I don't understand where the "expected points" basis of his arguments comes from. Exactly what is the algorithm for expected points.
I always look at goal difference as to where a team should be placed - there are inconsistencies in that metric, but I've generally found it to be a good guide.
In a nutshell:
Expected goals for and against is a statistical model for the quality of chances.
It all started when Opta statisticians got a "bright" idea that the quality of chances could be calculated statistically.
So, what they did?
They had a look at their database with all the shots from the big leagues and "found out" that if a shot was close to the goal it had a higher probability to be converted.
They simply calculated how many shots from the exact positions in and around the box had been converted in the past and how many hadn't. Then they said "Ok, now we can apply that to all future situations."
Soon they realised that their calculations are far away from the reality on the pitch. Why?
It doesn't matter whether millions of goals have been scored in the past from the exact position in the box or not - the vast majority of new chances is unique. I'll give you an example. Zlatan's famous bicycle kick from forty yards is a better chance than a kick from six yards if you have seven defenders and a goalkeeper between you and the goal.
So, what happened next? Opta (and others) started fine-tuning their models and adding other parameters:
distance, angle, defensive pressure, type of the shot (kick, header, volley etc.), weather conditions (snow, wind etc.) and other things. They only need to add players haircut to make their models completely ridiculous.
Only recently
Opta added two dimensional goal opening to their calculations as something like the fifth or sixth most important parameter. The trouble is that
football is played in 3D, not 2D.
A couple of years ago,
one of the prominent statisticians, who has developed his own expected goals model, finally gave up and said "It's impossible to precisely calculate the quality of chances. Let's give every shot exactly the same expected goals value." Lol. But he is wrong too. The quality of chances can be calculated.
So, what's wrong with the expected goals models? Nothing wrong with the models except that they are all
fundamentally wrong.
A valid theory can be built on false premises. Same with the expected goals statistical models. They are probably valid, all of them, but at the same time they are fundamentally wrong. In the past the human race have developed some valid theories about the Universe based on the false premise that the Earth is flat. The same with the statistical expected goals models.
Statistics depends on valid data, and if data is crap, then the results from the most sophisticated and valid statistical models will be crap as well.
Apart from chances where a player has an open goal in front of him, which is a rare situation in football, all other
chances are unique. Even penalty kicks are unique, which I am going to explain in a separate thread.
All football players and fans know instinctively what a (good) chance is:
the wider goal is open the better chance is. That's why we talk about "defenders closing down opposition players, blocking shots, throwing bodies on the line, goalkeepers narrowing the angle etc.".
When the expected goals models were introduced I was as excited as many other fans and pundits. Finally we got a model which could measure the real quality of football teams. But how wrong I was! As soon as I started comparing expected goals values from the statistical models with the real life situations on the football pitch I realised that there was
a ridiculous discrepancy between their calculations and the reality.
Is it possible to exactly calculate the quality of chances in football? Yes, but not by using statistical models. What we need is cameras around and above the pitch, we need geometry, and we need a perfect player. As I said, I'm going to explain everything in a new thread.