Deep divefrom Episode 9Training5 November 202620 min

Stop the Set Early? Velocity Loss, Jumps and Sprints

A deep dive into one paper: what was known before, what the researchers did, what the numbers say and what they don't.

The paper: Effects of different velocity loss thresholds on jump and sprint performance (Frontiers in Physiology, 2026)

Chapters

Transcript

The script the two AI voices read from. The times are worked out from the script, not heard, so a line can start a few seconds early or late.

Read the whole episode

Cold open

Host 1You heard the trailer — this is the whole story. Welcome to the Hastrup Review of Science — where artificial intelligence reads real science, so you don't have to pretend you understood the paper. Today, one question from the weight room: if you want to jump higher and sprint faster, should you stop a set of squats early, while the bar is still moving fast?

Host 2The paper is a systematic review and meta-analysis by Jie Zhang, Mingyue Yin and colleagues from several Chinese sport universities, published in Frontiers in Physiology in October twenty twenty-six. It pools twelve randomised trials in trained people. The short answer is that stopping earlier came out slightly ahead, and the authors are careful to call that advantage small.

Host 1Before we get into it, what is the actual thing being tested here, in plain words?

Host 2It is a rule for when to end a set. You measure how fast the bar moves on every repetition, and you stop once the speed has dropped by a chosen percentage from the fastest rep. That percentage is called the velocity loss threshold.

Host 1So if my threshold is ten percent and my first squat moves at one metre per second, I stop the set as soon as a rep comes in slower than about point nine?

Host 2Exactly that. A twenty or forty percent threshold lets you grind on much longer before you stop, so you do more reps and get more tired. The paper compares lower against higher thresholds.

Where this comes from

Host 1Why did anyone start measuring bar speed in the first place, instead of just counting reps?

Host 2For most of the history of strength training, the plan was written as percentages of your one-rep maximum and a fixed number of repetitions, say five sets of five at eighty percent. The wider picture in the field is that this ignores daily variation: your true maximum moves around with sleep, stress and fatigue, so eighty percent on paper can be a very different effort on a bad day. Velocity-based training grew up mainly in Spain from around twenty ten, led by researchers such as González-Badillo and Sánchez-Medina, once small sensors could read bar speed cheaply.

Host 1And bar speed tells you something useful about fatigue?

Host 2Yes, that was the key observation behind the method. As you tire within a set, the bar slows down, so the drop in velocity works as a running gauge of how close you are to failure. The paper cites that early work for the idea that a lower threshold means the set ends earlier, with fewer reps and less fatigue.

Host 1Who is this aimed at, really? Is this something an ordinary person at the gym would ever use?

Host 2The question matters most to athletes in sports built on short bursts, such as football, basketball, volleyball and sprinting, and to their coaches. That said, bar-speed sensors and phone apps have become fairly common, so a keen recreational lifter can do this too. But the review only covers trained people, which I will come back to.

What was known before

Host 1What did earlier reviews already say about which threshold is better?

Host 2Broadly, that low-to-moderate thresholds look good for jumping and sprinting, and that the extra volume from high thresholds does not seem to buy extra speed. One earlier review suggested twenty percent or less for jumping and ten percent or less for sprinting. But a network meta-analysis found both low and high velocity loss beat other training approaches, without a clear winner between them.

Host 1So the earlier picture was a lean in one direction, but nobody was sure?

Host 2That is a fair summary. The authors list three gaps. Many earlier reviews compared velocity-based training against classic percentage training rather than thresholds against each other; strength, jumping and sprinting were often lumped together; and the statistics did not handle studies that report many results at once.

Host 1What does it mean for a study to report many results at once, and why is that a problem?

Host 2A single trial might measure a countermovement jump, a squat jump, a ten metre sprint and a thirty metre sprint, sometimes across three different threshold groups. Those results come from the same people, so they are not independent pieces of evidence. If you treat them as independent, one busy study can count several times over and make you more confident than you should be.

The open question

Host 1So the question this team asked is narrower than before: in people who already train, does a lower threshold beat a higher one for jumping and sprinting specifically, and for which tests?

Host 2Yes, and whether anything changes that answer, such as being an athlete or not, how often you train, what exercise you use, how long the programme runs, or how far apart the two thresholds are.

What they did

Host 1How did they find the studies? Did they just pick the ones they liked?

Host 2No, it was a registered, systematic search. They logged their plan in advance on PROSPERO, the international register for reviews, then two researchers independently searched five databases, including PubMed and SPORTDiscus, up to April tenth, twenty twenty-six. They also checked reference lists, ResearchGate and Google Scholar.

Host 1How many papers did that turn up, and how many survived?

Host 2Nine hundred and sixty-five records from the databases plus twenty-four from other sources. After screening, twelve randomised controlled trials remained, ten from the databases and two from the other searches. One detail worth noting: one hundred and ninety-seven reports could not be retrieved at all.

Host 1Is that a lot to lose? Should that worry us?

Host 2It is a large number, and the paper reports it without further explanation. We cannot tell from the text how many of those would have been eligible, so it is a gap in what we can check rather than proof of bias.

Host 1What counted as trained? Does someone who goes to the gym twice a month count?

Host 2No. Either competitive athletes with a documented sport background, or non-athletes with at least one year of regular resistance training. Youth athletes counted only if their training history was documented, and untrained, clinical and rehabilitation groups were excluded.

Host 1And who were the people in those twelve trials, and what did they actually do?

Host 2Youth athletes, professional soccer players, highly resistance-trained individuals, and trained non-athletes. Most trials used squats; a few used sled sprints, loaded jumps, or squats combined with bench press. Programmes lasted four to eleven weeks, mostly four to eight, at one to three sessions a week, and the trials were published between twenty sixteen and twenty twenty-six.

Host 1Were these big trials, or just a handful of people each?

Host 2No, the authors describe the sample sizes as generally small. That is typical for training studies, where you have to supervise every session for weeks.

Host 1Which thresholds were being compared against each other?

Host 2Pairs such as five against twenty percent, ten against twenty, ten against thirty, fifteen against forty, and twenty against forty. Sometimes a trial had three arms, for example ten, thirty and forty-five percent, which produces several comparisons from one study.

Host 1How do you combine a jump measured in centimetres with a sprint measured in seconds?

Host 2You convert each result into a standardised effect size, here called Hedges' g. It takes the difference in improvement between the two groups and divides it by the spread of scores, so everything ends up in the same unit. For sprints, faster times were flipped to count as positive, so a positive number always means the lower threshold did better.

Host 1Is there a rule of thumb for what counts as a big or small value of that g?

Host 2The common convention in the field is that around point two is small, point five is medium and point eight is large. Those cut-offs are rough, but they are how the authors frame their result as small.

Host 1You mentioned the statistical problem of studies with many results. How did they deal with it?

Host 2They used a three-level model. Level one is ordinary measurement noise in each result, level two is how results vary within the same study, and level three is how studies differ from each other. This lets a study contribute several results without being counted as several independent studies.

Host 1Was there anything they had to guess at when doing the calculations?

Host 2Yes, one important thing. To work out the spread of each group's improvement you need the correlation between a person's before and after scores, and most studies do not report it. They assumed point five, then re-ran everything at point three, point seven and point nine to see if the answer changed.

Host 1And did they rate how trustworthy the individual trials were?

Host 2They used the Cochrane risk-of-bias tool, version two, and rated overall certainty with a system called GRADE. All twelve trials came out as having some concerns overall, mainly around how randomisation was reported, deviations from the planned training, and whether analyses were registered in advance. Outcome measurement itself was rated low risk in every trial, because jumps and sprint times are fairly objective.

Host 1One more practical thing. Did they pull any numbers off graphs rather than tables?

Host 2For one study only. Sex-specific jump data from a twenty-versus-forty percent trial by Rissanen and colleagues were read off a figure with digitising software, and a second researcher checked them during revision. Everything else came from reported numbers.

The results

Host 1So what was the headline result?

Host 2Across twelve studies and forty-two effect sizes, lower thresholds beat higher ones with a g of point two nine, and the ninety-five percent confidence interval ran from point one seven to point four two. By the usual convention, that is small. It was statistically clear, with a p-value below one in a thousand.

Host 1Can you put point two nine into a picture I can feel, like centimetres on a jump?

Host 2The paper does not translate the pooled number into centimetres or hundredths of a second, so I will not invent one. What it does give are examples from single trials. In one, a twenty percent group did about one hundred and eighty-six total reps against about three hundred and eleven for a forty percent group, yet improved its countermovement jump by nine and a half percent versus three and a half percent.

Host 1So in that trial the group doing roughly forty percent fewer reps jumped better?

Host 2Yes, in that one trial, and it also did less total work, about one hundred and twenty-eight against two hundred and one kilojoules. A ten-versus-thirty percent trial shows a similar pattern: about one hundred and ten reps against two hundred and twenty-eight, with jump gains of nine point two versus five point four percent. These are individual studies used to illustrate, not the pooled result.

Host 1Did jumping and sprinting both come out ahead?

Host 2Both did: jumping had a g of point two seven, from twenty-two effect sizes across all twelve studies. Sprinting had point three two, from twenty effect sizes across ten studies. The authors rated certainty as moderate for both, and stress that both effects are small.

Host 1Did every specific test agree, or were some of them shakier?

Host 2They were not uniform. The countermovement jump, the ordinary dip-and-jump, was clear at point two nine. The ten metre sprint at point three four and the thirty metre sprint at point three nine also reached significance, but only just, with p-values of point zero four one and point zero four seven.

Host 1And which ones did not reach significance?

Host 2The squat jump, where you hold a half-squat and jump without a dip, came in at point two five with a very wide interval from minus point five two to one point zero two. The twenty metre sprint was point two nine and narrowly missed, at p equals point zero five three. Both point the same way, but the evidence is weaker, and certainty for the squat jump was rated low.

Host 1Why would a jump with a dip respond and a jump without one not respond?

Host 2The authors offer a possible reason, not a tested one. The dip uses the stretch-shortening cycle, where muscles and tendons are loaded quickly and release that energy, and that depends on speed and coordination that low-fatigue training might preserve. A squat jump from a still position relies more on raw force, but with so few data points this could just be noise.

Host 1Did it matter how far apart the two thresholds were, like ten against thirty compared with twenty against forty?

Host 2Ten versus thirty showed point four three, and ten versus twenty showed point three four, both significant. Twenty versus forty was smaller at point one five and not significant. But the formal test for differences between comparison types was not significant either, with p equals point three nine eight, so you cannot rank them.

Host 1So it's not simply a case of the bigger the gap between thresholds, the bigger the benefit?

Host 2The same was not true for programme length, which showed a small positive slope of about zero point zero two seven.

Host 1Did athletes and regular gym-goers respond differently?

Host 2Both came out positive, athletes at point two nine and trained non-athletes at point two six, with no significant difference between them. Twice-weekly training and squat-based programmes showed clear effects within their groups, but no subgroup beat the others in a formal test. The authors call all of this exploratory and probably underpowered.

A word from our sponsor

Host 1We will get to how much to trust all this. First, a quick word from our sponsor.

Host 2The Hastrup Review of Science is brought to you by clothes pegs. Each peg holds one item of laundry, and the evidence for adding a second peg to the same sock is limited.

Host 1Clothes pegs work best when you stop hanging laundry before the line starts to sag. We recommend a low sag-loss threshold.

Host 2Clothes pegs are sold on Amazon at ordinary prices. The link is in the show notes.

How solid is it

Host 1Back to the paper. How sure can we be that this small advantage is real and not a fluke of how they did the maths?

Host 2The direction held up well: dropping any single study left the effect between point two five and point three three, always significant. Fitting simpler models gave the same point two nine three. And under every assumed before-after correlation the effect stayed positive, ranging from point two five to point four six.

Host 1You also said the heterogeneity was tiny. What does heterogeneity mean, and is near zero a good thing?

Host 2Heterogeneity is how much the true effect seems to vary from study to study beyond chance. Here it was estimated at about zero point one percent, meaning the studies looked remarkably consistent. But that estimate was fragile: at an assumed correlation of point seven it rose to about thirteen percent, at point nine to about forty-four percent, and the range of likely effects in a new setting then crossed zero.

Host 1So if I have this right, the claim that lower thresholds help on average is sturdy, but the claim that they help in every setting depends on a number nobody actually measured?

Host 2That is a precise way to put it. With so many small trials, low measured heterogeneity can also reflect low power to detect differences rather than true uniformity.

Host 1Could the result be inflated because studies that found nothing never got published?

Host 2They checked with funnel plots and Egger's test at the study level, and found no significant sign of small-study effects, with p-values of point one six, point six one and point two one. But most effect sizes sat in low-power regions, and with twelve studies those tests are weak. The authors say plainly that publication bias cannot be ruled out.

Why it matters

Host 1What is the core idea a listener should walk away with?

Host 2For speed and power, more reps done tired are not automatically better. Stopping a set before the bar slows much seems to give an equal or slightly better jump and sprint result with substantially less work, at least in the trials here. The authors frame low velocity loss as a quality-control tool, not a magic number.

Host 1So for a coach, the takeaway is that cutting the fatigue may cost nothing for speed, and might help a little?

Host 2Yes, for speed and power phases in trained athletes. The authors also note that higher thresholds may still have a place for building strength, muscle size or volume. One included trial, by Myrholt and colleagues, found the higher-threshold group gained slightly more thigh muscle thickness, about six percent more.

Host 1And for an ordinary person who lifts to stay fit, does any of this apply?

Host 2Only loosely. Everyone in these trials was already trained, and the outcomes were jumps and sprints, not health or general fitness. If your goal is to move quickly, the wider picture from training research is that grinding every set to exhaustion is not required, but this paper does not test beginners or everyday goals.

Host 1What does this paper not let us conclude, even though people might be tempted to?

Host 2It does not show that lower is always better or that there is one optimal threshold. It does not show which threshold to choose, since the comparisons could not be ranked. And it does not show why it works: the explanations about nerve drive and movement quality were not measured, the authors label them possible pathways.

Caveats

Host 1Is the extra speed coming from the low threshold itself, or simply from doing less total work?

Host 2That is the big untangled question. In most of these trials the lower threshold also meant far fewer reps, so threshold and volume move together. The authors explicitly call for trials that hold total volume equal to separate the two.

Host 1Are there other limits worth flagging?

Host 2Several. Every trial had some concerns about bias, the samples were small, programmes ran at most eleven weeks, and devices, exercises and test methods differed across studies. Evidence for reactive strength, fifteen metre sprints, changing direction and repeated sprints was too thin to analyse.

Host 1Who is missing from the evidence?

Host 2The authors name women, youth athletes, beginners and athletes from many sports as under-studied. So you would not want to carry this straight over to, say, a fourteen-year-old who has never lifted.

What to watch next

Host 1What would a study need to look like to settle the remaining questions?

Host 2Larger randomised trials, ideally registered in advance, that equalise total training volume and test several thresholds. The authors also want longer programmes inside a real season, and measurements such as muscle electrical activity, tendon mechanics and motion analysis to check the proposed mechanisms.

Host 1So the honest headline is something like: stopping sets early looks at least as good and probably a little better for jumping and sprinting, in trained people, over a couple of months?

Host 2Yes, with moderate certainty for jumping and sprinting overall, and low certainty for several specific tests. Small, consistent in direction, not yet a prescription.

Sign-off

Host 1That is the deep dive: Effects of different velocity loss thresholds during resistance training on jump and sprint performance in trained populations, by Jie Zhang, Mingyue Yin and colleagues, in Frontiers in Physiology. The link is in the show notes.

Host 1Every Friday this same feed carries the roundup episode: five new papers as trailers, and you pick which become deep dives. This has been the Hastrup Review of Science.