They Meet on the Offbeat

Imagine a pianist and a drummer playing a line of eighth notes. The pianist hangs a little behind the drummer on the numbered beats, but the two arrive together on the notes between them. They separate, meet, separate, meet.

That is a different arrangement from a pianist who stays the same small distance behind on every note. Both might get described as playing “behind the beat.” The description conceals the interesting part: behind on which notes, and together on which others?

I went looking for what a swing ratio leaves out. The useful answer was not that music needs more mistakes. It was that the timing of one player cannot tell us everything about the timing between players.

The triplet is a shortcut

The familiar way to explain swung eighth notes is long-short, long-short. Instead of dividing a beat in half, give the first interval more time and the second less. A triplet approximation makes the first interval twice as long as the second: a swing ratio of 2:1.

Here “interval” means the time from one note's onset to the next, not necessarily how long a finger holds a key. And I will use downbeat in the timing studies' local sense: the numbered part of each beat, not just the first beat of a bar. The offbeat is the eighth-note position between numbered beats.

The triplet picture is useful notation. It becomes misleading when treated as a permanent setting.

In a 1999 author account for the Acoustical Society of America, Anders Friberg and Andreas Sundström described measurements of ride-cymbal timing in recordings by four jazz drummers. Their measured ratios approached even eighths at fast tempi and became more unequal as tempo slowed. A separate task asked 34 listeners to adjust the ratio in synthesized excerpts of “Yardbird Suite”; its averages showed a similar tempo trend. That is evidence against one ratio for all speeds—not a license to prescribe one curve to every drummer.

Their later KTH ensemble demonstration adds a more revealing relationship. In its medium-tempo measurements, soloists' ratios were smaller than drummers' ratios. The soloists lagged on the downbeats but met the drummer on the offbeats.

The players were not simply applying the same swing setting to different instruments.

Same lateness, different relationship

A little arithmetic makes the distinction visible. This is an invented timing diagram, not a transcription or a listening test.

At 120 beats per minute, one beat lasts 500 milliseconds. Give our imaginary drummer a 2:1 pattern:

Event	Drummer	Soloist, everything 30 ms late	Soloist, only downbeats 30 ms late
Numbered beat	0 ms	30 ms	30 ms
Offbeat	333.33 ms	363.33 ms	333.33 ms
Next numbered beat	500 ms	530 ms	530 ms

In the middle column, both soloist intervals are unchanged. The whole pattern has moved, but its ratio remains 2:1. The players never quite meet.

In the last column, the soloist's first interval shrinks to about 303.33 milliseconds and the second grows to 196.67. The ratio drops to about 1.54:1. The soloist is closer to even eighths than the drummer is, yet the pair has acquired a patterned discrepancy: apart on the beat, together between beats.

That last sentence matters more than the decimal. A player can become less unequal internally while the relationship between players becomes less uniform.

None of this arithmetic tells us which version swings. For that, somebody has to judge the music.

What the experiment actually moved

In a 2022 study in Communications Physics, Corentin Nelias and colleagues combined an analysis of 456 jazz solos with a separate listening experiment.

The observational analysis found that positive average soloist downbeat delays were common, though not universal, and tended to get smaller as tempo increased. But its database supplied drum downbeats, not drum offbeats. It could not, by itself, establish that the players met between the beats. Nor could finding a delay establish what listeners preferred.

The experiment supplied a different handle. Eighteen professional and nineteen semiprofessional jazz musicians contributed to the main analysis, rating swing in manipulated versions of piano performances with standardized backing. These were rendered MIDI performances with a quantized rhythm section—not untouched recordings of a live group.

One version aligned the piano and backing to a chosen long-short grid. Another delayed the entire piano part. The third produced delayed piano downbeats and shared offbeats.

Here is the detail I nearly lost in the shorthand: the researchers kept the piano's own swing ratio unchanged across the versions. They did not simply move its downbeats as in my last table column. They shifted the whole piano part, then moved the backing's offbeats to meet the piano's delayed offbeats. That increased the backing's swing ratio while leaving the pianist's internal ratio intact. Their Figure 3 and Methods spell out the construction.

The imposed delay was 85 ticks out of 960 ticks per beat—about 8.9 percent of a beat, or roughly 30–35 milliseconds at the three retained pieces' tempi. It was not an instruction to add 85 milliseconds to every jazz recording.

The version with delayed downbeats and shared offbeats received higher swing ratings than either the aligned version or the uniformly delayed piano version. Simply shifting the whole piano part did not produce a statistically significant improvement over the aligned baseline. That last result is not proof that the two are equivalent; it says the study did not establish a difference between them.

The revealing result is that the piano's internal long-short ratio could stay put while judgments of the ensemble changed.

A result is not a recipe

There are reasons to resist turning this into “the secret of swing.”

The main analysis retained three pieces after a fourth, “Serenade to a Cuckoo,” was excluded because a poorly chosen backing ratio appeared to affect responses. The experiment favored passages rich in eighth notes, used piano with standardized backing, and asked experienced jazz musicians for four-point judgments. It did not measure a dancer's movement or establish the intentions of the musicians in the observational corpus.

There is also a limit inside the manipulation itself. Moving the backing offbeats changes its ratio as well as its alignment with the piano. Potential controls separating those contributions were rejected after a small professional prestudy rather than carried into the full main comparison. The evidence supports this configured relationship; it does not completely isolate every ingredient of it.

That is still a substantial lesson. “Humanize” is too vague a theory. So is “play it late.” A discrepancy can have structure, including places where the players stop disagreeing.

What changed my understanding was the meeting point. I had been thinking of swing as an uneven ruler, then imagining each musician sliding that ruler a little left or right. The more interesting picture has players sharing some arrivals while allowing others to differ.

The next question I want to carry into listening is not merely how far behind a player sits. It is where the players come back together—and what accents, articulation and different instruments do to that meeting. This experiment gives me a sharper question, not a universal answer to it.

Sources and limits

Friberg and Sundström's ASA account and KTH page belong to one research program, not two independent confirmations. I read those complete accounts, not the inaccessible full 2002 journal paper. For Nelias and colleagues I read the complete main article, methods, main captions and Tables 1–2; supplementary files, figure images, audio and raw data were not inspected. The arithmetic above was checked, but I did not conduct a listening or performance experiment. The earlier observations are independent-team context, not a replication of the 2022 manipulation.