How often does the higher-rated player win?
We looked at 7,198 USTA league matches played in 2026, using the dynamic rating every player had going into each one. The higher-rated side won 72% of the time. A small edge is close to a coin flip; a big one almost always holds. Beyond the ratings, recent wins and a team’s record matter most.
The short answer. Favorites won 72% of matches. With a rating gap under 0.10 they won about 58%; with a gap of 0.30 or more, about 91%. Our match predictor turns any gap into a win chance, and on matches it had not seen, its chances held up.
What we looked at
We used 7,198 adult league matches played from January 9 to November 2, 2026: 5,336 doubles and 1,862 singles. For each one we used the estimated dynamic rating each player had going into the match, so nothing that happened in the match leaked into the prediction. In doubles we compared the two teams’ average ratings.
We left out matches where a player had no rating yet (new and self-rated players), defaults and retirements, and matches with fewer than six games. Ratings and results are as published on TennisRecord.com, which estimates the same unpublished USTA number this site does.
How the rating gap changes the odds
The favorite is the side with the higher rating going in. Here is how often the favorite won, grouped by how far apart the two sides were. Half of all matches had a gap of 0.15 or less.
| Rating gap | Matches | Favorite won (95% range) | Predictor said |
|---|---|---|---|
| 0.00 to 0.05 | 1,279 | 55% (53–58%) | 54% |
| 0.05 to 0.10 | 1,185 | 61% (58–63%) | 60% |
| 0.10 to 0.15 | 1,132 | 69% (66–72%) | 67% |
| 0.15 to 0.20 | 838 | 74% (71–77%) | 73% |
| 0.20 to 0.30 | 1,316 | 82% (79–84%) | 80% |
| 0.30 to 0.40 | 681 | 88% (85–90%) | 88% |
| 0.40 to 0.60 | 551 | 93% (91–95%) | 95% |
| 0.60 or more | 126 | 98% (93–99%) | 99% |
Two things stand out. First, small gaps barely matter: a favorite by less than 0.05 wins about 11 times in 20. Second, the odds climb quickly after that. By a gap of 0.20 the favorite wins about 4 times in 5, and past 0.60 an upset is rare (3 of 126 matches).
Singles and doubles behave about the same. Favorites won 74% of singles matches and 72% of doubles matches.
Can anything beat “higher rating wins”?
The simplest way to predict a match is to pick the higher-rated side. To test fairly, we set up our match predictor on the first 5,398 matches (through August 11) and then scored it, once, on the 1,800 matches played after that. It had never seen them.
| Method | Picked the winner | Score for its chances (lower is better) |
|---|---|---|
| Coin flip | 50% | 0.250 |
| Higher rating wins, stated as certain | 72% | 0.271 |
| Higher rating wins, always “72%” | 72% | 0.199 |
| Our predictor, old settings | 72% | 0.190 |
| Our predictor, new settings | 72% | 0.186 |
The score in the last column (the Brier score) measures how close the stated chances came to what happened. Saying 50% every time scores 0.250. Saying 100% for the favorite scores worse than that, because the favorite loses more than a quarter of the time and being sure and wrong is punished hardest.
For picking winners, nothing beat the higher rating. Our predictor picks the higher-rated side every time, so it was right just as often: 72%. Ratings alone can’t do better than that, because a quarter of favorites lose.
Where the predictor helps is in how sure to be. It tells a 55% toss-up from a 90% lock, and on the 1,800 later matches its chances matched what happened:
| The predictor said | That side actually won |
|---|---|
| About 15% | 13% |
| About 35% | 31% |
| About 50% | 48% |
| About 65% | 67% |
| About 85% | 82% |
So the rating gap alone can’t pick more winners. The next section asks whether anything else can.
What else predicts a win?
For every match we measured what was known before it started and tested each factor on top of the rating gap: recent results, how fast each rating was moving, experience, rest, how often a doubles pair had played together, partner balance, the team’s record, court, level, league and time of season. We looked at each factor two ways: on its own, as in the table below, and in a logistic regression with all of them together, so that each one is measured with the others held equal.
| The side that… | Matches | Won | Rating gap alone said | Difference |
|---|---|---|---|---|
| won 2 or more extra of its last 5 matches | 1,249 | 73% | 63% | +10 |
| came from the team with the better line record (by 25+ points) | 519 | 75% | 67% | +8 |
| had a win streak 3 or more matches longer | 2,360 | 67% | 60% | +7 |
| had risen 0.15 or more further since the season started | 1,258 | 63% | 56% | +7 |
| had risen 0.10 or more further over its last 5 matches | 1,317 | 57% | 52% | +5 |
| had played 2 or more extra matches in the last two weeks | 2,716 | 59% | 54% | +5 |
| was a doubles pair that had played together more | 1,323 | 61% | 56% | +5 |
| jumped 0.08 or more further in its last match | 953 | 51% | 52% | −1 |
| was a lopsided doubles pair (say 4.0 + 3.0 against 3.5 + 3.5) | 855 | 50% | 48% | +2 |
What held up once the factors were tested together:
- Recent wins matter most. Ratings move slowly, so a player who keeps winning is usually better than their rating says. Winning two more of your last five matches than the other side is worth roughly 0.05 to 0.08 rating points. In singles a long win streak counts too.
- Rising players do a little better than their rating, as you’d expect. A rating that has climbed steadily since the start of the season tends to keep climbing.
- One big jump doesn’t. After a match that moved a rating up a lot, about a third of that jump or more is given back the next time out: the rating overreacts to a single great day. So steady climbs count; one-off spikes don’t.
- Strong teams beat their ratings. Players from the team winning more of its lines won more than their ratings predicted, even after their own recent results were counted.
- Doubles chemistry is small. Pairs that had played together more did a bit better, but most of that went away once recent results were counted. A lopsided pair did as well as a balanced one with the same average, so averaging the two ratings works.
- The gap counts more in some places. At gaps of 0.10 to 0.30, favorites won 76% of doubles matches but 72% of singles; 78% at the 3.5 to 4.0 level against 73% below 3.0; and 79 to 80% in 55 & over and tri-level leagues.
- No effect: days of rest, court 1 against the other courts, playoffs, and the time of season.
Did these make predictions better? On the 1,800 later matches, adding recent form to the rating gap made the stated chances more accurate: the score in the table above fell from 0.186 to 0.183, and the ability to rank winners above losers (AUC) rose from 0.79 to 0.80. It did not pick more winners: still 72%. Separate models for singles and doubles did no better than one model that learns from both and lets the gap count a little differently in each. A more flexible machine-learning model (gradient boosting) did no better than the simple one either; there aren’t enough matches yet for the extra flexibility to pay off.
What changed in the match predictor
The dynamic rating formula says that a player rated 0.1 higher should win about 9 percentage points more of the games. In real matches only about 70% of the rating gap showed up on court, because every rating is an estimate with some error in it. The predictor now uses that 70% (it used 60% before, set from just 68 matches) and a little less blur for good and bad days. With the old settings, a 0.15 favorite was given about 65%; the new ones give 70%, close to the 69 to 74% that favorites actually won.
From the expected share of games, the predictor works out every way the points, games, tiebreaks and sets can fall, so it also gives the most likely scores and how often a match goes to a third set (27% of these matches did).
What it means on court
- Don’t read much into a gap under 0.10. Those matches are close to a coin flip, so form, matchups and the day matter more than the number.
- A 0.20 to 0.30 favorite still loses about 1 match in 5. Upsets are normal, not proof that a rating is wrong.
- For lineups, think in chances, not picks. Three courts where you are a 60% favorite win 1.8 courts on average, not 3. Captain mode works from each court’s chance for this reason.
What this can’t tell you
The ratings here are estimates, not the USTA’s own numbers, which are never published. Most matches come from one part of the country, and the same players appear in many matches, so the true ranges are a little wider than the ones shown. Only matches where every player had a rating are included, so brand-new players are left out.
The matches came from the histories of 500 players chosen because they had played the most. Players chosen that way beat their ratings slightly (captains keep playing the people who win), so every factor above was tested with that taken into account, and checked again on the 1,446 matches where both sides had the same share of those players. The main findings held. We will re-run the study as more matches come in.
Common questions
How often does the higher-rated player win in USTA league tennis?
In 7,198 USTA league matches from 2026, the side with the higher estimated dynamic rating going in won 72% of the time: 74% in singles and 72% in doubles.
Is a 0.1 rating difference a big deal?
Not much. With a gap of 0.05 to 0.10 the higher-rated side won 61% of the time, and under 0.05 it won 55%, barely better than a coin flip. The odds get lopsided around 0.3: favorites by 0.30 to 0.40 won 88% of the time.
Does the higher-rated team win in doubles?
Yes, at about the same rate as singles. Comparing the two teams' average ratings, the higher-rated team won 72% of 5,287 doubles matches. Giving the stronger partner extra weight did not predict better than a plain average.
Do players on a hot streak beat their rating?
Yes, a little. The side that had won two or more extra of its last five matches won 73% of the time when the rating gap alone said 63%. Ratings move slowly, so a player who keeps winning is often better than their rating shows. A single big jump in the last match is the exception: about a third of it tends to be given back.
How accurate is the match predictor?
On 1,800 matches it had not seen, it picked the winner 72% of the time, the same as picking the higher-rated side. Its stated chances matched what happened: when it said about 85%, the favorite won 82%; when it said about 50%, 48%.
Got a match coming up? Put in the players on the match predictor to see the win chance and the likely scores, or sync your matches to see your own estimated dynamic rating.