Do Groundstrokes Break Down?

The human brain is a trend-spotting machine. A couple of groundstroke errors in a row, and fans and commentators conclude that a player’s stroke has gone off the rails, perhaps because her opponent has exerted just the right pressure. Shots, we say, can “break down,” and even that opponents can “break them down.”

Except… an awful lot of the trends we spot are mirages, misguided attempts to make sense of noisy data. Yesterday, we learned that first serves aren’t streaky. For pros, missing one doesn’t make it more likely that you’ll miss the next; women even tend to make slightly more first serves after misses. So, what about groundstrokes?

It’s a trickier phenomenon to pin down, because unlike first serves, forehands (or backhands) are not all created alike. After missing one forehand, you might not see one on the next point or two. Or if you do, it could be at a different angle, a different height, a different court position, and so on. Still–and long-time readers are going to struggle to believe me on this–there is a detectable effect! Really!

Based on women’s matches since 2020 logged by the Match Charting Project, if a player makes a forehand unforced error on one point, she is 0.45 percentage points more likely to make a forehand unforced error on her next point with a forehand in it. It’s a tiny shift, representing about a 2% tick upward from the typical error rate around 20%, but it is real. The results are about the same (slightly weaker, but in the same neighborhood) for backhands. Men are in the same neighborhood, as well.

But wait: If we control for set, establishing our expected error rates for a single set rather than the entire match, the effect shrinks, and for men, it vanishes almost entirely.

This is good news and bad news. There’s little or no short-term influence of one error on the next opportunity to hit that shot. But error rates do drift over the course of the match.

Drift away

Splitting matches into chunks of 20 forehand points, error rates move around quite a bit more than that sub-half-point shift. Without controlling for set, the typical WTAer sees her forehand UFE rate jump around 1.4 percentage points more than would be expected from chance alone. That’s still not detectable within a single match, but it is a meaningful difference in the long run. Over the last 52 weeks, the range in UFE rate between the most error-prone player (Anna Kalinskaya) and the least (Belinda Bencic) is only 9 percentage points. 1.4 points represents a jump of several places at most positions on the leaderboard.

A bunch of subsequent tests reveal some aspects of that drift. For one thing, it’s usually not a single stroke that ebbs and surges. When a player starts making more errors, both wings suffer. There’s a hint in the men’s data that forehand and backhand error rates don’t quite move in lockstep, but the data is inconclusive.

It’s also clear that players seize on their opponents’ sloppiness. You can probably imagine situations where gaffes seem contagious and no one can keep the ball in the court, but typically, one player tightens up his (or her) game while the other one struggles. I don’t know which direction the causation points: Does one player “break down” the other? Does she take advantage of her opponent’s errors and play more conservatively? Anything along those lines is just speculation at this point.

The strongest effects, though, were somewhere I didn’t intend to look. Error rates shift chronologically, both across sets and within them, in predictable ways.

Tightening up

These graphs show the trend in relative unforced error rates by set, for both men and women:

On both wings, players make more errors in the first set. Women improve their forehand error rate by almost a full percentage point between the first and second sets. While there’s not much movement after the second set, the rates stay well below the first-set standard.

Before we speculate about the causes, let me show you another pair of plots, now comparing relative error rates within sets, by pairs of games:

Again, we start high then tack lower. Though it’s not an unequivocal downward trend, the first two games are consistently the most error-prone, and tiebreaks are the least. (Players become more conservative in tiebreaks by just about every possible metric, from serve speed to rally length to errors.)

Combine those two effects, and you have some sloppy opening games. Here’s one more plot, using WTA data, showing each set broken into chunks of 10 forehand points. The top left corner is the beginning of the first set; the bottom right is the end of the third. Red means more errors than expected (the numbers are percentage points), blue means less:

Every set tightens up as it goes along, but my oh my are those first ten points brutal. These shifts account for much of the “drift” within matches. Error rates move for reasons other than chronology, but chronology is a single factor that seems to affect a large number of players.

Sloppy or aggressive?

High error rates are not inherently bad: Jelena Ostapenko has made a career out of winning matches when she loses one of three points with a (wild) miss. The trick is to pair errors with winners, to take the good aggression with the bad.

That’s not what’s happening here! For women, on both wings, winner rates are at their lowest in the first set. Within sets, winner rates are lower in the first two games than they are until 5-5, when tiebreak caution starts to take over.

These numbers strongly point toward a “warm-up” effect at the beginning of the match and a smaller “re-set” effect at the start of each subsequent set. Maybe that’s understandable: No amount of actual warming up can prepare you for the reality of a new opponent’s individual mix of speed, spin, and angles. And despite the dark red of that upper-left-hand heatmap cell, we’re still only talking about a shift of one percentage point: Enough to pick up across 4,000 charted matches, but far too small to notice even in a fortnight of attentive tennis-watching.

What about the players?

Up to this point, we’ve been talking about averages. We don’t have enough data to identify tendencies for most players, but we can find some.

One of the strongest individual-player findings for women is that Coco Gauff’s error rates drift quite a bit within a match: about three percentage points compared to less than two for the average player. Gauff is slightly more prone than average to a sloppy start, as well, but the full-match metric excludes the first 20 baseline points, so her variance goes far beyond that.

By contrast, Aryna Sabalenka and Iga Swiatek are remarkably steady, showing effectively no in-match variation in error rates beyond what would be expected from chance.

Sabalenka also stands out in that her aggression (winners plus errors) goes up under pressure, defined as the ninth game of a set or later, with the score within one game. The effect is small–a bit more than one percentage point–but it is striking because the average trend under pressure is negative. Gauff, Elena Rybakina, and Jessica Pegula all score about one percentage point in the opposite direction, more conservative under pressure than the average player. Iga is almost precisely average.

I haven’t said much in this post about men, because the trends are generally the same across both tours. Still, there are some interesting individual numbers for ATPers.

Both Stefanos Tsitsipas and Grigor Dimitrov see even more error-rate drift within a match than Gauff. Lorenzo Musetti is the slowest starter, committing the most errors (relative to match averages) in his first 20 chances per match of all men for whom I have a lot of data. It’s suggestive, though hardly conclusive, that those three outliers all have one-handed backhands. I can sympathize: If I play a match on Monday morning, my backhand starts rounding into form on Wednesday afternoon. (Tsitsipas and Dimitrov, for what it’s worth, aren’t slow starters; their error-rate drift comes later.)

A slow start isn’t necessarily bad. Second on that list behind Musetti is Carlos Alcaraz, even though Carlitos scores about average for general within-match drift.

Jannik Sinner’s career of bloodless executions turns up in the data as well. His within-match drift is the lowest (tied with Felix Auger-Aliassime, surprisingly enough) of all players with enough data. In the big moments, he grinds you down, committing fewer errors under pressure and improving his winner-errors ratio. He isn’t the most conservative in the big moments, though: That honor belongs to Novak Djokovic, with the perpetually cautious Alexander Zverev right behind him.

Circling back to where we started: Pro groundstrokes don’t really break down. They take a few games to reach optimal form, and on occasion, they rattle a bit. Most of the apparent trends we spot as fans are fake. But in this case, there are indeed a few drops of water in the desert of statistical noise.

Did Jimmy Connors Choke in the 1975 Wimbledon Final?

From our vantage point almost a half-century later, it’s easy to forget just how big an upset Arthur Ashe scored with his 1975 Wimbledon victory over Jimmy Connors. Connors was the top seed and defending champion, still riding high from a 1974 campaign that ranks among the best ever. Ashe was a few days short of his 32nd birthday, had a reputation of coming up short in finals, and had lost to Connors in their three previous meetings.

(For what it’s worth, my Elo algorithm thinks it was a much closer match than the bookies did at the time. It rated Ashe the second-best player in the tournament on grass courts, and gave the underdog a 39% chance of winning.)

Ashe ran away with the first two sets and held on to win in four, 6-1 6-1 5-7 6-4. Perhaps because the two men didn’t get along–apart from striking personality differences, Connors and his manager targeted Ashe with one of many lawsuits–the veteran was uncharacteristically critical of his opponent after the match. Ashe claimed that Connors missed many of his shots into the net (rather than long), a sign of choking.

Connors denied it, of course. It later came out that Jimmy was dealing with a foot problem which probably affected his play that day. In any case, fans and pundits surely had their fun debating whether Connors was a choker. I don’t know of anyone who took the question beyond simple speculation. No amount of statistical analysis can settle whether a player choked, but we can often answer adjacent questions to shed more light on the issue.

Counting errors

A couple of years ago I charted the Wimbledon final for the Match Charting Project, so we have a full count of errors–forced and unforced, serves and rallying shots, net and deep–for the entire match. We also have similar shot-by-shot stats for 25 other Connors matches for comparison. (Unfortunately, 24 of the 25 are chronologically later than the Ashe match, because there’s not much full-match footage from the early 70s.)

Here’s the tally: Excluding serves, Connors committed 13 unforced errors, 10 of them into the net. I recorded the type of error for 65 more forced errors: 32 into the net, 33 other. (Ashe was a netrusher, so many of Jimbo’s mistakes were failed passing shots.) On serve, he missed 29 first deliveries: 16 into the net, 13 otherwise. And his two second serve faults were split between one into the net and one elsewhere.

The unforced error split of 10-to-3 means that 77% of his UFEs were netted. That’s the most extreme of any of his charted matches; on average, his unforced errors were half nets, half others. While suggestive, that’s an awfully small sample from which to draw any conclusions.

Using larger samples that include forced errors and serves, the Wimbledon final doesn’t particularly stand out among other charted Connors matches. 54% of his non-serve errors (forced or unforced) in that match were netted, compared to 52% over the whole sample. 55% of his service faults against Ashe were hit into the net, versus 49% across the 26 matches. Altogether, Connors made 54% of his total errors and faults into the net in the Wimbledon final, compared to 51% in the broader sample.

Does it matter?

You’ve probably heard the tennis coaching conventional wisdom that it’s better to hit long than to hit into the net. Like most tennis shibboleths, this one has been around for a very long time. Ashe had surely heard it, which partly explains why he made the comment he did. Arthur didn’t have a printout with match stats generated by a consulting company with a gargantuan marketing budget, so he probably recalled a few key points and generalized from there.

If error types matter, we’d expect to see at least a mild correlation between results (say, percentage of points won) and error types. Let’s stay focused on the 26 charted Connors matches for today’s purposes. Here’s a version of the Ashe hypothesis, stripped of emotional content:

When Connors hits more errors than usual into the net, it’s a sign that he’s playing below his standard level.

It turns out that this theory is wrong–or, at best, possibly correct if narrowly defined. I considered five main stats as indicators of errors and faults going into the net:

  • Unforced errors (excluding double faults) into the net as a percentage of total unforced errors
  • Total rally errors (forced and unforced) into the net as a percentage of total errors
  • First serve faults into the net as a percentage of total first serve faults
  • All serve faults into the net as a percentage of all serve faults
  • All errors and faults into the net as a percentage of all errors and faults

The second (total rally errors) and last (all errors and faults) seem like the most valid of the five, because they give us a decent sample of error types for each match. There is almost exactly zero correlation between the last stat and total points won. And there is a very weak negative correlation (r^2 = 0.05) between the second stat and total points won.

In other words, the Ashe hypothesis might be on to something very minor if our focus in on rally shots. But the correlation is so weak that no human observer would ever notice it, unless they lucked into it by watching a few confirming key moments after being primed by the conventional wisdom.

He didn’t choke like that

I said above that statistical analysis couldn’t settle issues like whether a player choked. We can study what happened, but without machines hooked up to a player’s brain, we can’t tell what was going on inside their heads that might have caused it.

So we can’t say that Connors didn’t choke in the 1975 Wimbledon final. But we have seen that his percentage of into-the-net errors wasn’t that unusual for him (except for the small sample of unforced errors), and we’ve recognized that the number of mistakes he made into the net didn’t have much to say about his level of play that day. If Connors choked, then, it didn’t have anything to do with the low trajectory of his missed shots.

—

I learned of Ashe’s post-match comment in Raymond Arsenault’s excellent biography, Arthur Ashe: A Life.