Modeling vs Subjective Opinion

[alekaras]

Not sure where this belongs, but since it talks about identifying the right players (and not about football tactics), I figured I'd put it in the Transfers subforum. Here's part 1 of the article and you'll find the link to part 2 under Recent Articles. Talks about how a even a simple model can outdo subjective assessments. I know a lot of people believe that models are as worthless as the screen they're typed on (which in my case would make for a very expensive "worthless" model!), but there have been studies to prove otherwise. There is a degree of subjectivity, however, in how one grades the variables that comprise a model. The inputs to the formulas, in other words, can be subjective. In any case, read away, if you have the time and enjoy the subject.

http://frontoffice.report/how-to-build-a-simple-football-scouting-algorithm-part-1/

 

[Raziel]

Worth a read and makes sense, so many professional players now, need some sort of filter (as first, but not final viewpoint). Reality is probably based on

- Can you define that criteria required = in a system like Poch's? yes
- other variables (example showed injury free time), pace, distance covered, cost bracket, age, current league would probably all go into our system

I would say, subjective opinion in the past (with good scouts) probably did some of that anyway.

Like a lot in football, seems common sense, but so many clubs still make head scratching buying decisions.

 

[Charlie]

Watch the movie Moneyball. The story is about the use of modeling in MLB Baseball. I think the conclusion is that modeling may work well with smaller clubs. However, subjective evaluation and gut instinct are still required at the top level.

 

[alekaras]

Watch the movie Moneyball. The story is about the use of modeling in MLB Baseball. I think the conclusion is that modeling may work well with smaller clubs. However, subjective evaluation and gut instinct are still required at the top level.

I don't see what the difference in club size would have to do with it. The article just states that you don't need an army of data analysts to do any decent modelling, so a small club can do it too, but that modelling beats subjective opinion anytime. If what you said is the case, then there would be no reason for bigger clubs to have entire departments dedicated to this.

 

[scaramanga]

Watch the movie Moneyball. The story is about the use of modeling in MLB Baseball. I think the conclusion is that modeling may work well with smaller clubs. However, subjective evaluation and gut instinct are still required at the top level.

Really? That's not the message I got from it at all.

 

[Raziel]

My impression of moneyball had nothing to do with size of club, it obviously has some value to smaller clubs or teams with less budget as the idea was buying lower cost players with a better chance at success due to evaluating certain statistics/characteristics of the player.

Baseball has a lot of detailed stats to ingest into a prediction model, Football less so and you might have to correlate data to create new stats.

Basic idea is of all of these things is

- How do I buy a player with better odds of success and/or the most potential upside.

 

[SpurMeUp]

Money ball and stats can help identify possible targets. But then you need to watch someone play, and pick up all the subtle nuances that stats don't capture. So it's never the case of one or the other, but ideally a combination of data with subjective investigation. In other words watching a player play football!

 

[scaramanga]

Money ball and stats can help identify possible targets. But then you need to watch someone play, and pick up all the subtle nuances that stats don't capture. So it's never the case of one or the other, but ideally a combination of data with subjective investigation. In other words watching a player play football!

I think more importantly, stats are showing the subtle nuances that escape fallible humans.

 

[Gutter Boy]

I think more importantly, stats are showing the subtle nuances that escape fallible humans.

Yeah. Just a simple example from a different area. The staff in a hospital always complain that they are really busy on the three days before Christmas and that they need extra staff rota-ed for this coming Christmas. But managers have got 20 years historic data from their check-in system showing that this is only the case when a Friday or Saturday falls in those three days, which has happened the past three years, but won't this coming year. They therefore make their staffing decisions based on quantitative over observational data.

The whole point of big data is that it uses past trends to predict future patterns, processing it at a level thousands of humans combined could never do. It will spot things like say players who are 6'1" perform on average 3% better over their careers than players who are 6'2", and that could be calculated into the modelling.

Another example - at some point in the future weather prediction won't be based on observation like they are now. There will be enough data available from past trends to calculate all the components that determine the exact whether at any point and place in time - future or past - like we can already do with star maps.

 

[SpurMeUp]

I think more importantly, stats are showing the subtle nuances that escape fallible humans.

Such as?

 

[scaramanga]

Such as?

Like anything that goes against our natural bias, anything that requires large amounts of data, filtering patterns from randomness, etc.

Our brains are not equipped for that kind of processing, we've evolved specifically not to be analytical by nature - rather to see patterns that aren't there and confirm our own biases.

 

[SpurMeUp]

Like anything that goes against our natural bias, anything that requires large amounts of data, filtering patterns from randomness, etc.

Our brains are not equipped for that kind of processing, we've evolved specifically not to be analytical by nature - rather to see patterns that aren't there and confirm our own biases.

I understand the theory, but what's an example of something you would measure with stats in football to negate such biases?

 

[scaramanga]

I understand the theory, but what's an example of something you would measure with stats in football to negate such biases?

Quite literally everything.

There is nothing your brain records in memory as an event that doesn't pass through a perception filter first. Every point of light that hits your retina is passed through a filter to ensure that it fits what you expect to see. If it doesn't, your brain either makes something up or rejects what it sees before it even hits memory.

If you want truly unbiased, unadulterated facts you have to take the human brain out of it as much as possible.

 

[Gutter Boy]

Quite literally everything.

There is nothing your brain records in memory as an event that doesn't pass through a perception filter first. Every point of light that hits your retina is passed through a filter to ensure that it fits what you expect to see. If it doesn't, your brain either makes something up or rejects what it sees before it even hits memory.

If you want truly unbiased, unadulterated facts you have to take the human brain out of it as much as possible.

There's even arguments being made for removing human witnesses and juries from court cases (and replacing them with only scientific evidence and expert juries), because both are ridiculously unreliable.

The human brain shortcuts - it remembers a couple of things and then invents the rest to fill in the gaps. While juries have been shown to heavily bias taller people and those with more symmetrical faces, far more than anything that's actually said by anyone.

 

[scaramanga]

There's even arguments being made for removing human witnesses and juries from court cases (and replacing them with only scientific evidence and expert juries), because both are ridiculously unreliable.

The human brain shortcuts - it remembers a couple of things and then invents the rest to fill in the gaps. While juries have been shown to heavily bias taller people and those with more symmetrical faces, far more than anything that's actually said by anyone.

You're right, witness statements are pretty much worthless. Any decent lawyer can use the fact that memory alters over time to make it appear as if a witness is altering their statement and make them appear less credible.

It will be some time before we can remove "spotters" from the collection of data, but until we can, reducing human input will be the best we can do.

 

[SpurMeUp]

Quite literally everything.

There is nothing your brain records in memory as an event that doesn't pass through a perception filter first. Every point of light that hits your retina is passed through a filter to ensure that it fits what you expect to see. If it doesn't, your brain either makes something up or rejects what it sees before it even hits memory.

If you want truly unbiased, unadulterated facts you have to take the human brain out of it as much as possible.

Having trouble finding an example? Of course perception and bias are massive, placebo effects, self-fulfilling prophecies etc are well documented and show human judgment is clouded by many things. But we're not discussing psychology, we're looking at how statistic can inform on scouting players and how it fits or improves upon subjective player analysis.

What stats might a modern scout look at which would cut through bias? If its all theory and no practice then we have one answer to the original question.

 

[scaramanga]

Having trouble finding an example? Of course perception and bias are massive, placebo effects, self-fulfilling prophecies etc are well documented and show human judgment is clouded by many things. But we're not discussing psychology, we're looking at how statistic can inform on scouting players and how it fits or improves upon subjective player analysis.

What stats might a modern scout look at which would cut through bias? If its all theory and no practice then we have one answer to the original question.

The problem isn't finding an example, it's restricting it to one - literally everything that happens on a pitch would be better analysed with data than opinion.

 

[SpurMeUp]

The problem isn't finding an example, it's restricting it to one - literally everything that happens on a pitch would be better analysed with data than opinion.

Everything and nothing. There is so much, we can't think of one thing. Blinded by the enormity of it?

 

[scaramanga]

Everything and nothing. There is so much, we can't think of one thing. Blinded by the enormity of it?

Absolutely.

So let's start from the simplest first and work our way forward. Scouting a player that is 193cm is better than scouting a player that is "quite tall". Scouting a player who creates 3 open play chances and 1 assist per 90 is better than scouting a player who "is good at passing". The list is literally endless.

 

[SpurMeUp]

Absolutely.

So let's start from the simplest first and work our way forward. Scouting a player that is 193cm is better than scouting a player that is "quite tall". Scouting a player who creates 3 open play chances and 1 assist per 90 is better than scouting a player who "is good at passing". The list is literally endless.

So in other words, you would use stats to identify something, but then 'scout' them. Would this be watching them play by any chance?