Training load monitoring and the Goldilocks strategy

Sports scientists face a flood of data from which, they are tasked with selecting the most pertinent and impactful metrics to analyse, track, and report. This include many different data streams, but often load monitoring is in the spotlight. It may be one of the most important areas to translate data into value and action, given the burden it places on athletes (through wearable technology and/or data entry), as well as the expectation of key stakeholders to receive meaningful information.

A combination of internal (i.e. physiological cost such as heart rate or rating of perceived exertion) and external load (i.e. work done such as time-motion analysis) measures have been recommended to capture the training load dose ⁽³⁾. These may then be tracked over a variety of thresholds and timeframes. Therefore, the pool of metrics to be considered for analysis and reporting increases substantially.

Consequently, which metrics should you select is a frequently asked question. In the Age of Technology, we are encouraged to find the simplest solution by reporting as few metrics as possible ⁽¹⁾. On the other hand, authors have warned against being reductionist and urged a multivariate approach to capture the complexity of the training process ⁽²⁾. Therefore, we strive to select enough metrics that value is not filtered out, but not so many metrics that the message is lost. One of our greatest challenge is in selecting the number of metrics that are ‘just right’. My intention here is to briefly discuss four key components of critically selecting the most appropriate metrics for an athlete monitoring system.

Observing the Sport

First and foremost, understanding a sport’s demands is paramount for practitioners. That does not necessarily mean you need to have played the sport to a high level or even played it at all (I can testify to that). However, becoming a student of the game is essential, not just for training load monitoring or athlete monitoring purposes, but for many sports science responsibilities.

Consider the sport demands, not necessarily as metrics, but initially as movements. Put aside the numbers seen on a screen, the variables read on a report and contemplate the physical activities required for an athlete to be successful on the field of play. Use a “beginner’s mind” and put aside biases and expectations and observe the sport with curiosity and openness.

Consider jogging, sprinting, jumping, accelerating, decelerating, cutting, throwing and kicking. Maybe even think about skating, skiing and swimming. Consider the context within which these movements are being executed; space, playing surface, proximity to and contact with other athletes, length of time, and continuous or intermittent timeframes. Delve deeper in observation to consider the demands across teams, playing position, gender, age group, and other categories.

How do physical demands change with the pitch size and numbers of players changes in academy football? How do the demands of cricket differ across playing positions or game formats? How have policy changes, such as no body checking in youth ice hockey or kick-off rule changes in American Football, affected the demands of a sport? How do the high-speed running and collision demands differ between men’s and women’s rugby?

Next, contemplate the types of technologies and metrics that can (or cannot) capture these demands. If relevant, consider the technology you already employ in your own sport and critically appraise its ability to capture these sport-specific demands. This is especially important because as technology continues to evolve, more sport and/or position-specific metrics have been developed, including jumps in volleyball, goalkeeper load, and the quantification of collisions. 

Try not to be biased by your past or current experiences and practices but be led by your observation of the sport.

Does the total distance a goalkeeper covers matter? Is the maximum velocity a golfer reaches travelling around a course relevant? These are of course extreme examples. What about the total distance a centre back covers? Or the maximum velocity a quarterback reaches? Is it appropriate to use wearable technology placed between the shoulder blades to capture lower body stride or skate variables?

We may use different tracking metrics for different purposes, as we will come to later, but critically observing the requirements of a sport is an important step to selecting the most appropriate metrics of an athlete monitoring system. Before we can apply any of this information however, we must seek to understand the measurement precision associated with the metrics.

Understanding Precision

Capturing the movement demands of a sport is an appealing prospect. However, it is our responsibility as scientists to comprehend the accuracy with which a tool measures. As such, understanding measurement precision is a critical aspect of selecting suitable load monitoring metrics. Let’s first define some key terms:

Validity refers to whether a specific variable measures what it is supposed to measure ⁽⁴⁾.

If a technology reports that an athlete hit 20mph (32.2km/h), how accurate is that compared to their actual maximum velocity?

Reliability refers to how repeatable this measure is ⁽⁴⁾. 

If an athlete hit 20mph (32.2km/h) on eight different occasions, how consistent is the maximum velocity measured by the technology? 

Sensitivity refers to the ability of a measure to detect training effects ⁽⁵⁾.

If a training programme improves an athlete’s speed, is the technology capable of capturing this response (i.e., is the signal greater than the noise)?

Technology generally evolves faster than validation research, so it is particularly important practitioners appreciate these concepts. The desire for innovation may drive teams and individuals to become early adopters. While companies may provide in-house white papers or partnered research papers, these should be consumed with scepticism. Systematic bias in outcomes in favour of those funding the research has previously been demonstrated in the pharmaceutical industry ⁽⁶⁾. Therefore, a critical approach is required to avoid biased narratives and objectively assess the merits of the research.

In a brief report entitled “Monitoring Accelerations with GPS in Football: Time to Slow Down?”, Buchheit and colleagues ⁽⁷⁾ questioned the usefulness of acceleration and deceleration measures based on between –model and –unit differences. Not long after this, a survey of high-level football clubs reported acceleration variables to be among the top rated athlete training load metrics used by practitioners ⁽⁸⁾. While metrics may be attractive in relation to the demands of the sport, we need to constantly question: “Are we actually measuring what we think we are measuring with our player tracking or athlete monitoring system?”.

Research has consistently demonstrated that measurement precision of athlete tracking technology decreases with increasing velocity or movement complexity ^{(9, 10)}. This is unfortunate given that such high-speed or high-intensity movements are often those with the greatest physical demand and impact on the game. While these metrics are appealing based on the observation of the sport, it is essential their accuracy is quantified before being incorporated into use.

This caution remains important given the emerging technologies and position-specific variables. Just as Global Positioning Systems moved ahead of independent research earlier this century, new technologies follow a similar pattern as their development moves faster than academia. For instance, McLean and colleagues ⁽¹¹⁾ have expressed concerned regarding the absence of validation work on the NBA’s league-wide, game tracking system, so it may remain up to practitioners to determine themselves.

Precision itself is not an endpoint; it varies depending on a host of factors. Understanding the methodologies and calculations of how a technology actually captures the data can enable users to appreciate when accuracy may fluctuate or even be compromised completely. For instance:

How do different garments affect the precision of wearable technology? What frequency does your heart rate belt use to collect data and how much error was there during the data collection? How is the line of sight from a GPS receiver to satellite affected when you train/play in your stadium?

Precision should be a non-negotiable in the application of sports technology and is a duty we owe to our athletes to ensure any metrics we are reporting about them are accurate. Only then, can we consider the myriad ways we can utilise the data in our setting.

Establish the Purpose

Even with the relevance and precision of metrics considered, a multitude of training load monitoring measures probably remain. As such, a critical step is to establish the purposes we intend to use the metrics for, which will promote the transformation of data into meaningful information that impacts practice. 

A recent review outlined a multitude of ways training load data can be used in the applied setting, as well as applications it should not be used for that (mostly pertaining to injury prediction) ⁽¹²⁾. The use of tracking physical training loads was centred around informing athlete preparation and management, with five overarching levels of applications proposed:

• Long-term use
• Season planning
• Day-to-day planning
• In-session adjustment
• Feedback

This framework reinforces that there is a sliding scale of decisions that this information can support. The daily session report describing pertinent, valid and reliable measures of load is important, but is the tip of the iceberg for how this information can be analysed and used. 

Let’s discuss how metric selection may vary by briefly exploring three examples provided in the review by West and colleagues ⁽¹²⁾.

• Is a player(s) hitting a targeted physical threshold? (in-session adjustment)

Regular exposure to maximal velocity may provide a protective effect from injury ⁽¹³⁾ and consequently, target speeds might be used when tracking live. This application can determine who requires “top-up” running based on session targets, or to monitor the maximum velocities hit in the top-up drills themselves ⁽¹⁴⁾. A critical process is required to determine the most appropriate high-speed threshold to use in the specific setting.

Alternatively, during rehabilitation the target may be to stay below a particular threshold and may incorporate other or additional measures to high-speed. For instance, if rehabbing a quadriceps injury, tracking the volume and intensity progression of decelerations may be of interest.

• Assess player response to previous load and inform decision-making process on risk/reward bias (day-to-day planning)

Measuring individual response in-situ has been demonstrated as a means of using the training environment as a testing platform. Changes in accelerometery variables ⁽¹⁵⁾ and heart rate measures ⁽¹⁶⁾ during a submaximal run test have been demonstrated as sensitive measures in response to training load that could be used.

• Prepare for sport-specific average and maximal competition demands (season planning) 

In an indoor sport setting without access to a position-based tracking system, external load might be tracked over time by an accelerometery-derived measure of load. Alternatively, in competitions where tracking systems vary depending on the venue, reliance on an accelerometery-derived measure (if wearables are permitted) can provide consistent longitudinal data. As well as assessing longitudinal load, a critical thought process is required to establish how intensity demands will be assessed (e.g., temporal duration by segments, moving averages, worst case scenario).

These examples very briefly demonstrate the wide-ranging scope for utilising load metrics. Clearly, there is no one-size-fits-all solution. So, while we seek to report as few metrics as possible to each stakeholder, we must acknowledge the wide variety of metrics that we can apply in different ways to support athlete management. 

Establishing the purposes of different measures should not be a solo venture. Collaboration with stakeholders, which could include management and scouting, coaches, players, medical staff, fitness/strength and conditioning practitioners, is vital to uncover what is of use to each department or colleague. Some may have clarity of the information they want to see, and others may need guidance in how this can support their own processes. This reinforces the fundamental skillset of building relationships and communicating with others. 

This process still requires the practitioner to make decisions regarding the intricacies of the data. For instance, if tracking high-speed running is of interest, the individual tasked with managing the data will need to decide the type of threshold to use, which bands to report, the timeframes over which to analyse, and the number of targeted exposures. There is no avoiding the critical and informed approach required to make these decisions. However, one way this process may be supported is through data reduction.

Data Classification and Reduction

Grouping our load measures may provide a simple way to think about the data. Internal and external load is an important distinction. Further, external load measures have been grouped into three distinct levels in the Gray classification ⁽¹⁷⁾, which uses three distinct levels:

1. Distances covered in velocity zones (e.g., 3426 m in total distance, 489 m above 19.8 km/h)
2. Events related to changes in velocity i.e. acceleration, deceleration, changes of direction (e.g., 18 accelerations above 2.5 m/s/s for a distance of 216 m)
3. Events derived from inertial sensors (e.g., 350 arbitrary units in a manufacturer-specific accelerometry-derived measure, stride imbalances)

There may also be hybrid measures, such as Metabolic Power (levels 1 and 2) or PlayerLoad^TM per metre (levels 1 and 3). Although this is a useful overview from which to consider tracking metrics, we are still left pondering which metrics within each level to select. 

As we have discussed, employing observation of the sport with a critical mindset is one of our most important skillsets. This however, does not mean an analytical approach is not also warranted. Data abundance can be condensed with data reduction approaches. Therefore, we can potentially support the process described throughout this article (observe the sport, understand precision, establish purpose) by using statistical tools that help to reduce the dataset. 

“More things should not be used than are necessary” – William of Ockham ⁽¹⁾

Utilising the most parsimonious dataset is useful with respect to the need to seek the fewest metrics to report. This is especially true in datasets that exhibit substantial multicollinearity. 

Multicollinearity exists when an independent variable is highly correlated with one or more of the other independent variables, which can undermine the statistical significance of the variable ⁽¹⁸⁾. 

There is inherent multicollinearity within load data. For example, velocity (m/s) and acceleration (m/s/s) are reported as separate variables, despite being the first and second differentials of distance and therefore, highly correlated ⁽¹⁹⁾. Analysis has shown different variables are often measuring the same phenomena, such as PlayerLoad^TM over 7 days and Total Distance over 7 days ⁽¹⁹⁾, so it begs the question if we should report both of these. Clearly, there is benefit to assessing multicollinearity and striving to reduce redundancy in the measures we chose to analyse and report.

Principle Component Analysis (PCA) is a dimension-reduction method that reduces a dataset to a smaller one, while still preserving as much information as possible. The application of PCA on subjective measures of training load in rugby union demonstrated three principle components; ‘cumulative load’, ‘changes in load’, and ‘acute load’ ⁽²⁰⁾. Research in rugby league demonstrated the first Principle Component (PC) explained 70% of the total information provided by five training load measures during small sided games ⁽²¹⁾. However, the proportion of variance explained differed according to training mode. For a detailed review of PCA application with athlete monitoring, see Weaving and colleagues ⁽¹⁹⁾ available in full here.

While the reduction of data through such statistical approaches is worthwhile, we must exact caution in overly reducing datasets. Human physiology and the training process are complex entities. When relationships between metrics diverge may be informative. It is important to seek the simplest solutions possible but without losing context or information in the process. It may be appealing to surmise the complexity of training load into fewer metrics, or even a single number or ratio, but such a reductionist approach conceals important information. Once again, we are striving to get the balance of selecting metrics ‘just right’.

Conclusion

Like with many questions in sports science, the answer to “which metrics should I use” is “it depends”. It depends on the sport, the playing position, the measurement precision, as well as the reason why. There is no single answer.

The framework presented here (see Figure 1) is not a cookie cutter solution. It seeks to provide four overarching considerations as part of the metric selection process. There will always be more metrics, more thresholds, and more calculations that a practitioner could consider. The challenge for practitioners is to utilise a critical and informed mindset that allows them to select the most relevant, precise, and meaningful measures across a variety of applications that support athlete management. How these metrics are then reported and communicated is equally important, but that is a topic for another day…

• Beware of the bear with too many metrics that lose meaning, overwhelm stakeholders, and include redundancies.
• Beware of the bear with too few metrics that do not capture the complexity and intricacy of the training process.
• Strive to select the number of metrics that are ‘just right’ for the specific setting, whilst remembering that this is a never-ending, iterative process.

Practical Applications

• Strive to present as few metrics as possible to each stakeholder, but analyse and manage a wider dataset behind the scenes to enable a variety of applications.
• Observe your sport – scrutinise the demands of the sport and playing positions. Consider how this may translate to load monitoring.
• Understand the precision of metrics through independent, peer-reviewed literature and, if possible, in-house assessment. Remember precision is not fixed and therefore, how and when this fluctuates should also be understood.
• Establish the purposes such data can serve in your specific environment to support athlete management. This requires collaboration with stakeholders to understand and/or demonstrate how such information could be useful to their specific processes.
• Where possible, collaborate with data scientists to explore relationships and redundancies in the data, with a view to objectively reducing the dataset.