Asymmetry Percentage: Noise vs. Real Difference in Strength Testing

A high school strength coach retests an athlete's quad strength in week six of preseason and the printout shows a 14% left-right gap on the handheld dynamometer — up from 3% at the initial baseline. He pulls the athlete from unilateral squat work that afternoon and flags him for the athletic trainer. The next morning, same protocol, same tester, the number comes back at 4%. Nothing changed in the athlete's legs overnight. What changed was which side of ordinary day-to-day variation the test happened to land on.

That swing isn't a rare glitch. Every asymmetry percentage that comes off a dynamometer, a dual force plate, or a pair of IMU sensors carries a layer of measurement noise before any real strength difference gets stacked on top of it. Treat every number on the printout as equally certain and two expensive mistakes follow: pulling an athlete from training over a gap that was never real, or clearing one whose genuine deficit happened to land inside a lucky test day. The fix isn't a better cutoff number — it's knowing how much of a given test's own noise sits underneath whatever cutoff gets used, and this piece works through the statistics that draw that line: typical error, standard error of measurement, and the minimal detectable change threshold that separates noise from something real.

Most of the fixed asymmetry thresholds that circulate in strength and conditioning rooms — 10%, sometimes 15% — got adopted because they're easy to remember, not because every test they get applied to carries the same amount of built-in noise. Bishop, Turner, and Read (2018), reviewing the inter-limb asymmetry literature across jump, sprint, and strength tests in a systematic review published in the Journal of Sports Sciences, found studies calculating asymmetry with different formulas, on different equipment, then comparing the result against the same generic cutoff regardless of which metric produced it. A 10% gap on a metric that's normally reproducible to within 2-3% carries a very different meaning than the same 10% on a metric that swings by 8-9% between two clean trials on an uninjured athlete — landing-force asymmetry, covered in more depth in ground reaction force asymmetry research, is a common example of the noisier end of that spectrum. The review's limitation is worth stating directly: it's a descriptive synthesis of heterogeneous studies, not a controlled trial, so it can document the inconsistency without handing back one universal number to replace it — which is itself the point. There isn't one number to replace it with; there's a formula, applied per test.

The trial-to-trial side of the same problem shows up directly in Exell, Irwin, Gittoes, and Kerwin (2012), who tracked lower-limb joint kinetics across repeated sprint trials in the same group of athletes. Asymmetry scores for the same joint moment moved meaningfully from one trial to the next in athletes whose underlying strength hadn't changed between trials, to the point that a single trial could flag an athlete as asymmetric while the very next trial on the same day wouldn't have. Their sample was small and sprint-specific, so the exact size of that swing doesn't transfer directly to a dynamometer or a force-plate protocol, but the underlying point does: a single asymmetry reading is a point estimate sitting inside its own margin of error, the same way a single body-weight reading on a slightly different morning doesn't mean someone gained or lost real mass.

Hopkins (2000), in his widely used reliability framework for sports science published in Sports Medicine, gave that margin of error a name: typical error (TE), the spread you'd see if you tested the same unchanged athlete repeatedly under identical conditions. TE isn't a flaw to eliminate — biological day-to-day fluctuation, sensor placement, warm-up state, and tester cueing all contribute to it — it's a property of a specific test on a specific population that has to be measured before any single reading from that test can be trusted.

Weir (2005), writing in the Journal of Strength and Conditioning Research, turned that idea into arithmetic a strength staff can run without a statistics department. From repeated-trial data, the standard error of measurement is SEM = SD × the square root of (1 minus the intraclass correlation coefficient). From SEM, the minimal detectable change at 95% confidence — the smallest change you can call real rather than noise — is MDC95 = SEM × 1.96 × the square root of 2, which works out to roughly 2.77 times SEM. Below one TE, a reading is statistically indistinguishable from the athlete's own baseline noise. Between one TE and the MDC95 line, a gap is trending toward real but hasn't cleared the 95% confidence bar yet. At or above MDC95, the gap is a statistically real difference, not a coin flip.

Put concrete numbers on it: an athlete whose own repeated testing produces a TE of 3 percentage points has an MDC95 of roughly 8.3 percentage points (3 × 2.77). A 5% asymmetry reading on that athlete sits inside the grey zone — bigger than pure noise, not yet confirmed. A 9% reading has cleared the bar. Those numbers illustrate the math, not a fixed value to borrow — the actual TE has to come from that specific test, on that specific equipment, ideally on that specific athlete, which is exactly what the next section walks through building.

Establishing a personal noise floor doesn't require a research lab, but it does require resisting the urge to skip straight to a single test day. Run the same asymmetry test — a countermovement jump on a dual force plate or bilateral IMU setup, or a handheld dynamometer strength test — on three separate occasions at least 48 hours apart, same time of day where the schedule allows it, with the athlete rested rather than mid-fatigue block. Same tester, same setup, same verbal cueing every time; changing any of those between sessions inflates the very noise this protocol is trying to measure honestly.

Record three trials per side per session and take the best trial per side, the same way the underlying test is normally scored. Calculate asymmetry as (stronger side minus weaker side) divided by stronger side, times 100, for each of the three sessions. Take the standard deviation of those three session-level asymmetry readings — that figure is a rough, early estimate of the athlete's own TE for that test. Three sessions is a workable starting point, not a finished number; Hopkins recommends considerably more repeat trials than three for a precise TE, so treat this as version one of the estimate and let it firm up as more regular test data accumulates across the season rather than locking in the three-session number permanently.

A useful field version of the same test lives in single-leg CMJ asymmetry testing, which gives a faster, lower-fatigue readout between full dynamometer sessions once the noise floor is established.

Retest cadence should track how fast the situation can change, not a fixed calendar. During return-to-play after injury, weekly retesting catches a closing or widening gap early enough to adjust the program; general in-season monitoring can run monthly without missing anything that matters. Treat a single reading above MDC95 with real but measured concern — confirm it with a second test within a few days before changing training, the same way a single high blood-pressure reading gets a second check before a diagnosis changes. Two consecutive readings above MDC95 is a much stronger signal than one, and that pattern, not the single number, is what should drive a training decision.

Statistical reality and practical importance aren't the same question, and it's worth holding them separately. A gap that clears MDC95 on a very precise test — one with a TE under 2%, common for jump height on a good IMU or force-plate setup — might still be small enough in absolute terms that it doesn't change training. Meanwhile a test with wider inherent noise needs a considerably larger raw percentage before it clears its own MDC95, so borrowing a tight threshold from a different, more precise test would flag that noisier metric constantly for gaps that were never statistically distinguishable from normal variation in the first place — a mismatch worth checking against the broader threshold research summarized in limb symmetry index cutoff research. The practical rule that holds up across both cases: let MDC95 answer whether a number is noise, and let the accumulated injury-prediction and asymmetry research for that specific population and test answer whether a confirmed real difference is actually worth acting on. A reading needs to clear both bars before it changes a training program, not just one.