REF-1478 · Swinton et al., 2024 · Sports Medicine · Paper · PubMed · Free full text

How heavy for strength, and how light for jumps?

Across 295 studies: heaviest for 1RM, about 30% 1RM for jumps, 40-70% for power. Uncontrolled effects, %1RM estimated from reps if unstated; it gives no rule for when to add load.

Main caveat: Indirect and uncontrolled. Within-group changes pooled across studies; no control required.
moderate evidenceMeasured on non-athletes to trained athletes

Swinton and colleagues reused a database of strength and conditioning studies (1962 to 2018) and modelled the improvement in each outcome against duration, outcome domain and training dose, adding predictors only if they improved leave-one-out predictive accuracy. Of 295 studies and 535 groups, 372 groups did resistance training alone.

Analyses covered time points up to 26 weeks, the median being 8 weeks. The best model held duration, the average number of sets, and an interaction between outcome and load as a smooth function of %1RM. Maximum strength responded most (0.79, 95% CrI 0.74-0.84) and sprint performance least (0.39, 0.32-0.46), with jump, power and change of direction near 0.5.

Each added week was worth about 0.03 SMD and each added set per session about 0.05. The load profile differed by outcome: for strength it rose with load, with an inflection near 70% 1RM; for jump it fell with load, largest near 30% 1RM; for power it was parabolic, greatest at about 40-70% 1RM; for sprint it rose with load; change of direction showed no clear pattern.

Sessions per week, number of exercises, sex and training status did not improve the model. Our reading: a broad map of what loads suit what outcome. It describes averages over whole programmes and says nothing about how to move load or reps from one session to the next. Its averaging is the authors' first-named limitation.

Protocol

Data source: database built for an earlier larger review; searches of Embase, Medline, Web of Science, SPORTDiscus and Google Scholar, plus hand-searching, to January 2018. Studies 1962 to 2018.
Inclusion: any resistance or majority-resistance study of ≥4 weeks; healthy trained or untrained participants with a mean age between 14 and 60 years; a group of at least four; pre and post means and SDs; enough detail to quantify intensity, volume and frequency. No control group required.
Who: n = 6710 participants in 535 groups. Male-only 65%, mixed 23%, female-only 12% of groups. Untrained 54%, recreationally trained 41%, highly trained 5%.
Training: 372 groups resistance only (n = 4,664); 163 combined with other modes (n = 2,046). Durations 4 to 208 weeks, 97% of the data up to 26 weeks; the analysis was limited to that. Median measurement at 8 weeks (IQR 6-12).
Dose coding: load as mean %1RM; when not stated, estimated from reps per set using Haff and Triplett's table. Overall intensity: low, medium, high, also counting effort for ballistic and plyometric work. Volume: reps per set, sets per session.
Analysis: SMD from baseline with small-sample correction; nested four-level Bayesian model fitted in brms; candidates compared by ELPD-LOO; median and 95% credible intervals. 61 outliers removed. Sensitivity analysis with resistance-only interventions was consistent.
Not done: "Risk of bias assessment was not conducted"; "No attempts were made to assess certainty in the body of evidence".
Funding and interests: no funding received. B. J. S. serves on the scientific advisory board of Tonal Corporation, a manufacturer of fitness equipment; the others report no conflicts.

Key Results

PredictorEstimate (95% CrI)Model comparison
Pooled effect, null model0.55 (0.51-0.60); between-study SD 0.33reference
Each week of training0.03 (0.02-0.04)improved (-12.1, SE 4.2)
Outcome: strength / jump / power / change of direction / sprint0.79 (0.74-0.84) / 0.53 (0.47-0.59) / 0.52 (0.45-0.58) / 0.51 (0.39-0.63) / 0.39 (0.32-0.46)improved (-46.8, SE 12.0)
Each extra set per session0.05 (0.03-0.07); resistance-only 0.07 (0.04-0.09)improved, linear
Overall intensity, low to medium / low to high0.07 (0.00-0.15) / 0.10 (0.01-0.19)improved (-6.6, SE 2.2)
Load as %1RM (smooth)monotonic, non-linear; reduced incline from 65 to 100%improved (-18.0, SE 4.5)
Reps per set (smooth)effect falls with more reps, slowing after tenlimited evidence (-6.4, SE 3.7)
Sessions per week / exercises per sessionno improvement+3.4 (SE 2.0) / +1.0 (SE 2.6)
Sex / training statusno improvement+2.9 (SE 2.3) / +0.2 (SE 2.3)
Outcome by load interactionprofiles differTable 2: -11.7 (SE 4.8)

Negative differences mean a better model. Effect sizes are standardised mean changes from baseline in single arms, not differences against a control.

Load profiles (Fig. 3, authors' descriptions): strength, monotonic rise with an inflection at about 70% 1RM; sprint, monotonic rise; jump, monotonic fall, largest at about 30% 1RM; power, parabolic, greatest between about 40 and 70% 1RM; change of direction, no clear pattern.
Practical line from the authors: focusing on a smaller number of key exercises with multiple sets at an intensity suited to the outcome may beat many exercises at a higher frequency.

Mechanisms

Measured: only change in the outcome against average programme descriptors.
Authors' interpretation: heavy loads raise strength partly because 1RM testing matches heavy training (specificity); they note that some authors have suggested gains from greater load "may be inflated" for that reason. Jump gains favour light loads moved fast. Sprint responded least, perhaps because most programmes train vertical, bilateral, slow force.
Inference: a smooth reaches its highest point at the heaviest loads, but the slope there is small; it does not say a cyclist should go to the top of it. Heavy loads for strength are a supported direction, while the exact load is a trade against the fatigue of the rest of the week, which this model does not see.

Practical Application

How heavy for strength, and how light for jumps? Across 295 studies: heaviest for 1RM, about 30% 1RM for jumps, 40-70% for power. Uncontrolled effects, %1RM estimated from reps if unstated; it gives no rule for when to add load.
Goal decides the load: do not progress every lift the same way. A strength lift moves toward heavier; a jump or power exercise stays light and fast, and adding kg there may move it away from its goal (inference from the profile).
Fewer exercises, more sets on them: sets per session predicted better results, while more exercises or more sessions a week did not. Within a short session, this favours moving sets before adding exercises (inference; single-arm comparison).
Give a scheme time: the model adds about 0.03 SMD per week across a median 8-week programme. It supports not reading a block's result early, not any given block length.
No %1RM in the app: "≥85% 1RM" or "40-70% 1RM" cannot be read from a log. Not from this paper; the base's own practice: low reps at a hard logged RPE stand in for a heavy load, and fast, easy reps for a light one.
Not a progression study: sets and load here are programme averages. Do not cite it for how much to add, or when.

Limitations

Indirect and uncontrolled. Within-group changes pooled across studies; no control required; differences in load across studies come with differences in everything else. Learning and test familiarity sit inside these effects. No risk of bias or certainty rating.
Averages. The authors name it: an average number of sets can hide substantial changes across a programme's progression model. Nothing here tests a progression.
Who. 54% untrained and only 5% highly trained, and 65% of groups male-only. The authors warn that lack of evidence for a training-status effect may reflect this mix and short durations. No endurance outcome.
%1RM. Estimated from reps per set when not stated, and overall intensity and volume coding carried, in the authors' words, "a high degree of subjectivity".
Frequency and volume. The authors expect these to matter in specific settings even though they lacked predictive power here.
Duration. Up to 26 weeks; no inference about seasons or years.
Things that do not add up. The text gives the same ELPD-LOO difference (-0.6, SE 3.5) for the no-improvement sets interaction and for the improving load interaction; Table 2 shows -11.7 (SE 4.8) for the latter. The abstract speaks of adults, while the inclusion criterion is a mean age from 14 to 60 years.
Interests. One author on an equipment maker's advisory board.
How this entry was checked and changed
Why this level

read from the full text (open access, CC BY). The size would suggest A or A-; held at B+ because control groups were not required and effects are single-arm changes compared indirectly across studies, risk of bias was not assessed, no attempt was made to assess certainty, and %1RM was estimated from reps per set wherever a study did not state it (share not printed)

Full text (checked 2026-10-30)
Data from a total of 295 studies comprising 535 groups and 6,710 participants; of the 535 groups, 372 comprised resistance training only (n = 4,664) and 163 combined (n = 2,046) (full text)
Sixty-five percent of groups male-only, 23% mixed sex, 12% female-only; Fifty-four percent untrained, 41% recreationally trained, 5% highly trained (full text)
Durations ranged from 4 to 208 weeks, with 97% of the data from interventions ≤ 26 weeks; median time of measurement 8 weeks (full text)
Risk of bias assessment was not conducted; no attempts were made to assess certainty in the body of evidence; 61 outliers removed (full text)
Each week: 0.03 [95% CrI 0.02–0.04]; each set: 0.05 [95% CrI 0.03–0.07] (full text)
Strength: 0.79 [0.74–0.84]; Jump: 0.53 [0.47–0.59]; Power: 0.52 [0.45–0.58]; COD: 0.51 [0.39–0.63]; Sprint: 0.39 [0.32–0.46] (full text)
Sessions per week: + 3.4 (se: 2.0); exercises per session: + 1.0 (se: 2.6); sex: + 2.9 (se: 2.3); training status: + 0.2 (se: 2.3) (full text)
Inflection point ~ 70% 1RM for maximum strength; jump largest at ~ 30% 1RM; power greatest between ~ 40 and 70% 1RM (full text)
Conflicts: BJS serves on the scientific advisory board of Tonal Corporation; no funding received (full text)