Do UFC Judges Favor the Previous Round Winner? What Data Says

Quick answer

The best controlled study says they might: Paul Gift analyzed live, round-by-round scores from major Nevada and California MMA events. After accounting for recorded fight performance, judges still leaned toward the fighter who won the previous round. That result makes previous-round bias a real concern for UFC judging, not a proven rule. The study is old, nonexperimental, and doesn't give fans a simple win-rate shortcut. Current Unified Rules require judges to assess every round on its own, so five-minute-old action should carry no official weight.

Data snapshot: 4,155 potential judge-round observations from major Nevada and California MMA events, 2001–2012. A separate 2024 model covered 309 judges. Current rules were amended August 6, 2025.

Do UFC judges favor the previous round's winner?

Gift's study found evidence that they do after controlling for measured performance within the round being scored, and that last part matters.

A raw comparison would tell us little because fighters often win several rounds through better skill, fitness, or planning. Winning round two after winning round one isn't suspicious by itself.

Gift instead used detailed performance statistics from post-fight video analysis and matched them to each judge's live round score. His models then tested whether outside signals still moved the score after the recorded action was considered.

The previous-round winner effect remained after performance controls, while the paper also found favoritism toward larger betting favorites and insurmountable leads. It found no support for a titleholder advantage or extra punishment after a point deduction.

The table doesn't rank effect sizes, and the public abstract gives no clean percentage such as “previous-round winners take 55% of the next rounds.” Any article claiming one from this paper needs the full estimate and model context.

The sound conclusion is narrower: sequence appeared to matter when the rules say it shouldn't.

What does “previous-round bias” mean?

Previous-round bias means a judge's last decision may influence the next one, but the same fighter can still fairly win consecutive rounds.

Imagine Fighter A clearly takes round one while round two stays close. If the first verdict nudges the judge toward A again, the second score reflects more than round-two action.

Gift's paper can't look inside a judge's mind, but it reports a residual association after statistical controls. The author describes the evidence as nonexperimental, which rules out a clean causal claim.

The controls aren't a perfect copy of judging: a database can count strikes, takedowns, attempts, and positions. It can't fully capture a fighter being badly hurt, a near submission, or the view from one cageside seat.

An unexplained score isn't automatically wrong because some of the gap may be real action that the variables missed. Some may be judging bias, and the study separates those possibilities better than a fan poll without erasing them.

Are UFC rounds supposed to be scored independently?

Yes. The Unified Rules say at least three judges evaluate each round individually for the more effective fighter.

The 10-point must system gives 10 to the round winner and nine or fewer to the loser; a rare even round is 10-10. Judges first weigh effective striking and grappling; aggressiveness and fighting-area control matter only when that work is even.

Nothing there rewards winning the previous round, and a judge shouldn't “make up” for an earlier score with a later close round.

The same rule explains why a fighter can lose four minutes yet win the round: one damaging sequence may outweigh lower-impact volume. That is round scoring, not memory bias.

This distinction gets lost during disputed UFC decisions because fans often total strikes across the whole fight. Judges don't: they produce five separate round scores in a five-round bout, then add the points.

Why could the previous round affect the next score?

The study establishes a pattern rather than its cause, and several explanations fit the evidence.

• An expectation forms: after choosing Fighter A once, a judge may see the next close exchange through that earlier view.
• Performance carries over: the same speed, pressure, or fatigue can shape consecutive rounds without appearing fully in recorded stats.
• The running score changes the fight: a fighter who believes they're ahead may defend more, while their opponent may accept greater risks.
• Judges value actions differently: one judge may reward damaging counters, while another may put more weight on grappling advances.

The final point has newer support: Benjamin Holmes, Ian McHale, and Kamila Żychaluk modeled individual MMA judges in a 2024 paper. Their data included 309 judges, with a median of only 12 scored rounds and a maximum of 1,573.

They found personal judging preferences for different actions, and those preferences can decide a bout in their model. That doesn't retest Gift's exact result, but it warns against treating every judge as one average decision-maker.

What Jones vs Reyes shows about score sequences

Jon Jones's UFC 247 decision over Dominick Reyes is a useful scorecard example, but it cannot prove previous-round bias decided the fight.

All three judges gave Reyes round one and Jones rounds four and five, leaving the disagreement in the middle. Chris Lee gave Reyes round two and Jones round three; Marcos Rosales reversed them, while Joe Soliz gave both to Jones.

MMA Decisions collected 8,864 fan scorecards, and 71.6% had Reyes winning 48-47. Fans gave Reyes round two on 88.3% of submissions and round three on 84.7%.

Those numbers show public disagreement, but they don't isolate why any official chose Jones in a close round. A fan's view may include commentary, replays, and broadcast angles, while officials scored live.

The card sequence still teaches a practical lesson because a final score hides the path that created it. Studying UFC scorecard bias requires judge-by-judge rounds, rather than only the announced winner.

Adesanya vs Gastelum is a useful counterexample

Israel Adesanya's UFC 236 win over Kelvin Gastelum shows why a strong finish isn't automatically recency bias.

All three judges produced the same 48-46 card, with Gastelum winning rounds one and four. Adesanya won rounds two and three, while each judge scored the fifth 10-8 for him.

That last round changed the total through its wider scoring margin, but it didn't rewrite the first four rounds. This is exactly what the 10-point must system allows.

Fans sometimes call any late swing “recency bias,” but that usage is too loose. True previous-round bias concerns one completed round affecting the next; a deserved fifth-round 10-8 is different.

Can fighters or prediction models use this finding?

Fighters shouldn't build a strategy around a possible judging error because winning the current round remains the useful target.

The finding supports clear round endings because a late damaging exchange can carry legitimate value within the full five minutes. It also leaves less room for a close call.

Prediction models need a sharper distinction: a model of performance should exclude reputation, current score, and the prior round's official winner. Those fields describe context, not present-round effectiveness.

A model built to forecast an actual judge's score may include them because it predicts human decisions, including possible bias. It isn't declaring what the score should be.

Don't turn the result into a betting rule because the paper gives no simple probability edge for the next round. Even a real average effect can be too small, too old, or too dependent on the judges involved.

How solid is this data?

Gift's study is the best direct evidence we found because it links performance measures to individual live scores. Its 4,155 potential observations offer far more information than a list of controversial decisions.

The limits are concrete: the sample covers major MMA events in Nevada and California from 2001 through 2012. It isn't a modern UFC-only dataset, and judging language, fighter tactics, commissions, and data collection have changed since then.

The 4,155 figure is the potential judge-round count, though models may use fewer observations when a required variable is missing. The study's public page also doesn't provide an easy effect size for readers to convert into a win rate.

Measured actions are imperfect stand-ins for damage and danger, while judges share the same bout from different angles. Repeated scores from the same officials also mean the observations aren't 4,155 fully independent people.

The 2024 Holmes study shows that judge preferences differ, but it doesn't directly replicate the previous-round effect under today's rules. Its sample is uneven: the median is 12 rounds, but one judge has 1,573.

Treat previous-round bias as a credible historical signal that deserves a modern audit. Don't treat it as proof that a disputed UFC card was corrupt or incompetent.

FAQ

Do UFC judges score each round independently?

Yes: the Unified Rules tell at least three judges to evaluate each round individually, with effective striking and grappling first. A fighter's score, reputation, and result in the previous round aren't official criteria.

Does winning one UFC round help you win the next?

It may affect a close score, according to Paul Gift's 2001–2012 study. The prior winner retained a favorable residual association after performance controls, but consecutive round wins aren't usually biased.

Is previous-round bias the same as recency bias?

They overlap, but previous-round bias specifically tests whether the last round's winner affects the next score. Recency bias can also mean overweighting action near the end of the current round.

Can one controversial UFC fight prove judging bias?

No, because one fight can reveal disagreement but not its cause. You need round-level scores, performance data, many bouts, and controls for fighter quality before claiming a stable UFC judging bias.

Should an AI UFC model account for judge bias?

Only if it predicts judges rather than ideal scores, since a performance model should judge the current round's action. A scorecard forecast may use judge history and sequence, while clearly labeling that different goal.