How Often Does the Better Team Win in the NBA?

Basketball is supposed to be the sport where the better team almost always wins — no flukey bounces, no goalie standing on his head, just talent grinding out over 48 minutes. The 2023-24 data mostly agrees, with one big asterisk: the better team won 67.9% of games across the season, but that headline hides a steep gradient. In a toss-up, the favorite barely cleared a coin flip (55%); in a true mismatch, it won four times out of five (81%). How deterministic the NBA is depends entirely on how far apart the two teams are.

How we defined "the better team"

For every game we labeled the “better team” as the one with the higher season net rating — points scored minus allowed per 100 possessions, the cleanest one-number measure of team quality. Then we took the gap between the two teams’ net ratings, bucketed every game by how big that gap was, and asked: how often did the higher-rated team actually win? That turns a vague question (“does the better team win?”) into a measurable one (“the better team wins X% of the time when it’s this much better”).

The exhibit: predictability rises with the gap

The staircase is the whole story. When two teams are within 2 points of net rating — genuinely matched — the “better” team won just 55%, barely better than chance. A slight edge (2–5) bumped it to 58%. A clear edge (5–8) jumped to 71%. And in mismatches of 8-plus points per 100 — a contender hosting a tanker — the favorite won 81% of the time. The NBA is highly predictable, but only when the teams are actually unequal; among peers, it’s nearly a coin flip.

Why 68% is high — and why it isn't higher

Compare that 67.9% to other sports and the NBA’s reputation holds up. Favorites win far more often in basketball than in, say, baseball or soccer, where a single game is dominated by variance and the better team might win 55–60% at most. Five-on-five over 100-plus possessions gives talent a lot of room to express itself; the law of large numbers works inside a single game.

So why doesn’t the better team win 80%+ across the board? Because most NBA games aren’t mismatches. The leaguewide 68% is a blend dominated by the many close-matchup games sitting near a coin flip, dragged up by the minority of blow(out)-prone mismatches. The single number is real but misleading on its own — it’s the average of a 55% world and an 81% world, and which one you’re watching depends on the marquee.

A worked example: reading a single matchup

Say the Celtics (net rating around +11 in 2023-24) host a roughly average team (net rating near 0). That’s an 11-point gap — deep into the mismatch bucket — so the model says Boston wins about 81% of the time, before you even add home court. Now put two +3 and +1 teams together: a 2-point gap, the toss-up bucket, and the “better” team is only about 55% to win. Same league, same night, wildly different certainty. This is why “the better team always wins in the NBA” is true for the Celtics-versus-cellar game and basically false for two playoff-caliber teams trading blows.

A second yardstick: season record, with the judged game left out

Net rating is one way to name the better team. A stricter one uses each team’s season win% excluding the game being judged, so no game gets to vote on its own favorite. On that yardstick, across the 1,230 regular-season games of 2023-24 (the In-Season Tournament final is left out so every team sits at exactly 82), the better team won 70.8% of the 933 games where the two records were at least five points of win% apart. The two headline numbers, 67.9% and 70.8%, are not in conflict: the first labels every game by net rating, the second labels by out-of-game record and starts at a five-point gap.

Read the trend, not the teeth: adjacent buckets sit within a standard error or two of each other, but the staircase from the mid-fifties to ninety is the same shape the net-rating chart shows. Even in the 281 games with a gap of thirty points or more, the worse team still won about one game in six.

Why the first row reads 41.8%. One game is worth 1/81 ≈ 1.23 points of win%, and leaving the judged game out moves the two teams in opposite directions: the eventual winner’s remaining record drops and the loser’s rises, a combined swing of about 2.5 points. When the true gap is smaller than that, the exclusion itself decides who wears the favorite label, and it hands the label to the team that lost. The below-coin-flip “favorite” is manufactured by the measurement, which is why every conclusion on this yardstick starts at the five-point line. One more rule shapes that row: in the 40 games where the two out-of-game records are exactly level, the home team is labelled the favorite.

The upset ledger. Tied at the top, at a 54.3-point gap: the 21-win Hornets beating the 64-win Celtics 121-118 in November, and the 14-win Pistons handling the 57-win Thunder 120-104 in January. San Antonio appears three times in the top eight. Two of the eight came on April 12, the season’s final weekend, when contenders rest rotations, so some upsets are scheduling artifacts; the November Hornets game carries no such asterisk.

The favorite travels at a discount. Split the favorites by venue (the 40 level-record games counted as home favorites, as above) and the better team won 67.6% at home and 59.8% on the road, about eight points of win rate for the building, in line with the season-level home-advantage census. A mid-tier favorite on the road is barely a favorite at all.

Where this read has limits

• It’s in-sample. We used full-season net ratings to label games within that same season — the ratings already “know” how the games turned out. The out-of-game record yardstick above removes the judged game from its own label, at the cost of the small-gap distortion it explains. A true pre-game forecast (using only prior information) would be a little less accurate, so treat these as upper-ish bounds on predictability, not live betting numbers.
• Net rating isn’t destiny. A season-long average smooths over injuries, trades, rest, and schedule fatigue — the team that was +11 over the year might be +3 tonight with two starters out. The gap is a strong proxy, not the exact matchup truth.
• No home court in the buckets. We labeled the favorite purely by rating, ignoring venue. Home advantage would lift the home favorite’s number and cut the road favorite’s; folding it in is the natural next step (see home-court advantage).
• One season. 1,200-odd games is plenty to see the gradient, but the exact percentages will wobble year to year; the shape — rising with the gap — is the robust part.

The takeaway

“Does the better team win?” has a satisfying NBA answer: usually — 68% of the time — and much more than usually when the teams are far apart. But the honest version replaces the single number with a gradient. The favorite is a near coin flip among equals and a near lock in a mismatch, and almost every interesting game in the league lives somewhere on that slope. Predictability isn’t a fixed property of the sport; it’s a function of how unequal tonight’s two teams happen to be.

Reproduce it

Join each game in data_layer/nba_home_results.csv to the two teams’ net ratings in data_layer/nba_ratings.csv, label the higher-rated team the favorite, bucket by the absolute rating gap, and compute the favorite’s win rate in each bucket. The chart is regenerated by chart_better_team_win_rate.py — no network, nothing hand-entered.