Here are 10 popular non-scientific principles that I like to consider in practice.

They may not be absolute truth, but I think they are generally useful.

⬇️

Yes…missing the most important one.

Brandolini’s Law.
(aka The 🐂💩 Asymmetry Principle).

I'll leave that one for another day!

10 popular non-scientific principles to keep in mind

Sport tech companies / products are frequently bashed by coaches. There is something stoic about solving all problems with spreadsheets and some sort of badge of honor as a result of sticking to basics. 

As if using tech and taking care of basics were mutually exclusive.

Some tech/products are universally bad. Some are good. Some are good or bad depending on context.

The importance of the humans operating the products is highly relevant to the usefulness and success of any given technology.

If the best option is using excel over x or y software, or not using tech vs using it, then great. 

I am more interested in the critical process that led you to make that decision than the decision itself.

The tech field is expanding, and it will continue to do so.

Some companies/products have been a game changer. Some do honest marketing, some may oversell, just like in any other industry. Try buying a car.

If you are an experienced practitioner share your feedback with them.

Looking back, I probably had better discussions with companies I ended up not using.

Extreme positions are not usually helpful,  they are just the loudest and most common on twitter.

Be critical. Educate yourself. There are opportunities out there. 

Technology in Sport

“The idea of an operational process, without a philosophy or values, without a specific stance on how to look at the world, is highly problematic because every groundbreaking design has a philosophy behind it, good or bad”. Natasha Jen.

Source: https://www.fastcompany.com/90411683/two-top-designers-debate-the-value-of-design-thinking

Replace “design” with any other sport related term (i.e. “coaching program”). It still makes sense.

Drills, routines, certificates, metrics, tech…without a philosophy there is no direction.

What is your x or y philosophy is one of the most important questions you can ask.

Philosophy exists at both levels, organizational and individual…we’ve all heard before that “someone wasn’t a good fit”.

When I hear this it’s either a matter of personality (likeability) or a misalignment between that person’s philosophy and the group or organization’s one.

To cut it short - what I’m trying to say is that developing your philosophy is important, try to poke holes, be critical, explain to others, answer questions.

It’ll help later, i.e. job interviews, managing teams, etc.

By the way, “develop” is the key word. It can’t be learned.

Importance of Philosophy

10 things I learned working in pro sport as sport scientist.

A visual thread
👇

You won’t have to sell anything if you are able to show links to performance.

Recovery is more about training than about resting.

Problem solve first. Innovation will come naturally.

There is no shortage of questions. Learn the skills to answer them.

Collect data only for the things you can’t see.

The more complex is the environment the easier should be the solution.

Self-protection is detrimental in the long term. Hire the best and make adjustments if you have to.

Advantages are temporary. Continuously audit your processes to keep finding them.

Work-life balance is hard. Those who chase it end up being frustrated at both ends. 

Get in, sprint (for as long as you want) then get out.

Ensure those around you are onboard.

I am sure some of these are controversial.

I’d love to hear from others with experience in the field.

🤙

Lessons learned from working in pro sports as a sport scientist.

40 questions to ask a sport scientist.

⬇️

If you could solve only one problem in your environment, what would that be?

What’s your biggest contribution to the athletes / coaches / team you work with?

From all the data that you are collecting with your athletes, what is a must have and what would you be comfortable letting go?

If you quit your job today, what would they miss the most?

How do you go about getting buy-in from your athletes?

If you were to hire a sport scientist to join your department, what (cert, skills, etc) would you prioritise?

What do you want to achieve in the short (~6 months) and long (~2 years) term in your current job?

What KPIs do you use to evaluate / understand how your athletes are progressing through the training program?

What research would you like to do in your environment and why?

How much does your work impact the performance of your athletes / team and how do you (if possible) measure it?

If you can only invest in one technology or equipment what would that be?

What advice would you give to a fresh sport science graduate?

Given what you know now, would you still want to become a sport scientist if you were born again?

What are your top 3 principles for injury prevention?

Which one do you use more? For your selection, what are the main limitations?

• Excel
• Python
• R
• Matlab
• Other (which one?)

What’s the lowest hanging fruit in your environment? 

What’s the most common type of analysis that you have to do with your data?

List your top 3 recovery modalities for your environment? (besides sleep and nutrition)

How does your job (or part of it) helps minimize the risk of injury happening to one of your athletes?

How do you know your model is not overfitting? What do you do if it is?

What part of your job would you like to automate but you can’t?

Is there anything that you do knowing that it doesn’t make a statistical (quantifiable) difference but you still think it is important to keep doing it?

In your experience, what strategies work better to communicate data / results with athletes? What about other staff members?

What’s the weakest link in you current data infrastructure? What makes you think so?

What’s the most influential book you’ve read? 

What’s the reliability of the top 3 most important metrics you are collecting? Are you happy with that?

You have an athlete on the force plate and you have 240 characters to coach / teach / cue a CMJ assessment. How would you use them?

What’s one AMS feature that you really need but you can’t get?

What objective information do you collect to guide the rehab process for the most common injury in your sport?

How do you assess significance, magnitude and direction of change with your KPIs and athletes?

Is there any part of your athlete monitoring process that is negatively affecting the quality of your data? 

How are you supporting your medical/ATC team?

Is there anything that you do on game day (or race day, etc) that supports your athlete’s performance? If so, why do you believe so?

What content would you add to sport science degrees that is not currently being taught at universities?

Do you factor in ecological validity in you assessment / monitoring process?

Is there anything that you successfully implemented in a previous job that does not work in your current one? 

How much do you know (or don’t know) about the physical demands of your sport?

And a classic one to finish, 

What does success looks like in your role?

40 Questions to Ask a Sport Scientist.

I am more open to sharing my opinion here now than before.

I know I am going to come across people who disagrees with me more often, but I also get better interactions.

Being comfortable handling different opinions is not something you learn at university. It's a process.  {{ img:e64222 }}

Some of the most successful people I met in sport are really good at this.

You get better by observing and by self-reflecting about situations that could've been managed differently in the past

At times, I underestimated how important this is to work in high performance sport.

Those willing / able to share well reasoned opinions, become more valuable and progress faster within teams than those who don't.

Creating an environment that fosters this behaviour as well as knowing how, when (and when not) to express your opinion is just as important.

I read about managing disagreement (and other "soft topics") less than what I should, but this thread by @businessbarista is well worth a scroll as well as his 13mins podcast on the topic.

https://twitter.com/businessbarista/status/1495406816716595201?s=21

Navigating Disagreements in Sport

#NBA Schedule Density App
--

A couple of years ago I created this #shinyapp to explore the density of the schedule in the NBA. 

I just published an update to account for some of the schedule challenges due to COVID.

Live App: https://josedv.shinyapps.io/NBASchedule/

For more background 👇

1. Purpose
-
My goal was to come up with a way to explore the stress of the schedule for both, teams and players, using only publicly available data, mostly box scores.

Understanding periods of higher / lower schedule stress could help coaches manage work and plan accordingly.

2. Schedule Stress
-
There are many ways to look at this. In this case, I was constrained to only public data, so I focused on the interaction of various geographical and time related factors such as:

- Location
- Time Shifts
- Travel Direction
- Time between games
- etc.

3. Game Density Index
-
I won't go into calculations here (the code is available for those interested). Long story short, the combination of time + geographical factors results in a game density score (or index) for each game. A number between 0-100 applied equally to all teams.

4. Rolling Index
-
Along with the game index, a rolling index over n days can help evaluate the density of  the schedule over a period of time, which may be practically more useful than a single game index.

5. Early Prototypes
-
I started playing with this idea in excel. Below is an example of the schedule index for the Bulls in 2019. You can intuitively see periods of higher (red) and lower (green) density. 
-
Full link to this file here: https://github.com/josedv82/NBA-Game-Density-App/blob/master/NBA%20Schedule%20Density%20Excel%20Prototypes/Bulls%20Schedule%20Density.pdf

6. How much weight?
-
Users can change the weighing factors for all different schedule metrics, for example, I may score crossing two time zones as a 3, but for someone else it may be a 2.5

I left it open so users can explore. Also, some of the articles above may help with this.

7. From Excel to Shiny
-
Eventually I turned that into an #rshiny app to be able to look at multiple teams over various seasons, dynamically adjust schedule factors, look at teams & their opponents, etc. 

I'll drop a few screenshots of key parts of the app in the next few tweets

8. Team Season Report
-
This is basically the translation of the excel table above into a datatable in shiny.

9. Player Report
-
Looking at the combination of schedule factors + minutes played. The app displays individual loads for players in various ways.

10. League Density Heatmap
-
A visual to look at the density of the schedule for all 30 teams. The image below is for the 2019-20 season. You can see the "bubble", in green on the far right.

11. League Wide Player Report
-
Another report looking at individual player loads for all teams. There are filters to look at different factors.

12. Research / Articles
-
I added a table with ~174 research and media articles on anything schedule and basketball. This is available for download. 

I'd like to think this is one of the most extensive resources on this topic out there.

I try to update this table every 6 months

13. Limitations
-
There are always many limitations, even more if we are limited to only public data.

But I hope this is helpful for those like me, that are curious about this topic and have no access to team proprietary data.

14. Predicting Outcomes
-
This comes up often. I didn't create this index under the assumption that it could predict outcomes/injuries, but rather, I wanted a bird's-eye view at the beginning of the season of easy / hard schedule windows for planning purposes if I was a coach.

15. Prediction Research
-
Outside of the scope of this app, if interested in predicting game outcomes based on schedule, we wrote something that may be of interest here: https://www.researchgate.net/publication/357856729_Hiding_in_plain_sight_schedule_density_and_travel_influence_on_NBA_game_outcomes
-
Hiding in plain sight: schedule density and travel influence on NBA game outcomes.

16. Access Schedule Data
-
For those interested in downloading NBA schedule data, I created {airball}. 

An #Rstats package that makes it really easy to download various schedule related metrics in a tidy way. 
-
Available here: https://github.com/josedv82/airball

Finally,
-

If you want to check the code behind the app or if you spot any issues, it is all available on my GitHub page here: https://github.com/josedv82/NBA-Game-Density-App

NBA Game Density App (Rshiny)

Machine Learning in Sports
--

An ongoing list of reviews and commentary articles on the use of #MachineLearning / #ArtificialIntelligence within sport applications, with emphasis on sport science.

Artificial Intelligence and Machine Learning in Sport Research: An Introduction for Non-data Scientists.

https://www.ncbi.nlm.nih.gov/pmc/articles/PMC8692708/

Machine Learning in Sport Science. Challenges and Opportunities.

https://www.tandfonline.com/doi/full/10.1080/14763141.2021.1910334

A Narrative Review for a Machine Learning Application in Sports: An Example Based on Injury Forecasting in Soccer

https://www.mdpi.com/2075-4663/10/1/5/htm

Current Approaches to the Use of Artificial Intelligence for Injury Risk Assessment and Performance Prediction in Team Sports: a Systematic Review.

https://sportsmedicine-open.springeropen.com/articles/10.1186/s40798-019-0202-3

Machine learning in sports medicine: need for improvement.

https://pubmed.ncbi.nlm.nih.gov/33833038/

Black Box Prediction Methods in Sports Medicine Deserve a Red Card for Reckless Practice: A Change of Tactics is Needed to Advance Athlete Care.

https://pubmed.ncbi.nlm.nih.gov/35175575/

Machine learning methods in sport injury prediction and prevention: a systematic review

https://pubmed.ncbi.nlm.nih.gov/33855647/

Ideal algorithms in healthcare: Explainable,
dynamic, precise, autonomous, fair, and
reproducible.

https://journals.plos.org/digitalhealth/article?id=10.1371/journal.pdig.0000006

--
*not sport specific but still relevant.

Precision strength training: Data-driven artificial intelligence approach to strength and conditioning

https://osf.io/preprints/sportrxiv/w734a/

List of articles on ML / AI in Sport Science.

Answering "it depends" should be our last option as sport scientists. In fact, I prefer to avoid saying it altogether. 

If it really depends on various factors, then everyone in the room knows that already. In practical terms, it is not a helpful answer.

What to do instead?

⬇️

Problem solve
--
Yes - There are multiple factors to consider in every situation.

The athlete, the team, environment, circumstances, timing, goals, etc. To top things up we may only have incomplete information.

Finding the best answer is just a problem solving exercise.

What is the problem?
--
Many times questions are either vague or too complex. If this is the case, then consider the following:

- If too vague, help re-formulate the question. Add context.

- If too complex, try to break it down into smaller pieces.

Better Qs = Better As

Process > answers
--
What do you need to do first? Based on that, what do you need to do next, and so on...

Providing a plan that leads to an answer is the best answer you can give. If you are able to come up with good processes to answer questions you are very valuable.

Evaluate your resources.
--
What do you need to answer the question? Resources are anything from data, network, skills, experience, etc. 

You may need only one resource, or you may need to combine a few.

When in doubt, start with identifying what you need to answer a question.

Level of confidence.
--
I keep bringing this up in many of my threads. We should strive to provide uncertainty (confidence) estimations along with our answers.

For quantitative problems this is relatively easy. For qualitative problems, use your resources to help you with this.

Sort the list of options
--
For those scenarios with multiple possible answers. Don't just throw them at the decision-makers.

Order them, from better to worse (using your resources, judgment,  confidence levels, etc.). 

Provide direction.

Consider "I don't know".
--
If the above steps don't help you provide direction to a question, then just say "I don't know".

You may need to do more research, more time, or you may need to think from a different angle.

Step out. Come back later if the problem is still unsolved.

Job Interviews
--
Unrelated, but apply the above principles during job interviews. Any candidate can say "it depends".

Given a question, show what would you do and what is the process for you to come up with an answer, or some sort of guidance. 

This is what they care about.

Final Thoughts
--
Hope this doesn't come across the wrong way. I know in our environment 99% of the things we deal with are "it depends". 

But how can we minimize using it and continue adding value to the team? 

These are just a few of my thoughts.

“It depends” in Sport Science

Injury Prediction 
-- 
There were a few great discussions here on Twitter about injury prediction over the last few days which got me thinking. 

Here are my thoughts as a practitioner. A set of principles I tend to keep in mind.

Thread
⬇

Basics
--
Don't worry about injury prediction if an athlete doesn't lift twice a week.

Don't kill me yet...It's just a metaphor to make the point. 

Just wanted to get that out of the way before we keep going!

Algorithms
--
The algorithm, model, method or steps applied to the data have important practical implications. I'm not saying we have to go deep into the maths, but we should try to understand the overall workflow of the methods used. This will support our critical thinking.

Re Critical Thinking
--
Part of our job is to disseminate research to inform our practice. Often companies will knock on our door to sell this new injury prediction product.

You can ask your expert friend for help, but you may as well learn and get better at it.

Transparency
--
That links to the previous tweet. Show me the m̵o̵n̵e̵y̵ code. 

There is no secret sauce. Show the code or ask for code. It is often a good idea to get others to have a look.

Is it reproductible? How does it test in different populations? etc.

Simplicity
--
Less is more. Why? Two things come to my mind:

1. The risk of overfitting goes up as complexity increases (there are ways to deal with this I know, but still...)

2. Focusing on less metrics can help to learn faster about our interventions. 

Start small.

Performance
--
We don't need to have a perfect model. There is not such thing. Build your model, evaluate it, understand how it performs so you know how much trust to put on it. Many times, if it performs poorly, you may come up with ways to improve it

Keep iterating.

Machine Learning
--
ML is buzzing. There're some powerful things out there. Majority open-source. All becoming more user-friendly.

Simple first - Fancy later. If a simpler model offers a similar performance (even if slightly worse) stay simple. It'll pay off later.

Causality
--
When something correlates with risk, chances are that manipulating that something will impact the prediction scores. But this is not always the case. Most predictive models don't study causation, yet. Keep this in mind.

The practitioner's judgment is still relevant.

Probability
--
Predictions are not (or should not be) binary. What we are doing is estimating risk (probability distribution), based on the available data. 

All we can do is to add this information to our decision making process and use within context.

Check, Double Check
--
Don't oversell. 

Try to show (and explain) uncertainty.

Most of the time, especially early, it's better to be skeptical. Even if things look good.

Explainability
--
Many models are black boxes.

Banks just want to know whether someone will pay back or not. The reasons behind are irrelevant.

But coaches have to go back to the athlete and say: "so this is what we're doing and why".

Understand how models predict may help.

Keep calm and...
--
The most popular algorithm on Kaggle - XGBoost - is 7 years old. 

TensorFlow was released in 2015. 

The SHAP python library is from 2018. 

Data - still the biggest challenge - is more available and diverse now than 5 years ago.

We are making progress.

Standards
--
The level of confidence is low. The research published so far does not help. It is in general, of very low quality. What can we do?

1. Identify the flaws
2. Keep learning
3. Lean on experts when we have to

This applies to both, researchers and applied practitioners

That's it
--
That's more or less it.

Hope that helps. Very polarized topic. Would love to hear what others use, ideas, feedback, etc...please feel free to share.