RSI Inflated by Short Contact Time: How to Find and Fix the Measurement Error

Jump four of a six-drop-jump set from a 30 cm box reads RSI 3.10 - contact time 0.091s, flight time 0.282s. Jumps one, two, three, five and six are all sitting in a tight band between RSI 1.95 and RSI 2.15, with contact times around 0.19-0.21s. Watching the session live, jump four did not look faster or stiffer than the others. If anything, the athlete looked slightly more cautious coming down, not less. But the number says otherwise, and if that 3.10 gets averaged into the session or logged as a new personal best in reactive strength, it is now steering programming decisions off a jump that almost certainly never happened the way the number claims.

This is the pattern that catches people who trust RSI as a single clean output rather than as a ratio built from two separately measured, separately fragile timings. Reactive strength index is jump height divided by ground contact time, or flight time divided by ground contact time depending on which formula your device uses - either way, contact time sits in the denominator, and a denominator that is measured even slightly short inflates the whole ratio in a way that looks exactly like a genuinely stiffer, more reactive jump. The athlete did not get 45% more reactive between jump three and jump four. The mat, the plate, or the algorithm clipped part of the true ground contact window, and nothing about the resulting number warns you that it happened.

Contact times in a well-executed drop jump from a 20-40 cm box typically run 150-250 milliseconds for athletes training reactive strength deliberately, and can drop under 180 ms for well-trained jumpers emphasizing stiffness. At that scale, a timing error of just 20-30 milliseconds - well within what a noisy contact mat or a poorly filtered accelerometer signal can produce on a single rep - is not a rounding error. It is 10-15% of the entire denominator.

Because RSI is a ratio, that error does not just shift the number a little, it compounds against whatever jump height or flight time sits in the numerator. A jump with completely normal flight time of 0.28s and a true contact time of 0.20s gives an RSI of 1.40. Clip 30 ms off that same contact time - down to 0.17s from a mistimed touchdown detection - and the RSI jumps to 1.65, an 18% increase from a timing artifact alone, with zero change in how high or how reactively the athlete actually jumped. Flight time errors of the same magnitude barely move the number by comparison, because flight time is almost always longer and less prone to edge-detection error than the abrupt, fast-changing signal at touchdown and takeoff. Contact time is the fragile half of the ratio, and it is exactly the half most vulnerable to sensor and algorithm error.

The error is rarely random. It clusters around a handful of specific mechanical and algorithmic causes, and each leaves a slightly different fingerprint.

Of these, late touchdown and early takeoff detection are the most common causes of a single-jump outlier within an otherwise normal set, because they depend on the specific dynamics of that one landing - a slightly harder strike, a slightly different foot angle - rather than a systematic device limitation that would affect every rep equally.

Not every fast contact time is wrong, and treating every short reading as noise means throwing away real improvements in stiffness. Four checks separate a clipped reading from a genuine one.

• Check contact time and RSI together, never RSI alone. A genuinely stiffer jump shows a short contact time paired with flight time that is unchanged or only modestly higher. If RSI jumped mainly because contact time dropped sharply while flight time stayed flat or even dropped slightly too, that is the profile of a timing artifact, not a stiffness gain - a real stiffness improvement usually still produces comparable or better jump height, not a jump that got both shorter on the ground and no higher in the air.
• Compare against the set's own distribution. One contact time more than roughly 2 standard deviations below the session mean, surrounded by normal values on both sides, is a statistical outlier in most working sets of five or more jumps. A genuine neuromuscular improvement in stiffness across a session tends to show a gradual downward drift in contact time as the athlete grooves the pattern, not one isolated jump that is 30-40 ms faster than everything around it.
• Cross-check with video at 60 fps or higher if you have it. Count frames from first visible foot contact to the frame where the foot clearly leaves the surface. At 60 fps each frame is 16.7 ms, which is precise enough to catch a 20-30 ms discrepancy against what the device reported.
• Look at where in the rep sequence it occurred. A suspicious reading on the very first drop jump of a set - before the athlete has settled into a rhythm - or immediately after a rep with an obviously awkward landing is more likely to reflect an atypical foot strike than a genuine change in reactive ability.

This is not a theoretical concern about sensor precision - it shows up directly when contact-time devices are compared against force plates, the field's reference standard. Healy, Kenny and Harrison (2018, Journal of Biomechanics) compared contact mat, IMU and force plate measures of ground contact time and RSI during drop jumps, and found that while flight time agreement across devices was generally strong, ground contact time showed meaningfully greater variability between device types, with the contact mat in particular prone to underestimating contact time relative to the force plate on faster, stiffer landings - precisely the landings where the true contact window is shortest and hardest to resolve with a lower-resolution timing method. Because RSI is calculated from that contact time, the paper's device-comparison data show RSI values diverging between systems even when the same athlete performed the same jump, purely as a function of how each device's algorithm handled the touchdown and takeoff transition.

A second line of evidence comes from reliability work rather than device-agreement work. Markwick, Bird, Tufano, Seitz and Haff (2015, International Journal of Sports Physiology and Performance) examined the reliability of RSI derived from drop jumps across multiple box heights in team-sport athletes and reported that while RSI showed acceptable reliability at a group level, individual-rep variability was high enough that a single jump's RSI should not be treated as a stable trait value without confirming it against the surrounding set - single-trial noise, much of which traces back to contact-time measurement variability rather than genuine trial-to-trial changes in jumping strategy, was large enough to produce misleading single-rep conclusions if taken in isolation. Both papers point at the same underlying limitation from different angles: contact time is the less reliable half of the RSI calculation, and any single reading needs corroboration before it gets treated as real.

Run this whenever a single RSI reading looks meaningfully better than the rest of a set - roughly 20% or more above the session average is a reasonable trigger point.

1. Pull the raw contact time and flight time for the suspect jump alongside the same values for every other jump in that session at the same box height.
2. Calculate the session mean and standard deviation for contact time specifically, not RSI. If the suspect jump's contact time sits more than 2 SD below the mean, flag it.
3. Check flight time for the same jump. If flight time is flat or lower than the session average while contact time is the outlier, this is very likely a timing artifact rather than a real change.
4. Have the athlete perform 3 additional drop jumps at the same box height, ideally within the same session or no later than the next training day, with clear instruction to land and rebound at their normal pace rather than trying to consciously replicate the fast number.
5. If the retest jumps land back in the original 1.95-2.15 RSI band (using this example's numbers) and none reproduce anything close to 3.10, treat the original reading as a measurement artifact and exclude it from trend data. If two or more retest jumps land meaningfully closer to the original high reading, the improvement may be real, and it is worth reviewing video or force-plate data if available before fully accepting it.

The retest step matters because a single confirmatory jump is still just one more sample - three to five retest jumps give you a distribution to compare against rather than one number chasing another number.

A club-level volleyball player performed six drop jumps from a 30 cm box during a monthly reactive strength assessment. Raw data below, followed by the retest set performed two days later at the same box height.

The session mean contact time excluding jump 4 is 0.199s with a standard deviation of roughly 0.004s - jump 4's 0.091s sits over 25 standard deviations below that mean, an implausible physiological jump and a clear sign of a touchdown-detection failure rather than a real change in landing mechanics. Flight time for jump 4 (0.282s) is entirely unremarkable and matches the rest of the set closely, confirming the anomaly is isolated to the contact-time measurement. On retest two days later, five jumps at the same box height produced contact times between 0.192s and 0.207s and RSI values between 1.35 and 1.45 - nothing close to 3.10 appeared. The original jump 4 reading was excluded from the athlete's tracked RSI trend, and the touchdown-detection threshold on the device was adjusted downward slightly to reduce the chance of a repeat on future fast landings.