REF-1714 · Zhang et al., 2026 · Frontiers in Physiology, · Paper · PubMed · Free full text

Ending resistance-training sets at a lower velocity loss gave slightly better jump and sprint gains than letting the bar slow further

Hedges' g 0.29 (95% CI 0.17 to 0.42) for jump and sprint combined, in a three-level meta-analysis of 12 small randomised trials (42 effect sizes) in trained people, all trials at "some concerns" for risk of bias and certainty low to moderate by GRADE; clearest for the countermovement jump and 10 m and 30 m sprints, and for 10% against 20% or 30% loss, while 20% against 40% was not significant (g 0.15, −0.23 to 0.54); strength and muscle size were not outcomes. Read in full

Main caveat: The authors': only 12 trials, so moderator analyses had little power and a missing moderator effect is not evidence of none.
fairly strong evidenceMeasured on recreational to elite athletesRead in full

Zhang and colleagues asked whether stopping each set of resistance training at a smaller loss of bar velocity (fewer repetitions, less fatigue) improves jump and sprint performance more than allowing a larger loss, in people who already train.

They searched five databases to April 2026 and found 12 randomised trials published 2016-2026 that compared two or more velocity-loss thresholds, from 5% against 20% to 20% against 40%, mostly in the squat, with some using sled sprints, loaded countermovement jumps or squat plus bench press.

Because the trials reported several outcomes and comparisons each, the authors used a three-level model over 42 effect sizes. The lower threshold came out ahead by a small amount (g 0.29) with almost no heterogeneity (I² 0.1%); jump (g 0.27) and sprint (g 0.32) both favoured it.

By single outcome the countermovement jump and the 10 m and 30 m sprints were significant and the squat jump and 20 m sprint were not; by comparison, 10% against 30% and 10% against 20% were significant and 20% against 40% was not.

No moderator (athletes against trained non-athletes, weekly frequency, exercise, programme length, size of the threshold gap) changed the effect significantly, which the authors say may reflect too few trials. All 12 trials carried some concerns for risk of bias.

The authors conclude that a lower velocity loss is one reasonable option for speed and power work, not a universally better prescription.

Participants

Pooled sample: 12 RCTs; trained non-athletes (a year or more of regular resistance training), highly trained resistance-trained people, professional soccer players and youth athletes with documented organised training. Athlete and trained non-athlete samples were pooled and also compared as a moderator.
Not in the main text: total number of participants, sex split, ages and per-study sample sizes (supplementary Table S4); the authors call the samples "generally small" and say the evidence comes mainly from people with resistance-training experience, so women, youth athletes and beginners need further study.

Protocol

Search: five databases from inception to 10 April 2026 (965 records), plus 24 records from other sources. 345 database reports were sought, 197 of them could not be retrieved, 148 were assessed in full; 10 trials from databases and 2 from other sources were included.
Inclusion: RCTs comparing different velocity-loss thresholds in trained people (competitive athletes with documented sport training, or non-athletes with at least a year of regular resistance training; youth only with documented organised training); untrained, clinical and rehabilitation samples excluded; jump or sprint outcomes.
Comparisons in the trials: 5% vs 20%, 10% vs 20%, 10% vs 30%, 15% vs 30%, 15% vs 40%, 20% vs 40%, and low vs high velocity loss (Figure 5 also lists 20% vs 30%, 10% vs 45% and 30% vs 45%). Training: mostly squats; also resisted sled sprints, loaded countermovement jumps, squat plus bench press; 4-11 weeks (most 4-8), generally 1-3 sessions a week.
Outcomes: countermovement jump, squat jump, reactive strength index, and 10, 15, 20 and 30 m sprint; sprint times coded so that positive g favours the lower threshold.
Analysis: Hedges' g from between-group differences in change scores, pre-post correlation assumed 0.50 when not reported and re-run at 0.30, 0.70 and 0.90; three-level random-effects model (metafor, REML; effect sizes within studies); no robust variance estimation; leave-one-out; study-level and conventional random-effects models as checks; subgroups and meta-regression exploratory; power-enhanced and contour-enhanced funnel plots, Egger's test on study-level effects.One trial's sex-specific jump data were digitised from a figure.
Risk of bias and certainty: RoB 2 and GRADE (GRADEpro), two reviewers with a third when needed.
Funding and interests: no financial support received; no commercial or financial relationships declared.

Key Results

Comparison: lower vs higher velocity lossEffect sizes (studies)Hedges' g (95% CI)GRADE
Jump and sprint combined42 (12)0.29 (0.17 to 0.42), P < 0.001; I² total 0.1%; prediction interval 0.16 to 0.42not stated in the text
Jump performance22 (12)0.27 (0.10 to 0.45), P = 0.002; I² 0.0%in a figure only
Sprint performance20 (10)0.32 (0.13 to 0.50), P < 0.001; I² 0.0%in a figure only
Countermovement jump—0.29 (0.09 to 0.50), P = 0.005moderate
Squat jump—0.25 (−0.52 to 1.02), P = 0.523, not significantlow
10 m sprint—0.34 (0.01 to 0.67), P = 0.041moderate
20 m sprint—0.29 (−0.00 to 0.59), P = 0.053, not significantlow
30 m sprint—0.39 (0.01 to 0.78), P = 0.047low

By threshold comparison (no significant difference between comparison types, QM(8) = 8.374, P = 0.398):

ComparisonHedges' g (95% CI)
10% vs 30%0.43 (0.19 to 0.66), P < 0.001
10% vs 20%0.34 (0.06 to 0.62), P = 0.017
20% vs 40%0.15 (−0.23 to 0.54), P = 0.430, not significant
10% vs 45% (mostly single-study data)0.50 (0.01 to 1.00), P = 0.044
Moderators (none significant): athletes 0.29 (0.00 to 0.57) and trained non-athletes 0.26 (0.08 to 0.44), P for difference 0.625; two sessions a week 0.32 (0.18 to 0.46), frequency difference P = 0.686; squat-based training 0.35 (0.20 to 0.49), mode difference P = 0.426; programme length β = 0.027 per week (−0.068 to 0.121), P = 0.580; size of the threshold gap β = −0.0004 per point (−0.020 to 0.019), P = 0.966.
Pre-post correlation: pooled g 0.25 (0.13 to 0.38) at r = 0.30 up to 0.46 (0.19 to 0.73) at r = 0.90, all P < 0.001; heterogeneity rose to 12.7% at r = 0.70 and 43.6% at r = 0.90, where the prediction intervals crossed zero.
Other checks: three-level, study-level and conventional models all gave g 0.293 (0.166 to 0.421); leave-one-out g 0.247 to 0.334, all significant; Egger's test P = 0.16 (combined), 0.61 (jump), 0.21 (sprint).
Risk of bias: all 12 trials "some concerns" overall; all "some concerns" for randomisation, deviations from intended interventions and selection of the reported result; all low risk for outcome measurement; missing data low risk in 5 and some concerns in 7.
Examples from single trials, as the authors describe them: at 20% against 40% loss in the squat, 185.9 against 310.5 total repetitions and CMJ +9.5% against +3.5%; at 10% against 30%, 109.6 against 228.0 repetitions and CMJ +9.2% against +5.4%; in one trial with similar training volume, 1RM rose by about 7-11% in both groups, with a slight muscle-thickness advantage (about 6 ± 6%) for the higher threshold.

Practical Application

For jump and short-sprint carry-over, end lifting sets earlier: the lower threshold gave a small extra gain (g 0.29); 10% against 20% or 30% was significant (g 0.34-0.43) and 20% against 40% was not (g 0.15, −0.23 to 0.54), but the comparison types did not differ significantly from each other (P = 0.398), and a bigger gap between thresholds did not give a bigger effect. The overall advantage of the lower threshold was small (g 0.29) and consistent across trials, with fewer repetitions done; it held in athletes and in trained non-athletes alike.
This is a speed-power rule, not a strength or muscle rule. Maximal strength and muscle size were not outcomes of the pooled analysis; in the trials the base already holds, 20% loss gave the largest 1RM gains across 0, 10, 20 and 40% (REF-1085, B+), and one trial described here found a slight muscle-thickness edge for the higher threshold. Choose the threshold by the goal of the block, as the authors advise.
Fits [SYNTH-59] and does not contradict it. SYNTH-59 builds a sprinter's gym work on heavy squats and jumps but gives no rule for how far into a set to go; this review adds that ending squat sets at a low velocity loss favours jump and 10-30 m sprint gains. Transfer: the samples were trained lifters, youth athletes and soccer players, not track sprinters, and the 20 m sprint and squat jump estimates were not significant.
Fits [SYNTH-43] and [SYNTH-44]'s "stop when a rep clearly slows". It extends REF-584 (B+, 20% against 40%, one of the 12 included trials) and REF-516 (B+, ≤25% against >25%, abstract only) with a larger pooled set, read risk of bias and GRADE.
Inference, not tested here: without a velocity device, ending a set when the bar first visibly slows approximates a low velocity loss; for an endurance athlete the fewer repetitions also mean less fatigue carried into the next endurance session, which this review did not measure.

Limitations

The authors': only 12 trials, so moderator analyses had little power and a missing moderator effect is not evidence of none; trials differed in thresholds, exercises, loads, devices and tests; mostly people with resistance-training experience, so women, youth, beginners and other sports need study; 4-11 weeks only, so long-term effects in periodised training are unknown; too little data for reactive strength, 15 m sprint, change of direction and repeated sprints; total training volume differed between thresholds, so the threshold and the volume effects are not separated.
Certainty: all trials "some concerns" for risk of bias; GRADE moderate for the countermovement jump and 10 m sprint, low for the squat jump, 20 m and 30 m sprint; heterogeneity and prediction intervals depended on the assumed pre-post correlation.
Search (our reading): 197 of 345 database reports sought could not be retrieved, so eligible trials may have been missed.
Not in the main text: total participants, sex and age, per-study sizes, the domain-level GRADE ratings and the reasons for downgrading (supplementary tables we did not read).
Transfer: no endurance outcome and no endurance athletes; no track sprinters named among the samples.
How this entry was checked and changed
Why this level

a meta-analysis of randomised trials whose risk of bias and GRADE certainty we read in the full text; not A because all 12 trials were rated "some concerns" overall, the trials are small, and certainty is moderate only for the countermovement jump and the 10 m sprint and low for the other single outcomes