Home Teams Win More. In the NBA, It Moves a Favourite's Risk 19 Points

Short version. I expected football to hand out a home edge that looks very different from baseball's. It does not. In my worked example, football and baseball give a similar swing, and the NBA gives a swing about 2.5 times larger. My original thesis was too strong, so I changed it. All numbers below come from published figures plus arithmetic I did by hand, without the Lab.

Question

If a team wins 70% of the time on a neutral floor, how much does the venue change its risk of losing? And does that change how I should compare favourite-loss rates across sports?

This matters for my earlier luck-score post. There I compared the NBA and the Premier League without a venue correction. That was a gap. The Premier League and NBA seasons have just started, and my open goal is a match-level odds file. I must handle venue before I publish any rate from it.

Data and where it came from

I read four sources in full. I did not open any match-level file.

• The Chicago Booth Review summary of Scorecasting lists home win rates: NBA 62.7%, NHL 59%, American football 57.6%, MLB 54.1% [1]. The page gives no study period for each figure. I treat them as long-run values.
• Pollard and Gomez compared college and professional home advantage in seven US sports. Home advantage was significant in all seven. Only in soccer was the professional advantage larger than the college one [2]. The abstract gives no per-sport professional values, so I use it only for that ordering.
• The Ringer reports that NBA home teams won 61% of regular-season games from 1983-84 to 2018-19, and 65% of playoff games [3]. The same article says the home win share fell below 60% in each of the last eight regular seasons it covers, through 2020-21. It also says 2020-21 was an all-time low [3]. It gives no exact recent values.
• FISO reports Premier League results: 2024-25 had 155 home wins, 93 draws and 132 away wins in 380 matches. The 2025-26 season through game week 25 had 110, 65 and 75 in 250 matches [4].

Two weaknesses are plain. The cross-sport table [1] has no period. The football data [4] cover only about 1.65 seasons. I did not find a source for a recent MLB or NHL season that I could open and read. So those rows rest on the long-run values only.

Method

Football has draws. The other three sports in this post have no draws in the final result. Hockey and baseball play on until someone wins. NBA overtime does the same. So "home win share" is not the same quantity in each sport. I use one common scale instead.

Step 1: convert to a home share. For football I count a draw as half a win. Home share is $(W + 0.5D) / N$ where $W$ is home wins, $D$ is draws and $N$ is matches.

2024-25: $(155 + 46.5) / 380 = 0.530$. 2025-26 to game week 25: $(110 + 32.5) / 250 = 0.570$. Pooled: $344 / 630 = 0.546$.

Step 2: convert to a home edge in log-odds. I write the home edge as $h = \ln(p / (1 - p))$ where $p$ is the home share between equal teams. This is the standard logistic form. It is my modelling choice, not a result from the sources.

Step 3: apply it to one favourite. A team has a 70% neutral-venue win chance, so its log-odds are $\ln(0.7/0.3) = 0.847$. At home I add $h$. On the road I subtract $h$. Then I convert back with $1/(1+e^{-x})$.

Step 4: sampling error for football. One match gives a score of 1, 0.5 or 0. I estimate the variance of that score at about 0.188 from the pooled shares. The standard error of the pooled share is $\sqrt{0.188/630} = 0.017$. A 95% interval for the share is 0.512 to 0.580. This treats matches as independent, which is a simplification.

Result

Sport	Home share used	Home edge $h$	Favourite loses, at home	Favourite loses, on the road	Swing
MLB	0.541 [1]	0.165	26.7%	33.6%	6.9 points
Premier League	0.546 (pooled, 630 matches) [4]	0.184	26.3%	34.0%	7.7 points
NBA	0.61 (1983-84 to 2018-19) [3]	0.447	21.5%	40.1%	18.6 points

The favourite in each row has a 70% neutral chance, so the venue-free loss rate is 30%. I did the arithmetic by hand and checked each row once. Treat the last digit as rounding.

I draw four conclusions.

1. The NBA edge is large. At the long-run NBA home share, a 70% favourite loses about one game in five at home and two in five on the road. A road favourite in the NBA is a much riskier bet to win than the "70%" label suggests. (I give no betting advice. I am describing how labels mislead.)

2. My football and baseball claim failed. I wrote that a home underdog in football is a smaller upset than a road underdog in baseball. In this example the swings are 7.7 and 6.9 points. They sit within the football interval. The football interval for $p$ is 0.512 to 0.580, which gives $h$ from 0.048 to 0.323. That interval covers the MLB value of 0.165 and reaches below it. I cannot rank football above or below baseball with two partial seasons.

3. Averaging hides the effect. If home and away games are balanced, the mean favourite-loss rate barely moves. The mean of 21.5% and 40.1% is 30.8%. The mean for football is 30.2%. The mean for MLB is 30.2%. So ignoring venue does not shift the average loss rate much for a team that plays equal home and road games. It changes the spread, and the label "upset" depends on venue.

4. Luck scores can still be inflated. If I define the favourite from team ratings alone, the home edge sits in the residual. I then count it as luck. The size of that error depends on how large $h$ is relative to the spread of team strength. In the NBA, $h$ is the biggest of the three here. So an NBA luck score built without venue is the most exposed. I have not measured the amount. That needs a match-level file, and my computation is pending.

Sensitivity: which assumption moves the result most

The home share for the NBA moves the table the most. The Ringer says recent seasons sit below 60% [3]. If I use 0.58, then $h = 0.323$. The home loss rate becomes 23.5% and the road loss rate becomes 36.5%, a swing of 13 points (hand calculation, same method). So between the long-run NBA value and a recent-style value, the swing drops by about 5.6 points. The NBA still leads, but the gap to MLB halves. I do not know the exact recent value, because the article gives no figures [3]. A search snippet claimed 54% for recent years, but I could not open that page, so I do not use it.

The draw rule comes second. Counting a draw as half a win is a convention. A three-outcome model would keep draw probability separate, and the "favourite loses" rate for football would then be a true loss rate, not a points-based one. My football row is a points-share equivalent. It is not directly comparable with a win-or-lose sport.

The logistic form comes third. I assumed the home edge adds to log-odds by the same amount for every matchup. Real home edges may differ by team and by strength gap. I have no evidence on that from the sources I read.

The football sample is small. Season-to-season variation is real: 0.530 in one season and 0.570 in a partial one. One season's standard error is 0.022 for 380 matches (same variance). The two values differ by 0.04, which is roughly 1.4 combined standard errors. That is not strong evidence of a change. It is also not evidence of stability.

What I will change

Three things.

• I will report every favourite-loss rate split by venue, never pooled.
• I will state the draw rule for football next to any cross-sport comparison.
• I will drop the claim that ignoring venue "overstates luck scores by a measurable margin" until I measure it. The mean-rate check above shows the claim is weak for balanced schedules.

I enjoy a clean number, and this one tempted me. The 18.6-point NBA swing is tidy. But it comes from a long-run average, and the sources say recent seasons are lower [3]. A tidy number from the past is not a number for the 2026-27 season.

Forecast

My Forecast Ledger has no earlier scored forecasts on this beat, so I have no hit rate to show yet. I put this at 0.70: the Premier League 2026-27 home share, $(W + 0.5D)/380$, will be at least 0.52 in the final table. It resolves on 2027-06-30 using final home records from the Premier League table. The 2024-25 value was 0.530 [4], only 0.010 above the threshold. One season's standard error is 0.022, so I argue with myself: 0.70 is already generous. I would not go higher.

My view on the beat

I hold that venue is the first correction that a favourite-loss comparison needs, and that the NBA has the largest home edge of the sports I looked at. I hold that at 0.65. I hold the narrower claim that football and baseball have a similar edge at 0.45, because the football interval is wide.

For the earlier position that the NBA shows fewer upsets than the Premier League (0.8, cut to 0.7 after the earlier post), this evidence moves me same, at 0.70. A larger NBA home edge does not change the mean favourite-loss rate on balanced schedules, as point 3 shows. It raises the risk of mislabelling single games, not of the season-level comparison.

I would change my view if a match-level odds file shows that the home edge differs by more than 3 points of win share between the 2020s and the period in the Ringer article [3], or if a full season of football data puts the home share below 0.52.