The Conference Tax: The West Was a Home Court Better, and It Cost a Record One Win

Every autumn somebody points at the standings and says the West is a harder place to make a living, and every autumn the claim goes unpriced. Here is the price for 2023-24. The two conferences played 452 games against each other and the West won 261–191, a .577 winning percentage and a margin of +1.95 points a game. Eastern teams hosting Western ones went 105–121; in cross-conference play the East's home court did not exist. Fit a rating to all 1,230 games with home court taken out and the gap between the conferences is 1.96 points, which is 92% of a home court. But the gap is not where the argument usually puts it. Lay the two conferences side by side, rank for rank, and the best team in the league was Eastern, the second- through eighth-best chairs differed by a point and a half, and the entire case lived in ranks nine through eleven, where the West was better by 3.9 points a game. Priced in wins, being a Western team cost a record about one win. That is the conference tax: real, about a home court's worth of quality, and smaller than the conversation about it.

452 games, not 450

The format has every team play the 15 clubs in the other conference twice, once in each building, which should make 450 cross-conference games. The log holds 452. Two pairs met three times — the Bulls and Spurs (Chicago won the extra one 121–112 in San Antonio on December 8) and the Pistons and Grizzlies (Memphis won 116–102 in Detroit on December 6). Both dates fall in the In-Season Tournament's knockout week, when clubs eliminated in group play were handed extra regular-season games to fill out their 82, so those four teams finished with 31 cross-conference games rather than 30. I have kept them; they count in the standings. The tournament final itself, Pacers at Lakers 123–109, counts in no team's 82 and is excluded, as it is in every piece on this site. That leaves 1,230 games: 778 inside a conference, 452 across, and each conference hosted exactly 226 of the 452, so home court cancels out of the raw ledger by construction.

The hand-typed list of 15 Eastern teams is the one thing here not read from a file, and the script verifies it against the schedule itself: all 225 East-West pairings occur, 223 of them exactly twice, and inside a conference every pair met three, four or five times.

The raw ledger

West 261–191. On the win count alone that is 3.3 standard errors from a coin flip; on margin it is +1.95 a game with a standard error of 0.73, t = 2.69, and a 95% interval from about half a point to 3.4. No number in this piece is more secure than that one, and none of the rest should be read as more precise than that interval allows.

The venue split is the part I did not expect. Western hosts went 140–86 and won by 4.05 a game, the league's ordinary home edge with the conference gap stacked on top. Eastern hosts went 105–121 and won by 0.15 a game: level, on average, at home. Home court did not vanish in these games — the home team's margin across conferences was +2.10, league-wide +2.15 — it was simply spent. An Eastern team hosting a Western one started with a home court and ended roughly where a neutral-site game starts.

The right panel spreads the same 452 games across the 30 teams. Denver's 24–6 was the best cross-conference record; Boston's +13.87 a game on 23–7 the best margin. Eleven of 15 Western teams outscored the East; six of 15 Eastern teams outscored the West. The bottom of the panel is entirely teal: Toronto 7–23 at −11.73, Washington 4–26, Detroit 4–27, Charlotte 7–23, all four beaten by nine or more a game, while no Western team lost to the East by more than Portland's 5.17. The East had a basement. The West had a floor.

A rating with the home court taken out

A raw ledger of 452 games is one estimate of the gap. A second uses all 1,230, giving every team a rating r such that a game's expected home margin is r_home − r_away + h, with the 30 ratings constrained to sum to zero and h the home-court term. That is ordinary least squares — the linear-model approach David Harville laid out for NFL games in the Journal of the American Statistical Association in 1980, and the same family as Basketball-Reference's SRS, which adjusts margin for schedule. I fit it from scratch so every number traces back to the game log.

The fit puts home court at +2.13 (SE 0.39) and the residual standard deviation at 13.84 points, the usual reminder that most of a single NBA game is not explained by anything. Boston tops the ratings at +10.74, Charlotte closes them at −10.12, Oklahoma City leads the West at +7.36. They correlate at 0.998 with the bundled net ratings, differing by 0.36 points on average.

The West's mean rating is +0.98, the East's −0.98: a gap of 1.96 points, standard error 0.65, t = 3.02, with the raw ledger's +1.95 sitting on top of it by a different route. Against the model's own home court of 2.13, the conference gap is 92% of a home court. Drawing an average Eastern opponent instead of an average Western one was worth nearly what playing at home was worth.

Notice what the raw version of that number does. Average the bundled net ratings by conference and the gap reads only 1.42, the West at +0.71 and the East at −0.71, because 52 of every team's 82 games are inside its own conference. Each conference accumulates its margins mostly against itself, so raw margin flatters the East, and the adjusted gap is 1.38 times the raw one. It is the mirror image of the self-exclusion effect in the strength-of-schedule piece: a great team's schedule looks soft because it never plays itself; a conference's numbers look better than they are because it mostly does.

Where the gap lives

Sort each conference's 15 ratings and pair them rank for rank, and the left panel is what you get. Rank one goes to the East, and not narrowly: Boston +10.74 against Oklahoma City +7.36, a 3.4-point edge. Ranks two through eight go to the West by an average of 1.6 points — Minnesota over New York, Denver over Indiana, down to Dallas over Miami in the eighth chair, +2.30 against +1.10. Those two are the conference medians, and they say the typical Western team was 1.2 points better than the typical Eastern one.

Then ranks nine, ten and eleven. Sacramento +2.29 against Chicago −1.77: 4.06 points. Houston +1.24 against Atlanta −2.38: 3.62. The Lakers +1.02 against Brooklyn −3.02: 4.04. The average gap across the three play-in chairs is 3.9 points a game, more than double the gap across the seven chairs above them. Eleven Western teams rated above zero and eight Eastern teams did, and the same eleven and eight finished at .500 or better. Sacramento, the West's ninth-best team by rating, would have been the East's sixth. Ranks twelve through fifteen settle back to a 2.5-point gap: the East's basement was deeper, but both were basements.

So the headline claim is true and badly located. At the top the East had the best team in basketball; across the playoff seeds the West was a point and a half better, real and unremarkable. The gap that generates the argument every year sat in the three chairs where a season is decided by a play-in game, and there it was the size of two home courts. Drop each conference's best team and the mean gap widens from 1.96 to 2.35; the East's strength was one team deep. I think this is also why the East's top eight froze on December 16 while the West's last chair changed hands until April 12: a top eight freezes when the ninth-best team is four points worse than the eighth, and stays open when the two are a coin flip.

Pricing it: Golden State and Miami, both 46–36

Golden State finished 46–36, tenth in the West, and lost its play-in game. Miami finished 46–36, eighth in the East, and won through the play-in. Same record, different Aprils. How much of that was the conference?

The model rates Golden State +2.77 and Miami +1.10, so the Warriors were the better team by 1.67 points before any schedule is considered. Now give each a synthetic schedule: 52 games against the average of one conference and 30 against the average of the other, half at home and half away, with each game's win probability P = Φ((r_team − r_opp ± 2.13) / 13.84), the model's own residual scale. Golden State's Eastern pool averages −0.98; its Western pool, with itself removed, +0.85. Venue-averaged, the Warriors beat an average Eastern team 59.8% of the time and an average Western team 55.0%.

As a Western team: 52 × 0.5497 + 30 × 0.5977 = 28.58 + 17.93 = 46.51 expected wins. As an Eastern team: 52 × 0.5977 + 30 × 0.5497 = 31.08 + 16.49 = 47.57. The conference cost Golden State 1.06 wins. Run the model over the Warriors' actual 82 opponents and venues and it expects 46.58; they won 46. Miami's version: 55.8% against the East, 50.1% against the West — a coin flip — for 44.08 expected wins as an Eastern team and 42.82 as a Western one, a tax of 1.25; over its real schedule the model expected 44.73, and the Heat won 46.

Put the two together. Golden State was the better team by 1.67 points, its conference cost it about a win, and Miami's record ran a win or so ahead of its margin, the kind of gap the Pythagorean piece treats as noise. Swap conferences and the arithmetic says roughly 47–48 wins for Golden State in the East, fifth place among the three 47-win teams, and about 43 for Miami in the West, eleventh. Whether the Warriors' April would have gone differently is not a question a game log can answer. That there was about one win of conference in the difference between the two seasons is.

The tax barely varies: across all 30 teams it averages 1.11 wins, from 0.67 for Charlotte to 1.46 for Portland, smaller at the extremes because a team that wins a quarter or three quarters of its games sits on the flat part of the curve. A linear shortcut lands in the same place: the swap changes the opponent pool in 22 of 82 games (52 minus 30) by 1.96 points, 22 × 1.96 / 82 = 0.53 points of season-average margin, and the file's own wins-on-margin line, 41.000 + 2.3678 × margin — the line the points-to-wins piece rests on, recomputed here — makes that 1.25 wins. Curve or line, one win, give or take a quarter.

What this doesn't prove

One season. The sign flips in years the East is stronger, and the size — 1.96 points, with a model interval of roughly 0.7 to 3.2 and a raw-ledger interval of 0.5 to 3.4 — is not a constant of the sport. Nothing here says the West is structurally superior.

The ratings are end-of-season and include the games they judge. Every team's rating is fit on all 82 games, the 30 cross-conference ones included, so the model's gap and the raw ledger's gap are not independent estimates; the raw ledger is the cleaner number and the model adds a schedule adjustment, not a second witness. Rosters also move, and a single rating averages over that. The season's arc says as much: the West went 69–55 across conferences in 2023 at +0.52 a game and 192–136 from January at +2.49, the East actually leading November 27–25 — the late-season bottom-drop the margin-creep piece found — but the halves differ by 1.3 standard errors, so that is how the season felt, not something the data established.

The tax is a schedule-composition counterfactual and nothing more. Swapping a team's conference swaps 22 games' worth of opponent pool. It does not swap travel, time zones, rest, or division rivals; the rest ledger found that side of the schedule worth 0.73 wins across the whole league, so the omission is small, but it is an omission.

Seeds are not wins. The tax says the conference cost Golden State about a win. It does not say the Warriors would have been a fifth seed in the East; seeding depends on 14 other records, tiebreakers and a play-in game's single-night variance, none of which this piece models. The claim is about the record, not the bracket.

Reproduce it

chart_conference_tax.py reads the two bundled files, verifies the conference map against the schedule's structure, counts the 452 games, fits the 31-parameter least-squares model with a sum-to-zero constraint, pairs the conferences by rank, and prices the swap for every team. It pins each number above with an assert — 150 of them — and draws both panels. The regression is one numpy.linalg.lstsq call on a 1,231-row design matrix (1,230 games plus the constraint row); the pricing is a normal CDF. Nothing needs a network connection.

What I take from this is a narrower complaint than the one usually made. The West-is-harder argument is right, it is about a home court's worth of quality, and it was worth about one win to a record. The louder version — that the standings could not be compared across conferences — fails at the top and across the playoff seeds. It holds in exactly one place: the three play-in chairs, where an Eastern .500 team was a Western lottery team. If you want the conference gap to matter, look there and nowhere else.