Available in US
- The block-charge call that taught me to separate calls into two buckets
- Defining the line between discretionary and clear
- The Pelechrinis finding and where the bias actually lives
- How the discretionary share maps onto markets
- The referees whose discretionary patterns I track most closely
- The technical-foul connection and where it matters
- The data discipline of separating call types
- The trader's lens on discretionary calls
The block-charge call that taught me to separate calls into two buckets
Sacramento, late January 2023. Domantas Sabonis catches a pass in the post, a defender slides into position a quarter-second before contact, and the trail official points the other way. Block. Sabonis dunks. Free throw. The defender’s coach loses his mind on the sideline. I watched the replay six times and the call could have gone either way honestly. That is the moment I decided to start tracking discretionary fouls separately from clear ones in my own data set, because the two categories carry different information and produce different betting signals.
That separation is the most underused analytical lens in NBA refereeing, and the reason most public referee analysis is mediocre is precisely that it averages discretionary and clear fouls into a single number. The discretionary bucket is where referee identity actually expresses itself. The clear bucket is where referee identity barely matters at all. Let me walk you through what the distinction looks like operationally and how to read it for betting purposes.
Defining the line between discretionary and clear
A clear foul, in NBA officiating terminology, is a call where contact is severe enough or rule-coded enough that any qualified official would make the same call. Flagrant contact, obvious shooting fouls on jump-shooters with no other player nearby, dead-ball technicals after explicit confrontation – these calls have functionally zero discretionary content. The official is recording a rule violation that happened in plain sight.
A discretionary foul is a call where the contact is real but the rulebook leaves the application open to the official’s judgement. Block versus charge on a driving guard. Hand-check on a perimeter handler. Off-arm clearing space on a post catch. Continuation versus non-continuation on a hard fall after a shot release. Each of these contact patterns produces a foul call somewhere between sixty and eighty per cent of the time at the league level, with the remaining fraction either no-called or called the opposite direction. The discretionary content is the residual variance after rule coding is stripped out.
The discretionary share of total NBA fouls runs at roughly fifty to fifty-five per cent of the call count, with the share fluctuating game to game depending on style. A grinding defensive matchup with lots of physical post play produces a higher discretionary share than a perimeter-oriented matchup with predominantly clean catch-and-shoot offence. The discretionary share matters because it is the share where referee identity actually drives variance in the call count.
The Pelechrinis finding and where the bias actually lives
The Pelechrinis analysis of 7,498 personal foul calls in Last Two Minute reports identified persistent home bias in personal foul calling, with the bias concentrated in calls where the contact was contested or ambiguous on review. The bias was much smaller – close to zero in some sub-samples – on calls where the contact was clear and the rule application uncontroversial. That finding is the empirical foundation for the discretionary-versus-clear framework I am describing.
The mechanism is the same one we see in the broader cognitive psychology literature on expert judgement. When the rule application is unambiguous, the official’s identity matters very little because the decision is essentially automated. When the rule application is ambiguous, the official’s calling tendencies, the contextual cues from the crowd, the prior pattern of the game, and the body language of the players all combine to produce a directional bias that is detectable in the data. The Pelechrinis data set captures this because L2M reviews specifically identify which calls were correct, which were incorrect, and which were no-calls that should have been called. That granularity is what allows the bias to be measured on the discretionary subset rather than the headline foul-count number.
The directional component of the bias is also informative. The bias in the Pelechrinis sample favoured the home team specifically. Discretionary calls in close games in the L2M window produced fewer fouls against the home team than the clear-call sample predicted. The home crowd, the referee attentional load, the in-game tone – whatever combination of inputs produces the bias – operates almost entirely through the discretionary channel.
How the discretionary share maps onto markets
The betting application of this framework runs through three channels. First, the total fouls market on UK coupons where it is offered. A high-discretionary-share matchup produces more variance in call counts, which means the closing line on total fouls is less reliable. The market typically prices total fouls at the season-average rate for the two teams, which ignores the matchup-specific discretionary share. The bettor’s edge is in identifying matchups where the discretionary share will run high and pricing the total accordingly.
Second, the player-foul-trouble props that have survived the post-2025 reform environment. A star player who is foul-prone – Joel Embiid is the canonical case – sees their foul-trouble risk vary directly with the discretionary share of the matchup. Embiid against a perimeter team that does not initiate contact produces a low-discretionary game with manageable foul exposure. Embiid against a physical defensive front line produces a high-discretionary game where the discretionary share of his foul exposure runs significantly higher. The market does not price this matchup-specific variance well.
Third, the spread market through the foul-out and foul-trouble channel. A star fouling out in a close game has spread-relevant consequences. The probability of that fouling-out scenario depends on the matchup’s discretionary share, which depends partly on the referee crew’s discretionary tendencies. High-discretionary crews working physical matchups produce the highest foul-out rates and the largest unpriced spread variance.
The referees whose discretionary patterns I track most closely
The names matter less than the patterns. Some officials run a high discretionary share with a home-leaning directional bias. Some run a high discretionary share with a balanced directional pattern. Some run a low discretionary share regardless of matchup – these are the officials who tend toward letting marginal contact go and only blowing the whistle when the contact is clear. The third group produces lower foul counts overall but more predictable foul distributions, which is a different kind of betting input.
Eric Lewis’s 61.1 per cent road-team foul rate works almost entirely through the discretionary channel. His clear-foul call distribution is roughly balanced. His discretionary distribution skews heavily toward the road team. The same pattern reverses for Natalie Sago, whose 63.3 per cent home-team foul rate runs through her discretionary calls rather than her clear ones. That structural similarity between two officials with opposite directional signals is one of the cleaner examples of how discretionary versus clear is the right analytical cut, rather than the headline directional split.
The discretionary share is also what most cleanly separates crew chiefs from non-chiefs in their on-court impact. The chief slot involves more late-game replay calls, more block-charge primary positioning, and more end-of-quarter foul management. Each of those is a discretionary-heavy call type. The two non-chief officials handle a larger share of away-from-ball action and routine off-ball coverage, where the discretionary share is lower. That is the mechanical reason crew chief identity disproportionately drives crew-level outcomes – the chief is doing more discretionary work.
The technical-foul connection and where it matters
The relationship between discretionary fouls and technical fouls is a separate channel that interacts with the framework. A technical foul is partly discretionary – the official decides whether language or behaviour crosses the line – and partly behavioural, since the player is choosing to confront the official. The interaction matters because some referee crews are quicker to issue technicals in response to discretionary-call disputes, which compounds the foul-trouble effect on stars who are vocal complainers.
The post-call dispute pattern is a measurable input. Crews that issue technicals quickly after discretionary calls produce more cumulative foul disruption than crews that absorb the complaint and move on. The compounding effect on star foul trouble is the betting-relevant output. A star with five fouls and a technical against a quick-tech crew is in a different risk position than the same star against a patient crew. The deeper read on this specific interaction runs through the NBA technical fouls betting piece, which works through the markets that respond most directly to technical-foul patterns.
The data discipline of separating call types
Building a discretionary-share data set is more work than building a foul-count data set, and that is partly why the analysis is not more common in public referee research. You need access to play-by-play data, you need a coding scheme for classifying each foul call as discretionary or clear, and you need to apply that coding consistently across hundreds or thousands of games. The L2M reports give you the closest thing to a ground truth on which calls were marginal, but L2M only covers the final two minutes of close games. For the rest of the call sample, the coding is judgement-driven.
The practical compromise I work with is to use the L2M-coded subset as a calibration set, then apply a simpler rule-based classifier to the rest of the call data. Block-charge calls on drives are coded discretionary by default. Shooting fouls on jump-shooters with no other player nearby are coded clear. Hand-checks on perimeter handlers are coded discretionary. The classifier is imperfect but produces a discretionary-share estimate that correlates around 0.85 with the L2M-derived ground truth on the subset where both are available.
That correlation is good enough for betting work. The signal-to-noise ratio on referee patterns is high enough that the residual classification error does not destroy the predictive value, and the alternative – coding nothing and treating all fouls as equivalent – is much worse. The pragmatic discipline is to use the imperfect classifier consistently rather than to wait for perfect data.
The trader’s lens on discretionary calls
The reason discretionary fouls deserve their own analytical category is that they are where officiating identity, contextual pressure, and matchup-specific variance all interact. Clear fouls are essentially automated. Discretionary fouls are the bet-relevant subset. Every signal that matters in referee analysis – directional splits, home court advantage, late-game accuracy, foul-trouble propagation – runs through the discretionary channel rather than the headline foul count.
If you take only one analytical practice from this piece, make it the discipline of separating the two call types before drawing any conclusions about referee patterns. The work of building the classification is dull. The output is the analytical edge that separates serious referee-driven betting from the cartoon version where you just look up which crew has the highest foul rate and call it a model. The cartoon version produces noise. The discretionary-first framework produces signal.
Articles
Published by the nbarefbettin team.
