top of page

Thanks for submitting!

  • Writer's pictureJo Clubb

Defining high-speed, acceleration, and deceleration efforts in sport

Choosing suitable velocity thresholds receives a lot of attention, and rightly so. It is difficult to decide where to place a cut-off point that determines the change from “high-speed” to “sprinting”, or from “moderate” to “high” for instance.

In reality, this is continuous data that we are trying to bucket into discrete groups. The lack of consensus on standardised cut-off points has led to an array of thresholds being used, both in literature and in applied practice.

The figure below is taken from our editorial in Sports Performance and Science Reports (SPSR), entitled “Tracking Systems in Team Sports: Back to Basics” (Torres-Ronda et al., 2022), available open access here.

In this figure, we’ve applied different thresholds from the literature for “high-speed” and “very high-speed” and illustrated the difference in distance outputs when applied to the same session. These can be wildly different; 2176m vs 243m in the high-speed band for the same athlete depending on which threshold is chosen… basically a 90% difference!

Select high speed running thresholds (A) from a groups of studies with similar subjects (sport, sex) and a visual comparison of the distance outputs of corresponding bands (B) applied to a single hypothetical session using player tracking technology. Notes: Ab = absolute thresholds; Rel = relative thresholds; LS = low speed band (light grey); HS = high speed (dark grey); VHS = very high speed (black)

It is not just the bands that are different, but the type of calculation i.e. absolute cut-offs at particular raw velocity numbers (e.g., 17 km/hr) or relative to an individual’s max speed (e.g., 65% of maximum velocity).

Furthermore, there are differences in the very techniques that are used to determine the most suitable thresholds. While more advanced data science techniques, such as k-means, Gaussian mixture model (GMM), and Spectral Clustering, have been applied to try to determine optimal velocity zones, there has been debate regarding how appropriate these methods are (Park et al., 2018; Kavanagh and Carling, 2019; Lovell et al., 2019).

Minimum Effort Duration

Yet, a setting that can also directly impact the outcome measures but garners less attention, is the minimum effort duration (MED), otherwise known as ‘dwell time’. The MED determines the minimum duration by which the athletes must exceed a threshold to count as an effort (Varley et al., 2017).

Historically, a MED of 0.5 seconds has been suggested to detect high-speed running efforts (Sweeting et al., 2017). However, there is inconsistency in the literature, with 0.2, 0.4, 0.5, 0.6, 1 and 2 seconds all being reported (Delves et al., 2021).

Effort detection generally requires a one or two band drop below the current threshold post-MED before registering a new effort. This is paired with a time out window that specifies the minimum time required to elapse before an effort is counted as a separate effort to the previous one.

Without these rules, erroneous spikes might be captured, or a single effort might be counted as multiple, whereby the athlete’s velocity oscillates across the cut-off threshold.

Figure 2 below is adapted from another figure in our SPSR editorial (Torres-Ronda et al., 2022). Effort B counts as a high-speed effort but not a sprint effort as it does not spend enough time above the threshold, in other words the MED is not met above that threshold. In effort C, the MED is not met so a high-speed running effort is not counted.

Effort detection on a raw velocity trace based on MED for high-speed and sprinting efforts.

This illustrates the challenge of trying to dichotomise (bucket) continuous data. We need to draw the line somewhere… but too short may capture errors, while too long may underestimate the number of efforts.

The MED in Accelerations and Decelerations

The MED was brought back to the forefront of my mind due to a recent publication that investigated this setting within the context of acceleration and deceleration (Silva et al., 2023).

Global Positioning System (GPS) data was analysed from a season’s worth of training data from two professional men’s football teams in Portugal. They analysed acceleration and deceleration data without an MED, as follows:

The duration of accelerations was calculated by the difference between the end of the acceleration (when velocity stopped increasing) and the start of the acceleration (initial velocity), using the same procedure for decelerations but when the velocity stopped decreasing.

They found a high proportion of acceleration and deceleration efforts at lower durations. In fact, if a MED of 0.2 or 0.3 seconds was applied to this dataset (as is commonly used), only 36% of accelerations and 34% of decelerations would be reported. Clearly, this is a concern from the perspective of capturing an athlete’s movement demands and training load.

Another interesting finding in the study was the difference in outcomes between acceleration and deceleration. For example, they found that peak acceleration magnitudes were around 4 m.s−2 with durations of 0.5–1.3 s, while peak decelerations (between 7–10 m.s−2) lasted 0.1 and 0.2 s.

I’ve previously written for Sportsmith about the need to uncouple acceleration and deceleration demands. We frequently consider them together and use settings that mirror each other. Yet, they are fundamentally different in terms of movement demands, with decelerations demanding a higher biomechanical load than accelerations.

This study by Silva et al. (2023) adds to previous work (such as Oliva-Lozano et al., 2020) illustrating that footballers frequently reach a higher absolute maximum deceleration compared to the absolute maximum acceleration. Similarly, Damian Harper has demonstrated that the frequency of deceleration efforts is higher than acceleration efforts across a number of team sports (Harper et al., 2019).

There has been a great deal of attention placed on acceleration and deceleration demands in recent time. Yet, these findings relating to MED suggest that we may have been underestimating the number of efforts. There is no simple fix to the challenge of dichotomising continuous data, such as velocity and acceleration raw data. While research continues to seek the optimal approach, it is important that applied practitioners understand what settings are used in their own environment, and why.

Enjoying my free content? Say thanks here:


bottom of page