Metric Deconstruction

Understanding Metrics in Sports

A metric is a quantitative measure used to evaluate a system, process, or performance. In sports analytics, metrics are used to compare teams or players and determine their relative strengths and weaknesses. However, the overuse of metrics in sports can lead to flawed evaluations if they lack a strong theoretical or analytical foundation. As diligent and curious data scientists and analysts, we should strive to understand how metrics are constructed to ensure that they are meaningful, accurate, and fit for purpose.

Game statistics are the most recognisable type of metric in sports, partly because they were the first type of metric to be collected and presented to the wider public, and partly because of the influence Moneyball has had on professional sports, whose origins can be traced back to Bill James use of Sabermetrics in the 1980s. However, as sports analytics has evolved, so have the types of metrics being created and used to gain a competitive edge. A much broader category of metrics now exists which seeks to move beyond simple aggregations and discrete events to more succinctly capture the dynamics of gameplay. Modern metrics tend to capture the spatial and temporal dimensions of gameplay and performance. Common examples include expected goals (xG) and expected threat (xT).

While more advanced metrics have provided us with new methods to quantify gameplay dynamics and athlete behaviour, there is a tradeoff with their interpretability. In this instance, interpretability refers to a user's ability to break down a metric into its component parts to gain a deeper understanding of it. For example, PlayerLoad is a metric used to estimate whole-body mechanical loading in team sports. It is constructed using raw accelerometer data collected across three axial planes. If we want to gain a deeper understanding of the variance in this metric, we can easily deconstruct it into its component vector parts and investigate further.

PlayerLoad can be considered to have a moderate level of interoperability. Simple metrics like shot success rate are high on the interpretability spectrum, while metrics such as xG are low, given that they are model-based metrics that use learned parameters (more on this in a future post).

Deconstructing a metric is not the only way to evaluate its suitability and value, but it does serve as a logical first step. By understanding the data sources contributing to a metric and the processes taken to aggregate those sources, we can more effectively evaluate if the metric is comprised of meaningful and accurate information suitable for measuring performance, tracking progress, and making data-driven decisions.

It is essential to constantly re-evaluate metrics to ensure that they are still relevant and accurate in light of changes in the sport or available data. Additionally, it is important to remember that metrics are only one piece of information, often presented as a single scalar value or ratio. Where possible, they should be used in conjunction with other qualitative and quantitative information to gain a comprehensive understanding of the system or process being evaluated.

In summary, it is important to understand how metrics are constructed and which data sources they are derived from. While advanced metrics can provide new methods to quantify gameplay dynamics and athlete behaviour, there is typically a tradeoff with their interpretability. Deconstructing a metric is a good first step to evaluate its accuracy and relevance, but metrics should also be used in conjunction with other qualitative and quantitative information to gain a complete understanding of a system's (team or athlete) behaviour and performance. In future posts, we will explore the process of formulating new metrics from a theoretical and analytical perspective.

Thank you for reading this article on deconstructing metrics in sports. We hope that it has prompted you to stop and reflect on your current practices surrounding metric use. If you have any questions or feedback, please feel free to leave a comment below. Please subscribe for more articles on sports analytics.

MC