Velocity-based 1RM estimates wobble most on lower-body lifts
In short
A 2026 systematic review and meta-analysis of velocity sensors and velocity-based 1RM prediction pooled 63 sensor studies and 38 prediction studies and reported high relative validity and reliability, with ICCs of 0.90–0.92. But performance varied by device — linear position transducers were more consistent than inertial measurement units — and large heterogeneity in lower-body exercises significantly biased the prediction results. The authors state that the dearth of measurement error and agreement analyses prohibits final conclusions, so treat a velocity-derived 1RM as a working reference and confirm the number you actually record by lifting it.
The appeal of estimating a 1RM from bar speed is obvious: you get today's maximum without having to lift a maximum. A 2026 systematic review tested how much that method can actually be trusted, in two parts — the sensors themselves, and the prediction models built on them.
Are velocity sensors reliable?
Sixty-three studies covered sensor validity and reliability. Pooled validity and device agreement came out at ICC 0.91–0.92 (k=55 and k=439 respectively), and intra- and inter-day reliability at ICC 0.90–0.91 (k=228 and k=608) — good to excellent across the board.
Device technology moderated those results, though. Linear position transducers, which tether to the bar and measure displacement directly, generally performed more consistently than accelerometer-based inertial measurement units. There is no basis for assuming a phone app and a clip-on sensor hand you the same number.
How accurate is a velocity-derived 1RM?
Across the 38 studies on prediction models, reliability was ICC 0.90 [0.83–0.94] (k=124) and validity ICC 0.91 [0.72–0.98] (k=9). On the average, high numbers. The problem sits behind the average.
First, heterogeneity between studies was substantial, with results scattering by exercise complexity, intensity, sensor technology and modelling approach. Second, and specifically, large heterogeneity in lower-body exercises significantly biased the results. Squats and deadlifts move the bar further and change speed more across the range, so they do not produce the clean load-velocity line a bench press does.
Why is ICC 0.90 not the whole answer?
ICC is a relative measure. It tells you how well a group holds its ranking on repeat testing, not how many kilograms one person's estimate misses by. That is precisely what the authors flag: the dearth of measurement error and agreement analyses prohibits final conclusions. Their conclusion is correspondingly firm — velocity-based monitoring and 1RM prediction require cautious interpretation, and sensor- and exercise-specific evidence remains limited.
How should you use it in practice?
- Do not switch devices. Agreement between devices is decent, but comparing a trend requires measuring it the same way.
- Read each lift separately. A model that fits your bench press should not be assumed to fit your squat.
- Follow the trend, not the absolute. Whether speed at a fixed load rises or falls week to week is the usable signal.
- It is genuinely useful for in-session load adjustment. On a bad day, velocity tells you before the set does.
- Confirm the 1RM you record by actually lifting it.
How this connects to your Muscle Index
Your Muscle Index is computed from three numbers: squat, bench press and deadlift 1RMs. Two of those three are lower-body lifts, which is exactly the area the meta-analysis identifies as biased. Feed in a velocity-derived estimate and the error carries straight through into the score. As a training reference it is fine; the number you log deserves a real attempt.
For reading intensity from speed see what bar speed tells you about your percentage, for profiling from two loads see the two-point force-velocity profile, and for set cutoffs see velocity loss of 10% versus 20%. On when to test for real, pair the best time to test your max with why a fresh 1RM overestimates your fatigued max.
The review was preregistered in PROSPERO (CRD42025634595) and methodological quality was assessed with an adapted COSMIN. Its limitation is inherited: the included studies themselves did not report measurement error thoroughly enough.
Frequently asked questions
Is velocity-based 1RM prediction accurate?
On average, yes — a meta-analysis of 38 studies found reliability of ICC 0.90 and validity of ICC 0.91. But heterogeneity between studies was substantial, lower-body exercises significantly biased the results, and the authors withheld a final conclusion because measurement error and agreement analyses were lacking.
Which type of velocity device is more consistent?
Linear position transducers, which tether to the bar and measure displacement directly, generally performed more consistently than accelerometer-based inertial measurement units. That came from pooling 63 sensor studies, with device technology acting as a moderator.
Why is the error larger on squats and deadlifts?
The meta-analysis found that large heterogeneity in lower-body exercises significantly biased prediction results. Lower-body lifts move the bar over a longer path with more speed variation across the range, so they do not yield the clean load-velocity relationship upper-body lifts do.
If the ICC is 0.90, isn't that good enough to trust?
ICC is a relative measure: it describes how well a group's ranking holds up, not how many kilograms an individual estimate misses by. The authors could not draw a final conclusion precisely because analyses of measurement error and agreement were missing.
So how should velocity tracking be used in training?
Use it as a trend rather than an absolute. Keep the same device and the same lift, and read your condition and progress from whether speed at a given load rises or falls. Confirm the 1RM you log with a real attempt.
Source: PubMed