Every pickleball player I know believes in the bad side. You lose the coin flip, you squint into the sun for a game, the ball does something stupid in the crosswind, and you tell yourself the court owes you one. It is the most universally held belief in the sport that nobody has ever checked.
So I checked. I attached the weather at the hour of play to every professional game in my database, and then I went looking for the effect six different ways. This is the write-up of not finding it — which took considerably more work than finding it would have.
The reason nobody has run this analysis is that the ingredients live in three different places and none of them are labelled "weather."
Results feeds tell you who played and what the score was. They don't tell you that the 2026 Boise Challenger happened at The Flying Pickle at 43.615° north, 116.364° west — but buried in the tournament record, it turns out they do, in a latitude and longitude field nobody has any reason to look at. Same for the team leagues, where each MLP stop carries a full venue object with coordinates and a time zone. That gets you 110 events pinned to a map.
From there, a historical weather archive will hand you the conditions at any point on earth for any hour since 1940, free. So now every event has a weather diary.
The last piece is the one that matters most, and I nearly skipped it. A tournament day has a daily maximum wind — but a 25 mph gust at 4pm tells you nothing about the match that finished at 9am in dead calm. Day-level weather is mostly noise smeared over real conditions, and it will bury a true effect. So I swept the start time of every individual match: 20,405 of them, 13,248 with the referee's actual clocked start rather than the schedule. Now the wind attached to a game is the wind that was blowing while they played it.
Start with the most basic physical question. In side-out scoring, only the serving team can score. If wind makes the ball harder to control, the balance between serving and returning should move — serves and drives sail long, drops die short, and something in that ledger should show up.
It doesn't. Across 7,900 outdoor matches with full rally logs, the share of rallies won by the serving team is 0.4483 in calm conditions and 0.4511 at 14–20 mph. The fitted slope is +0.003 per 10 mph, with a confidence interval that comfortably straddles zero. The indoor control is equally flat, which is what it should be.
Whatever wind does to a pickleball in flight, it appears to do it symmetrically to both teams.
This is the one everybody actually wants answered, and there's an obvious problem: the data never says which end of the court anyone is standing on. Referee logs record which team served, not where they were. There is no column for "sunny side."
But the ends aren't assigned randomly and forgotten — they're governed by rules. Teams switch ends between games. In deciding games they switch again mid-game, when the first team reaches 6. MLP switches at 6 in every game. That schedule is a lever: if one end is better, then a team's performance before the switch and after the switch are measurements of two different places, taken minutes apart, by the same four people.
Which gives you the test. Take a team's share of points before the switch and their share after — 6–0 and then 5–7, say. If the court is fair, those two numbers differ only by luck. If one end is worse, the swings should be systematically bigger than luck allows.
One subtlety, and it's the crux of the whole design. You can't just average the swings and see if they're positive, because nobody records which end is the good one — the team that starts on the good side is effectively picked by coin flip, so the average swing is zero whether the effect exists or not. The signal isn't in the average. It's in the size. Bigger-than-chance swings, and bigger still on windy days, are the fingerprint.
Here's where it gets interesting. Real swings are bigger than chance — about 70% bigger. That looks like a smoking gun until you notice that side-out scoring bunches points into serving streaks: a team that wins a rally serves again, so points arrive in clumps, and clumpy points make lumpy halves even with a perfectly fair court and perfectly steady players. Simulate games with the correct serve mechanics and nothing else — no momentum, no ends, each rally an independent coin flip — and you reproduce almost exactly the excess seen in real games.
| Conditions at match time | Games | Swing size | Expected from serve clumping alone |
|---|---|---|---|
| Indoors | 2,383 | 1.67 | 1.76 |
| Outdoor, calm (under 8 mph) | 1,636 | 1.72 | 1.78 |
| Outdoor, breezy (8–14 mph) | 656 | 1.76 | 1.79 |
| Outdoor, windy (14+ mph) | 111 | 1.95 | 1.79 |
Indoors and calm outdoor games land at or slightly below what pure serve-clumping predicts. There is no room left over for a bad side — and, as a bonus finding nobody asked for, none left over for momentum either. The lumpiness of a pickleball game, the runs and the collapses, is fully explained by the fact that you keep serving when you win. Whatever teams feel out there, it doesn't leave a trace beyond the scoring mechanics. The windy row pokes above its own baseline — and it's the last honest hope in this entire piece — but it rests on 111 games and its uncertainty band swallows the gap whole.
The strongest way to state it: compare the groups directly to each other, no simulation involved. Windy games swing 0.22 more than calm ones, give or take 0.38. Indoors and calm outdoor are separated by 0.06, which is nothing at all.
A null result deserves more suspicion than a positive one, because the boring explanation for finding nothing is that you looked badly. So the same question went through three more designs, each attacking a different weakness.
Between games, using scores alone. Teams change ends between game one and game two, and the scores are the point counts — 11–6 is a share of 17 points. No rally logs required, which means all 14,000 matches instead of 5,000. Observed swings: 1.81 to 1.90 across weather groups, against simulated no-effect baselines of 1.84 to 1.88. Flat.
Between games, using margins. Same switch, different statistic: the difference between a team's game-one and game-two point margin, which cancels their skill exactly. Also flat, though this design is the weakest of the four — game-to-game noise is large enough that it could only see an end advantage worth more than a point a game.
At the level of individual rallies. The best of the four. Points clump; individual serve rallies mostly don't, so the whole clumping nuisance collapses and the baseline falls from 1.8 to about 1.15 — much less machinery between me and the answer. It also doubles the sample, because at rally level each team's serve record is separate information rather than the mirror of its opponent's. Verdict: indoors 1.10, calm 1.15, breezy 1.09, all sitting on baselines of 1.15–1.16. The windy bin runs hot again at 1.34, and the continuous wind slope is flat as a board — the elevation shows up only in the extreme bin and never as a dose-response, which is what noise looks like and not what physics looks like.
Forget the ends. Maybe wind is just an equalizer — the ball misbehaves, the better team's edges shrink, upsets go up. This is the version of the theory most people actually believe, and for about an hour I believed I'd found it.
Sorting games into wind buckets and comparing how often the favorite won against how often my model expected them to, the pattern looked clean and it looked real:
| Wind at match time | Games | Favorite beat expectation by |
|---|---|---|
| Calm (under 8 mph) | 17,139 | −4.0 pts |
| 8–14 mph | 6,447 | −4.0 pts |
| 14–20 mph | 1,151 | −6.0 pts |
Favorites underperform by four points in a percentage sense normally, six points when it's blowing. Over a thousand games in that windy bucket. That's a story.
It's also wrong, and two things killed it.
The first is that binning throws away most of the information and invites composition effects — different bins contain different tournaments, rounds, and mixes of players. Test the claim the way it should be tested, as a continuous interaction between skill and wind across all 24,819 outdoor games, and the answer is +0.002, a zero so clean it's almost funny. A favorite converts skill into points at precisely the same rate at 15 mph as in dead air. Run it again at the level of individual rallies, 98,000 of them, as a logistic model of whether the serving team wins each one: also zero.
The second is the one that settles it. I ran the identical test on indoor games, where there is no wind — and the indoor group showed a larger apparent effect than the outdoor group, in both the game-level and the rally-level version. An effect that's stronger where the cause cannot reach is not the effect. It's whatever else moves with the calendar: which venues host which rounds, which events my venue labels get slightly wrong, which weeks the tour visits Florida.
The last hypothesis is the most fun, and it comes from Formula One. Rain doesn't just add randomness to a Grand Prix — it reorders the grid, because car control in low grip is a genuinely separate skill from ordinary fast driving. Verstappen in the wet is a different and better driver relative to the field.
Is there a wind equivalent? Players who quietly beat their rating when the flags are stiff, every time, year after year?
For each of 552 players with at least 40 outdoor games, I measured how their results move with wind. Then — and this is the part that matters — I split each player's games randomly in half and measured them twice, independently. A real skill shows up in both halves for the same people. Noise doesn't. The across-player correlation between the two halves came out at +0.06, sitting inside a null band of −0.07 to +0.12. For calibration, the same test on the same data certified clutch play and durability as faint-but-real at 0.15 and 0.13. Wind skill doesn't reach that bar.
But that test has a blind spot worth naming, because it's the exact objection the F1 analogy raises. A split-half correlation detects a dimension — variation spread across the whole field. If the truth is one Verstappen and 551 ordinary players, he's a single point among hundreds and the correlation washes him out. Absence of a dimension is not absence of a genius.
So, separately: is there a standout? I scored every player's wind effect against its own error bar and looked at the extremes, comparing them to what you'd get by shuffling the weather. The most extreme player in the database reaches 3.27 standard errors — and 58% of randomly shuffled datasets produce someone more extreme. Players clearing two standard errors: 29 observed, 27 expected by chance. The tail is exactly as crowded as luck makes it. No Verstappen.
If anyone should own the wind, it's the defensive players — the resetters and retrievers whose whole game is surviving chaos. I named ten of them in advance, before looking: Tellez, Tardio, Patriquin, Hewett, Staksrud, Frazier, Callie Smith, Parenteau, Jorja Johnson, Irvine.
As a group, across 5,134 games, they get worse in wind, not better. Individually the picture is split — Staksrud and Callie Smith lean positive, Parenteau and Irvine anchor the other end — and the group result leans on players who share partners and games, so I wouldn't put a ribbon on the reversal. But the original theory is dead in the direction it was pointed.
There's a plausible reason it ran backwards, and I like it better than the theory it replaced. Wind punishes touch, not defense. Dinks, resets and third-shot drops are slow, precise, high-arc shots — a hollow plastic ball that weighs under an ounce, moving slowly, is exactly what a crosswind gets to push around. Drives and speedups spend less time in the air being argued with. If that's right, wind doesn't reward the defensive player; it taxes the finesse player, whichever side of the ball they're on.
Consistent with that: Anna Leigh Waters, the most offensively lethal player in the sport, sits at the wind-negative edge of her own noise band over 726 outdoor games — the most dominant player alive gets no help from chaos, which stands to reason when you already win most of what you play. And Jack Sock, who arrived from professional tennis with one of the heaviest forehands of his generation, is among the most wind-positive big names in the database at 263 games. Neither is statistically significant. Both are the sort of thing you'd want to see if the touch theory were true.
Six hypotheses, a dozen ways of testing them, roughly 37,000 games and 193,000 rallies — and the sport declines to react. No serve effect. No bad side. No momentum. No upset multiplier. No wind specialists, in the aggregate or in the tail.
Here is what I think that actually means, and it isn't "conditions don't matter." It's that the sport's rules already handle it. Ends switch between games and again mid-game precisely so that no one eats a bad side for long. Everyone on tour has played outdoors their whole life; anyone who couldn't handle a breeze was filtered out several rungs below the pro level. The equalizing mechanisms work well enough that 37,000 games can't find their failure.
And what I can't say is that the effects are zero — only that they're small. Each test has a floor: the per-player tests can't see an individual edge under about 3 percentage points per 10 mph; the between-games margin design can't see an end advantage under about a point a game. A real, tiny bad side could be hiding under all of it, doing nothing anyone would ever notice.
The one door still ajar: high wind. Every design flagged the 14+ mph bin as slightly hot, and every design had too few games there to say anything — 111 in the sharpest test. That's not a finding, it's a to-do. Windy tournaments accumulate on their own, and the scripts re-run in under a minute. Ask me again after a stormy season.
Until then: you lost that game because they played better. The court is innocent.
Every professional MLP and PPA doubles game since January 2024, scraped from the tour's own public results feeds — about 37,000 games across 110 events. Skill ratings come from PICKLES, my Bayesian rating model, which learns each player's per-point strength jointly across all three seasons. Rally-by-rally data comes from referee logs, which cover roughly 31,000 matches. Weather is the ERA5 reanalysis archive by way of Open-Meteo, joined at the venue's coordinates and the match's local start hour.
Two of the tables above compare observed swings to a simulation. That simulation exists to answer one narrow question — "how lumpy would halves be if nothing were going on?" — because there's no way to answer it from within the data. It's built from each match's own skill gap plus the league's measured serve-rally rate, and it contains no momentum and no ends by construction. Nothing was tuned to make it match. Every weather comparison in this piece, though, is data against data: group versus group, or a continuous slope. That's deliberate. If a claim needed the simulation to survive, I wouldn't make it.
Indoor and outdoor labels are educated guesses. I classify venues by tour convention plus name keywords, and a few clubs in the database have both indoor and outdoor courts. Mislabelled venues would blur the contrast between the groups — pushing results toward the null. This is the weakest link in the whole analysis and the first thing I'd fix with better information.
Weather is measured at the venue, not on the court. A reanalysis grid gives wind at a location, not the swirl inside a stadium or the shelter of a windbreak. Real on-court conditions are noisier than my numbers, which again biases toward finding nothing.
Ratings are current-form, applied backwards. Fine for asking whether an effect interacts with weather, which is all I do with them. I wouldn't quote the raw favorite-accuracy levels from this setup as a headline model number.
The leaderboards are noise. There is a list in my working files of the ten most wind-positive players in the database. Every one of them sits at the minimum sample size, which is the signature of a ranking built from randomness. I'm not publishing it, and neither should anyone else.
Since I found the touch-versus-power idea by rummaging rather than by predicting, it doesn't count for anything yet. So, on the record and in advance: I expect finesse-first players to underperform in high wind relative to power players, and I expect the 14+ mph bins to stay slightly hot as they fill out. Both are checkable against games that haven't been played yet, which is the only kind of checking worth much.