Expected Threat vs Expected Goals: Where the Markets Diverge

[Betting Forum]

Expected goals has become, over the last decade, the default analytical currency in football betting discussion. The upgrade from raw shot counts to quality-adjusted expected goals was a genuine improvement in predictive accuracy, and the widespread adoption of xG as a betting reference point has improved the general quality of market analysis significantly.

The problem is that xG has a specific and structural blind spot: it only measures the probability of scoring from the moment a shot is taken. All the value-creating play that precedes the shot - the passes that advance the ball into dangerous zones, the carries that break defensive lines, the combinations that draw defenders out of shape - is invisible to xG unless it directly produces a shot. Teams that excel at creating value in the pre-shot phase but are less clinical in the shot phase are systematically undervalued by xG-based market models. Teams that are chaotic in the pre-shot phase but create high-quality shots when they do shoot are systematically overvalued.

Expected threat exists to address this blind spot. It measures how much each action - pass, carry, cross - increases the probability of a goal being scored in the next few possessions, not the current shot alone. The divergence between a team's xT and their xG is information that the market doesn't adequately incorporate, and that divergence is the subject of this article.

Recommended USA sportsbooks: Bovada, Everygame | Recommended UK sportsbook: 888 Sport | Recommended ROW sportsbooks: Pinnacle, 1XBET

What xT Measures and How​

The expected threat framework was developed and published by Karun Singh in 2018 and subsequently refined by various analytics researchers and commercial data providers. The core concept is to value each on-ball action by measuring how much it changes the probability of scoring in the subsequent three or four possessions.

The calculation divides the pitch into zones and assigns each zone a probability of leading to a goal - both from that zone directly and through subsequent actions. A pass from the defensive third to the central midfield increases the probability of a goal being scored in the next few actions by a small amount. A pass from the midfield directly into the penalty area increases the probability significantly more. A dribble that breaks the last line of defence and carries the ball into the final third increases it more still. Each action's xT value is the difference between the goal probability in the zone you started from and the goal probability in the zone you moved the ball to.

The aggregate xT for a player or team across a match or season is the total of all these incremental probability increases from their actions. A team that moves the ball efficiently into high-threat zones accumulates high xT. A team that struggles to penetrate into threatening positions accumulates low xT regardless of how many total passes they play.

The key distinction from xG: xT rewards all of the value-creating actions in a possession sequence, not just the terminal shot. A possession sequence that progresses from the defensive third through midfield into the final third and then loses the ball without creating a shot has created real xT across the progression actions but generates zero xG. Over many such sequences, a team that consistently creates high xT without converting it to xG is doing valuable work that doesn't show up in the shot-based model.

The xT-xG Divergence and What It Means​

Teams whose xT significantly exceeds their xG are creating dangerous attacking positions more than their shot record suggests. The gap between the positions they're reaching and the shots they're generating means either their final-ball quality is poor, their shot selection is unusually good (they're waiting for higher-quality opportunities rather than shooting from all dangerous positions), or there's a structural reason their xT-to-shot conversion is lower than average.

Each of these explanations has different betting implications.

The poor final-ball quality version: a team that consistently reaches dangerous positions but fails to deliver the final pass or shot is likely to show xT-xG divergence that's structurally persistent rather than variance. Their underlying attacking play is good. Their execution in the final phase is poor. This is a predictable and stable characteristic rather than a temporary fluctuation. Betting the over for this team based on their high xT without understanding the final-ball quality deficiency produces errors.

The high shot selectivity version: a team that's choosing not to shoot from positions that most teams would shoot from, waiting instead for even higher-quality opportunities, produces low shot count with high xG per shot. Their xT is high because they're reaching dangerous zones frequently. Their xG per shot is high because they're only shooting from the best positions. This is a positive characteristic that the xG model captures correctly, and the xT excess is a signal of consistent value creation rather than a warning.

The structural conversion deficit version: some tactical systems are specifically designed to generate dangerous positions that other teams then shoot from. The team that creates the chance gives the ball to the player in the best position. That player's shot appears in the xG record. The build-up player's contribution appears only in the xT record. Teams with specific tactical designs around this type of play - elaborate combination sequences that culminate in a specific player's shooting opportunity - show xT-xG divergence that's structural and reflects the system rather than a quality deficit.

Understanding which version of divergence applies to a specific team is the analytical work that determines whether the xT-xG gap is a warning, a positive signal, or simply a structural feature of their playing style.

Teams With High xT, Low xG: The Undervaluation Case​

The betting case for teams with high xT and low xG depends on whether the divergence represents a conversion deficit that's likely to correct or a structural feature that's likely to persist.

The version with the clearest betting value is the high xT, low xG team where the divergence reflects finishing variance rather than structural deficit. A team creating dangerous positions consistently but failing to score at the rate their chance creation warrants is showing a gap that's more likely to close than persist. The xG itself captures some of this - a team that creates high xG without scoring is the standard over-xG-performers-vs-under-scorers analysis from the xPoints article. But xT adds an additional layer: a team creating high xT without even reaching high shot quality is one phase further back in the possession sequence from where the xG model starts measuring.

This team is undervalued by the xG model specifically because the xG model doesn't see the threatening positions they're reaching and losing the ball from. Their actual attacking quality - as measured by the positions they generate, not the shots those positions lead to - is higher than the xG record suggests. The expected table based on xG underestimates this team. The market, anchored to the actual table and the xG table, underestimates them further still.

The betting implication is directionally similar to the xPoints analysis: a team whose xT significantly exceeds their xG, whose xG exceeds their actual goals, and whose actual points are below their xG-expected points is a team with compounding underperformance across three layers of quality measurement. Each layer of underperformance that's likely to revert toward mean represents upward pressure on their future results that the market isn't adequately incorporating.

Finding this team - high xT, low xG, low actual goals, low actual points - and identifying the specific matches where the reversion is most likely to manifest is a specific and tractable analytical task. The matches where their high-xT attacking play faces defensive opponents that their type of build-up play tends to exploit most effectively are the highest-value early-season betting opportunities.

Teams With Low xT, High xG: The Overvaluation Warning​

The reverse case - low xT generating high xG and high actual goals - is the structural overvaluation scenario where the market may be anchoring to a positive results record that the underlying play doesn't support.

A team generating low xT but high xG is creating rare but high-quality shots without consistently building through dangerous positions. This profile most commonly arises from one of three sources: a specific counter-attacking style that generates high-quality shots from infrequent transitions, exceptional individual quality in the final-third players who convert low-probability carry-and-shoot sequences into xG-weighted events, or finishing variance that has produced high xG from a relatively small number of dangerous positions.

The counter-attacking version is a genuine tactical strength that can persist. Teams that defend deep and exploit space behind the press generate shots from counter-attacks that tend to be high-quality - few defenders between ball and goal, fast transitions - without requiring the gradual xT accumulation of a build-up team. These teams can produce low xT and high xG sustainably because the tactical approach is designed to skip the value-creation phase and arrive directly at the value-execution phase.

The individual quality version is less sustainable at the team level. If the high xG is driven by specific individual conversion quality - a striker converting at well above expected rate from carry-and-shoot sequences that most strikers would be below expected on - the team xG will revert when the conversion rate normalises, regardless of what the xT record shows.

The variance version - high xG from a small number of dangerous positions in a sample where those positions happened to arrive more frequently than the xT predicts is normal - is the clearest case for the over to the market eventually repricing downward. A team whose recent high xG record is built on a small number of exceptional matches rather than consistent positional threat generation is showing a fragile attacking foundation that the market's form-based pricing has interpreted as strength.

The Player-Level xT Application​

Expected threat has a specific and underused application at the individual player level for prop markets, distinct from the team-level application described above.

A player whose xT contribution is high but whose xG contribution is low is creating value in the pre-shot phases that doesn't show up in the goals, assists, and shots statistics that prop markets price from. This is a version of the pressed-from-front striker analysis - invisible contribution that the statistical record misses - but specifically for creative players whose value is in the building-phase rather than the shooting-phase.

A progressive midfielder whose xT per 90 is significantly above the average for their position but whose xG per 90 is average or below is a player whose true contribution to their team's attacking threat is systematically underpriced in any prop market that uses shots and goals as inputs. Their assist probability is more accurately estimated from their xT contribution than from their historical assist rate, because xT captures the quality of positions they're creating that their teammates then fail to finish - which eventually shows up in assists when the finishing variance normalises.

This is the specific individual player prop application of the team-level analysis. The high-xT, low-xG midfielder is the individual embodiment of the team-level divergence, and their prop markets for assists, key passes, and shot-creating actions are all priced below their true expected contribution level for the same structural reason.

FBref's player-level data doesn't directly carry xT but carries the component metrics that allow approximation - progressive carries, progressive passes, passes into the final third, passes into the penalty area. These are the specific actions that generate the highest xT values per action and their volume per 90 minutes provides a reasonable proxy for a player's xT contribution without requiring the full xT calculation.

Data Access and Calculation​

xT data is less universally available from free sources than xG, which is part of why it's less incorporated into market pricing. This accessibility gap is part of what keeps it as amber information in the taxonomy from earlier in this series.

StatsBomb's open data includes xT calculations for the matches in their open dataset, which covers certain competitions at certain time periods rather than comprehensive current season data. Their commercial data carries full xT but requires subscription access.

The Analyst and various football analytics platforms have published xT tables for major European leagues at the team level, typically mid-season and end-of-season. These are useful for the historical calibration described in this article but don't provide real-time match-level xT data.

The practical free-source approximation is to use FBref's progressive passing, progressive carry, and zone-entry data as xT proxies. These metrics capture the high-xT-value actions without the full calculation. A team with high progressive pass volume into the penalty area, high progressive carry volume into the final third, and high zone entries into the penalty area is generating high xT without the specific numerical calculation. The approximation is less precise but directionally reliable for the large-divergence cases where the betting value is concentrated.

The investment in commercial xT data - either StatsBomb's subscription or access through one of the data aggregators who license StatsBomb's output - is worthwhile for bettors who make enough prop market decisions in this area for the precision to matter. For bettors using it as one input among many in a small number of annual decisions, the FBref approximation is sufficient.

When xT and xG Converge: The Neutral Signal​

Worth explicitly addressing the scenario where xT and xG are broadly aligned, because this is the majority of cases and the neutral interpretation matters.

A team whose xT is roughly proportionate to their xG is converting their dangerous position creation into shots at an average rate. The pre-shot phase and the shot phase are both working at approximately normal efficiency. For this team, the xG analysis described in the xPoints article is the appropriate framework - their xPoints divergence from actual points reflects shot quality and conversion variance rather than any pre-shot phase issue.

The xT-xG comparison adds information primarily in the divergence cases. When the two metrics are broadly aligned, they're confirming each other rather than revealing a new dimension. The aligned team's analysis doesn't require the additional xT layer - the xG model is capturing their performance with reasonable completeness.

The temptation to find xT-xG divergence where it doesn't meaningfully exist - forcing an interpretation from small differences that are within normal variance - is the analytical error to avoid. The same discipline described in the xPoints article applies here: a small divergence over a ten-game sample is noise. A large divergence over twenty or more games is signal. The threshold for treating the xT-xG gap as betting-relevant information rather than statistical noise is higher than most bettors who encounter the metric for the first time would set it.

FAQ​

Q1: Is there a specific xT-xG divergence threshold above which the market's underpricing becomes reliably actionable, similar to the xPoints divergence thresholds described earlier?
The xT-xG divergence threshold is harder to specify precisely than the xPoints divergence because the relationship between xT and xG is noisier than the relationship between xG and actual results. A rough working threshold: a team whose xT-implied attacking quality ranks them three or more positions higher than their xG-implied quality, sustained over fifteen or more games, is showing divergence large enough to act on. Below this threshold the variance makes it difficult to distinguish genuine systematic divergence from the normal noise in both metrics. The three-position threshold is a rough calibration rather than a validated figure - the appropriate threshold for your specific application depends on the competition, the sample size, and how many other supporting variables are pointing in the same direction.

Q2: Does the xT framework apply to defensive analysis as well as offensive, and if so, what does high defensive xT allowed mean for a team's clean sheet probability?
The xT framework applies symmetrically to defence - the xT allowed represents how much positional threat opponents have been allowed to create against a team, capturing the pre-shot defensive quality in the same way it captures offensive quality. A team allowing high xT but low xG against is facing teams that are reaching dangerous positions but not converting them to shots - analogous to the offensive version. A team allowing low xT but high xG against is vulnerable to counter-attacks and direct play that generates efficient shots without the build-up phase. The defensive xT allowed metric gives a more complete picture of a team's defensive quality than xG against alone, specifically capturing whether defensive solidity comes from preventing dangerous positions or from preventing shots from those positions. A team with low xT allowed and low xG against is genuinely defending well in both phases. A team with high xT allowed but low xG against is defending the shot phase adequately while being outplayed in the build-up, which is a more fragile defensive model that's more likely to deteriorate under sustained pressure.

Q3: How do you reconcile xT analysis with the observation that many successful counter-attacking teams deliberately avoid building up xT in favour of direct play - does this make xT less applicable to those teams?
Counter-attacking teams are the specific case where xT analysis requires the most careful interpretation, and the reconciliation is in understanding what the low xT is telling you about the team rather than treating it as a universal negative signal. A counter-attacking team with low xT, moderate xG, and results broadly consistent with their xG has a coherent tactical identity where the low xT is a design feature rather than a quality deficit. The xT-xG gap for this team is structural and sustainable. The xT analysis becomes relevant for counter-attacking teams primarily in two situations: when their xG starts to fall below what even their low-xT, low-shot-volume approach should produce, suggesting even the counter-attack quality is declining; and when opponents have adapted to their counter-attack approach and the direct play is being increasingly neutralised, which shows as both xT and xG declining simultaneously. Counter-attacking teams that are genuinely under tactical pressure show declining metrics in both dimensions as the direct play stops working. Teams that are simply playing their designed style show stable low xT paired with stable moderate xG. The distinction is the trend rather than the level.