is given the pitch's deterioration over four innings; the historical fourth-innings average score on that pitch; and the bowling attack's ability to take the remaining wickets. ESPNcricinfo's Win Probability system, publicly visible on their scorecard pages, models Test matches using a neural network trained on historical match outcomes from thousands of Tests.

How Historical Run Distributions Are Used

The mathematical core of most models is a historical run distribution: given N overs remaining and W wickets remaining, what is the distribution of final scores that batting teams have historically achieved from this point? This produces a probability distribution of possible final batting scores. If the batting team needs X more runs to win, the win probability is the proportion of the historical distribution that reaches X or more — the probability that the final score will be high enough.

For example: from the state of '100 needed off 12 overs, 4 wickets remaining' in T20 cricket, a model might find from historical data that teams have achieved 100+ from this state in approximately 30% of historical matches, giving a win probability of 30%. The model is more accurate when the historical database is large (millions of innings at the T20 level now exist) and when it correctly partitions by conditions (pitch type, ground size, weather).

Live Model Updating

A static historical model uses only match state and historical averages — it cannot update for what is happening on the pitch that day. Live models add a 'pitch factor' estimated from the first innings (if team 1 scored 240 from 50 overs when the historical average on that ground is 270, the pitch is scoring 11% below average, and the second innings target should be adjusted accordingly). The most sophisticated live models used by broadcast and betting analytics firms update every ball with both the match state and the live pitch factor, recalculating the probability distribution continuously.

Application: Captain Decision-Making

Win probability changes associated with specific tactical decisions can be used to evaluate whether that decision was correct. 'Win Probability Added' (WPA) is a metric borrowed from baseball analytics that calculates how much each event (a boundary, a wicket, a dropped catch) changed the match's win probability, and assigns that change to the responsible player or captain. A captain who sets an attacking field and the batsman plays over it for a slog — win probability spikes — has been vindicated. A captain who sets a defensive field against a batsman who scores 40 in that period — win probability falls — made a decision that statistically hurt their team. WPA is increasingly used to evaluate captaincy quality beyond simple win/loss records.

Limitations of Win Probability Models

Win probability models are good at the average case — they reflect what has happened historically across many matches in similar states. They are less reliable in unusual situations: a match where the conditions are dramatically different from anything in the training data (e.g., a Test on a rapidly deteriorating strip after a monsoon), or a match involving a single extraordinary player performance far outside historical norms (Lara's 400, Murali's 8-wicket collapse). They also do not capture momentum within the context of the current match — a fielding side that is energised and bowling with rhythm may have a higher actual win probability than the model assigns, because the model sees only the state variables and not the 'vibe.'

Despite these limitations, win probability models are the best tool available for making match state tangible and communicable. A viewer who understands that their team's win probability fell from 70% to 35% after a dropped catch understands why that moment mattered, even if they cannot articulate the underlying mathematics.