Are we in a false position? Expected goals

[brady1993]

21 hours ago, Ellafella said:

I haven't looked at the guts of this model in detail. I will do and I will add anything that @brady1993 hasn't picked up but it sounds to me that the above is a good and thorough explanation of what this model purports to do. It basicially says, on the balance of possession and chances created / conceded, in a given space, we expect any team to score this many goals in a game and concede this many. It then adjusts "results" accordingly and re-calculates the table. It is a theoretical model. Does it predict "reality"? Yes, up to a point, but no way near 100". Does it mean we are in a "false" position? No. We are 2nd.

The whole thing is based on a "possession-based" football model. But we don't play possession-based football. 

I posted some charts from Ben Mayhew's Experiental 3-6-1 site the other day. I didn't really make any interpretative comments because I wanted others to pick out the issues; but there are many pointers to this "Expected Goals" issue; indeed Ben Mayhew also has an "Expected Goals" chart. I've re-produced one of the charts below:

This chart gives a cross-tabulation of a simple statistic. It shows the relative number of shots taken {at the opposition's goal} compared with the number of shots faced {at the team's goal they're defending}. What the chart does, is to show that teams at the top of the table (eg Wolves) tend to have relatively more shots at the opposition's goal, compared with how many the opposition have at their goal per game. The teams who cluster in the green space are the ones at the top of the table - mainly. It's possible then to "describe" the distributions of quadrants which are labelled "Busy attack, quiet defence" (teams at the top of the league), "Quiet attack, quiet defence" (teams just outside the top), "Busy attack, busy defence" (teams like Florist...mid-table but with poor defence, can score but concede a lot) and "quiet attack, busy defence" (teams fighting relegation". All the Championship teams are distributed in this 2-dimensional space and when you look at each team, you tend to think "yes, that's where you'd expect Cardiff  City to be". But there are 1 or 2 outliers that make you think, "how come they're there?" given where they are in the League?

For me, those outliers are:

Brentford, Ipswich and Derby County.

I tipped Brentford to be the late-coming "surprise team" and to get a point at Wolves last night (they lost 3-0!). Ipswich and Derby County both have something in common...they don't play possession-based football. So the model doesn't handle those. In fact, the point of showing this chart above is that according to it, Derby are very average and appear at the intersection of the cross-hairs - they take an average number of shots compared with all other Championship sides, and they face and average number of shots compared with all other championship sides. 

It does really boil down to quality of finishing or at least the quality of the possesssion. As @brady1993 explained, their xG model takes no account of defender proximity and last ditch defending. If it did, it would have predicted Derby to be in the top 2.

So, will we continue to do well? Yes, in all probability, although it does seem to be that a 3-5-2 formation can very much strangle "Rowettball". 

Cracking post. 

If you plan on delving into that model further let me know as I'd be very interested to see the mechanics of it. Looks a really cool idea (if albeit a bit misleading)

On 352 strangling us; I think we could beat it but itd require Rowett to gamble a little. Right now we play completely into the systems hands, offensivelt we play narrowly and don't push our full backs forward so we can't exploit the gaps out wide and defensively we bring the wide players right back on defensive leaving them with plenty of players (opposition cbs mostly free to step into midfield) around the defensive midfield area to outman us and control the game.

If it was me id encourage our wide forwards to stay forward most of the time, defensively encourage them and the centre forward to cut out the passing channels back to the defence to stop them recycling possession easily. Offensively look to transition quickly to enable a 3 on 3 with the opposition defence with the space available out wide to peel off into to pull the centre backs out of position.

[FindernRam]

Football is a series of discrete events: Tackle, pass, shot, each of which has a probability of success or failure for each player doing that action so in theory you could write a programme to predict the outcome of each game. Basically roll the dice a few thousand times and come up with a predicted result and confidence level. I suspect that's how a certain management game does it, and probably the BBC predictor and most betting companies. Of course if it was that simple we'd all clean up at the bookies.

The problem is not only that each action has a probability (not certainty) but there are secondary levels of influences- weather, form, injury status, management instructions, decision making, emotional level, etc. These all screw up the purity of the numbers and render even the most sophisticated analysis pretty much meaningless. So Xgs --rubbish!

In another life I was involved with a project to model machine reliability and with masses of highly precise historical data we still had a significant margin of error going forward.  

 

[Comrade 86]

1 hour ago, HantsRam said:

This is possibly Cardiff reverting to THEIR mean.

So logically, the same could happen to us. In fact, the statistics are suggesting that it is more likely we lose the next 4 than win the next 4, if we continue to play the same way. Note....."likely", not "certain".

I understand what the stats are suggesting @HantsRam I just beg to differ based on an alternative view of the many metrics one could apply. Perhaps we'll crumble and fall down to 13th as said stats foretell, but I think it's rather unlikely this time around.

[CumbrianRam]

Found a stat that is finnally more misleading than "assist" stats.

[Ellafella]

51 minutes ago, brady1993 said:

Cracking post. 

If you plan on delving into that model further let me know as I'd be very interested to see the mechanics of it. Looks a really cool idea (if albeit a bit misleading)

On 352 strangling us; I think we could beat it but itd require Rowett to gamble a little. Right now we play completely into the systems hands, offensivelt we play narrowly and don't push our full backs forward so we can't exploit the gaps out wide and defensively we bring the wide players right back on defensive leaving them with plenty of players (opposition cbs mostly free to step into midfield) around the defensive midfield area to outman us and control the game.

If it was me id encourage our wide forwards to stay forward most of the time, defensively encourage them and the centre forward to cut out the passing channels back to the defence to stop them recycling possession easily. Offensively look to transition quickly to enable a 3 on 3 with the opposition defence with the space available out wide to peel off into to pull the centre backs out of position.

Thank you @brady1993 For your kind comment. Perhaps we can share our thoughts on this and other stats stuff? 

[brady1993]

Just now, Ellafella said:

Thank you @brady1993 For your kind comment. Perhaps we can share our thoughts on this and other stats stuff? 

Sounds great to me. It's an area I'm actively looking to delve more into and was planning on seeing what football data I could source in the near future.

[Ellafella]

19 minutes ago, 86 points said:

I understand what the stats are suggesting @HantsRam I just beg to differ based on an alternative view of the many metrics one could apply. Perhaps we'll crumble and fall down to 13th as said stats foretell, but I think it's rather unlikely this time around.

I gather Cardiff have no fit midfielders left rather than a statistical "regression to the mean" issue going forward. 

[Ellafella]

8 minutes ago, brady1993 said:

Sounds great to me. It's an area I'm actively looking to delve more into and was planning on seeing what football data I could source in the near future.

That's great. Btw, following a discussion about possession and the relationship between it and winning, I ran a quick correlational analysis using prem results from Dec 8th and Champ results from 16 Dec. Remarkably the outcome was an r of -0.5! Possession was negatively correlated with 3 points. I checked the data and there were a couple of outliers {including Man Ure's result at Arsenal on 22% possession} but I'm staggered that possession has become such a non-factor in contemporary football! 

[The Scarlet Pimpernel]

There is a huge amount of difference in the nature of a shot. Are shots that are blocked counted? Assuming not our brilliant defence keeps teams in the main taking speculative long(ish) shots in desperation. Our shots taken are not speculative long shots but very realistic efforts. How is that massive difference calculated in the graph?

[Comrade 86]

24 minutes ago, Ellafella said:

That's great. Btw, following a discussion about possession and the relationship between it and winning, I ran a quick correlational analysis using prem results from Dec 8th and Champ results from 16 Dec. Remarkably the outcome was an r of -0.5! Possession was negatively correlated with 3 points. I checked the data and there were a couple of outliers {including Man Ure's result at Arsenal on 22% possession} but I'm staggered that possession has become such a non-factor in contemporary football! 

Fester?

31 minutes ago, Ellafella said:

I gather Cardiff have no fit midfielders left rather than a statistical "regression to the mean" issue going forward. 

The point I was trying to make is that I don't necessarily give tremendous weight to either argument. Stats are useful in this regard for forecasting and trying to analyse performance but that's where it starts and finishes, for me at least. We'd all be retired and nailing 10 team accas with Ladbrokes every week otherwise.

[brady1993]

1 hour ago, Ellafella said:

That's great. Btw, following a discussion about possession and the relationship between it and winning, I ran a quick correlational analysis using prem results from Dec 8th and Champ results from 16 Dec. Remarkably the outcome was an r of -0.5! Possession was negatively correlated with 3 points. I checked the data and there were a couple of outliers {including Man Ure's result at Arsenal on 22% possession} but I'm staggered that possession has become such a non-factor in contemporary football! 

That's actually really surprising and interesting. I wonder if it holds up over the entire season. My guess would have been that it'd have a general positive correlation with the curve actually taking something of an s shape i.e. much more likely to lose at very low possession levels, much more likely to win at very high levels but flatter in between.

[brady1993]

1 hour ago, Ellafella said:

That's great. Btw, following a discussion about possession and the relationship between it and winning, I ran a quick correlational analysis using prem results from Dec 8th and Champ results from 16 Dec. Remarkably the outcome was an r of -0.5! Possession was negatively correlated with 3 points. I checked the data and there were a couple of outliers {including Man Ure's result at Arsenal on 22% possession} but I'm staggered that possession has become such a non-factor in contemporary football! 

Interestingly just found this from today:

https://amp.theguardian.com/football/blog/2018/jan/04/premier-league-possession-manchester-city

[KCG]

Are we in a false position?

No!

[Topram]

To be fair we haven't preformed that well this season in terms of attacking football but the team spirit and effort from defending from the front has us in a deserved prostion we didn't start too well and yes teams have missed a fair few chances but at Cardiff away I thought they were terrible but they are still 3rd yet I thought Bolton were one of the worst sides I'd seen yet they have won a few since! Championship is about team effort and you need that luck, the luck we haven't had over past few years! Ok yes we've had a few descions go our way (dirties away) but it evens itself out! 

[Ellafella]

47 minutes ago, Topram said:

To be fair we haven't preformed that well this season in terms of attacking football but the team spirit and effort from defending from the front has us in a deserved prostion we didn't start too well and yes teams have missed a fair few chances but at Cardiff away I thought they were terrible but they are still 3rd yet I thought Bolton were one of the worst sides I'd seen yet they have won a few since! Championship is about team effort and you need that luck, the luck we haven't had over past few years! Ok yes we've had a few descions go our way (dirties away) but it evens itself out! 

Good Lord..I'm now out of breath. 

[RoyMac5]

1 hour ago, brady1993 said:

Interestingly just found this from today:

https://amp.theguardian.com/football/blog/2018/jan/04/premier-league-possession-manchester-city

The article suggests this, which I think is quite pertinent:

"In part that is because it is easier to hold possession now. The liberalisation of the offside law makes it much harder for teams to set their basic defensive line high up the pitch, increasing the effective playing area and so making the midfield less congested. With referees far more inclined to show cards, it is much harder now than it was even 15 years ago to bully creative players out of the game. The result is an era in which the likes of Xavi, Luka Modric and David Silva have thrived."

[PistoldPete2]

If we have clinical finishers how does that make us in a false position? Not that I woudlnt expect villa to catch us unless we improve our performances.   

[Van der MoodHoover]

2 hours ago, 86 points said:

I understand what the stats are suggesting @HantsRam I just beg to differ based on an alternative view of the many metrics one could apply. Perhaps we'll crumble and fall down to 13th as said stats foretell, but I think it's rather unlikely this time around.

I wrote in an earlier post that the expected value of 13th place only made any sense if you were looking at a season as a whole, from outset. To achieve 13th from where we are NOW, we would have to collapse to such a degree that our form for the rest of the season would have to be Burton-esque.

If we hit these expected levels for the remainder of the season we will finish 6-7th is my estimate.

[BobbyD]

Was it yourself who did a lengthy post a short while ago using detailed stats to explain why we had no chance of overtaking Bristol or especially Cardiff and that we would run out of games by the end of the season before we could achieve that ?

And then two weeks later we were ahead of both ?

Kind of sums up the danger of using irrelevant or peripheral statistical  information to make predictions and forgetting to inject some good old common sense

Although if it provides an interest to some, then fair enough I guess

For me half way through the season is long enough to suggest the league table is the only stat that matters as to whether or not we are in a false position - so, clearly, no.

[Van der MoodHoover]

1 minute ago, BobbyD said:

Was it yourself who did a lengthy post a short while ago using detailed stats to explain why we had no chance of overtaking Bristol or especially Cardiff and that we would run out of games by the end of the season before we could achieve that ?

And then two weeks later we were ahead of both ?

Kind of sums up the danger of using irrelevant or peripheral statistical  information to make predictions and forgetting to inject some good old common sense

Although if it provides an interest to some, then fair enough I guess

For me half way through the season is long enough to suggest the league table is the only stat that matters as to whether or not we are in a false position - so, clearly, no.

Not me feller.

I agree that there needs to be very clear and careful communication about what you are and are not saying, when you use stats to frame ANY prediction. It's my profession (actuary).

I also agree that arguing whether or not we are in a "false" position is facile. For me, I would be limiting the stat's use to giving an indication (note NOT a prediction) of whether the current position is - on balance - likely to be maintained or not.

FWIW I would NEVER have concluded that we could not catch teams 6 points ahead of us. Its not impossible to catch Wolves. But as the games go by and the gap doesn't close, then it becomes less and less likely.