The first presenter on Thursday was our host, Dave Tenney, Sounders FC Fitness Coach. His presentation was Creating the Player Monitoring Model – Can we accurately capture the fitness/fatigue cycle?
Question: Can a model of player performance be developed which will take into account the match, training, and recovery in order to predict player readiness to play and train? That was the question that Dave Tenney attempted to answer with the Seattle Sounders FC, and this presentation presents his concepts, background, and methodologies on that path towards an answer. In fact, he was able to develop a model which, given the match, training, and recovery inputs, he can predict the likelihood of a player being injured in a match and his readiness to train. His model is continuously being improved and tweaked as more data is being generated.
Note: Some of the information presented during the three day seminar was proprietary. Out of respect for the proprietary nature of some of the information, I am leaving out some details, which, frankly, are not germane to this discussion anyway.
The following are my notes of his presentation:
- Performance optimization in sport requires that coaches and trainers understand and control, as much as possible, the effects of the training and the match on the individual players.
- This is the next phase of improvement in sports science.
- Technology has advanced exponentially over the last several years, and this includes technological advancement in sports science.
- However, the coach’s ability to grasp and utilize the technology has not increased at the same rapid rate.
- This difference between what is technologically possible and what coaches and trainers can use based on ability has created a “Sports Science Gap”.
- We also have to be careful with measuring unnecessary things. Just because you can measure something does not mean you should, and the data you generate might not be applicable or relevant to the sport. The job of the sports scientist, coach, and trainer is to only measure what is valuable in terms of improving team and player performance. Anything else is a waste of time, energy, and resources. It can also lead to incorrect conclusions which can have a negative impact on performance.
Player Monitoring Process
- The player monitoring model must encompass the play->train->recover cycle
- Recovery must be included in the cycle: recovery from the game and recovery from training
- Highest priorities are match readiness and injury avoidance
- The data generated from the game, post-game recovery, and training is critical to the accuracy of the model. This data will serve as the inputs to the model of player monitoring.
- The data will come from different sources in different formats (units), including both objective and subjective.
Building the Model
- Levels of analytics and statistics: Descriptive -> Predictive -> Prescriptive
- Descriptive – Looks at past performance as a reason for success or failure
- Predictive – Looks at “what will happen” but not why.
- Prescriptive – “What, why, and when it will happen.”
- Most never get past the descriptive stage.
- Goal is to be prescriptive.
- Reference: Rationale and resources for teaching the mathematical modeling of athletic training and performance, by David C. Clarke, et al.
- Development of a model is a long-term process that must be institutionalized.
- During the first year all the relevant data is collected.
- After the data is collected, the model is created and tested.
- Afterwards the results of the model are continuously evaluated and modified.
- “Models evolve.”
- “All models are wrong, some models are useful…” George Box, Industrial statistician.
Data Collection (Inputs)
- Match data – reference: Monitoring Training in Elite Soccer Players: Systematic Bias between Running Speed and Metabolic Power Data, P. Gaudino, et al.
- When assessing a player’s readiness to play or train the best case is the true positive or true negative in which the player was accurately predicted to be ready or not to train/play.
- The second best prediction would be the false positive in which the model predicted that the player was not ready to train/play even though they actually were. This is obviously not an accurate prediction but the error will not cause injury to the player.
- The worst prediction would be the false negative in which the model predicted that the player was ready to train/play when in fact they were not. This type of error in the model can result in player injury.
- Reference: Physiological assessment of aerobic training in soccer, Impellizzeri, FM, et al, Journal of Sports Sciences.
- Determining the total training load of the player during training is critical. It is the sum of the internal and external load on the player in training.
- The external training load consists of the velocity load (distance covered in each speed zone in training) and the body load (acceleration/deceleration).
- The internal training load consists of the HR load on the player in training.
- Individual player characteristics will also have an effect on the training load, as well as pre-training fitness.
- Of particular interest was the recent research by Casamichana, et al on the difference in workload between 11v11 and SSGs.
- Reference: Comparing the Physical Demands of Friendly Matches and Small-Sided Games in Semiprofessional Soccer Players, D Casamichana, J Castellano, et al.
- Also reference the Julen Castellano interview by Mladen Jovananic. The links are shown above.
- In the research it was found that 4v4 SSGs had a higher body load and lower velocity load than 11v11 games.
- In other words, the SSGs involved a higher number of short sprints, thus more acceleration/deceleration. The higher player density resulted in a low total number of sprints.
- In contrast, the larger playing area (and correspondingly lower player density) resulted in the 11v11 games having more medium and long sprints, thus allowing players to reach a higher percentage of their top speed.
- Relative vs. absolute: “Measuring individual [speed] zones… it could be an option, but we have to think that football is a sport of absolute values, in other words, for the game we need to know who is faster than another and not if the players ran at their 90% max speed.” Julen Castellano.
Main Points & Takeaway Messages
- Player monitoring is constantly changing.
- Modeling only improves with more data.
- Not all data is relevant.
- Sport scientists must work closely with the coach/manager. Their aim must be to help fulfill the vision and needs of the coach/manager.
- Incorporate player monitoring into weekly & monthly game-training cycle.
- Strive to objectify results – give coach meaningful data regarding results of model. Let coach decide how to use.
- Be able to justify numbers and opinions.
- Work to go from simply descriptive to predictive then prescriptive.
- Best model takes into account variety of stressors including lifestyle, social, environmental.
Overall this was an impressive presentation of the current state of the art in professional sports. What Dave Tenney and his crew of sports scientists have been able to do is at the cutting-edge of player performance modeling. Not all big pro clubs have reached this stage, yet. However, this is also not something that can realistically be implemented at any other organization than a professional club in which the money and expertise are on hand and can be used in conjunction to push the limits of sports science. Of course, that does not mean that this is a wasted effort and serves only the elite. The results and methodologies developed by Dave Tenney and his crew will undoubtedly trickle down-stream and result in improvements for all evidence-based coaches and clubs.
Over the next few days (weeks?) I intend to continue to provide my notes and opinions regarding several of the presentations.