PDA

View Full Version : Palace and expected goals


victor43
12-10-2017, 04:32 PM
I'm sure a lot of people have started noticing 'expected goals' come up in the media more and more, so with that in mind, and our recent struggles in front of goal, I've written a blog post about how we've struggled and a closer look at Benteke.

There is some good news in there - at least we have been creating the chances! (just a shame no one has taken them and now Benteke is injured...)

Have a read! Cheers

https://andyjohnsonsmagichat.wordpress.com/2017/10/12/cpfc-versus-expected-goals/

Neckinger Eagle
12-10-2017, 04:56 PM
'Expected Goals' is only as good as the model that has been used to generate the statistic.

For instance, one statistician may divide the pitch up in quite a different way compared to another statistician. Also, one statistician may have used many more games to generate his base expectations for each area compared to another, or may have used different leagues' matches compared to another.

After the league match against Huddersfield, somebody was bandying about a statistic that Huddersfield only had an Expected Goals of something like 0.33 for the game. I found that hard to believe considering Joel Ward was virtually on the goal line when he 'scored' for them.

So, a nice pointer, but a lot of supposition, and hardly an empirical statistic.

Edit: what may be of more use is the hard to come by 'Expected Pass Success' statistic, which would filter out defenders poncing about amongst themselves rather than hitting a pass that hurts.

JDawg
12-10-2017, 05:03 PM
When shall we expect some?

ee-ay-ee-ay-ee-ay-o
12-10-2017, 05:05 PM
Yes but if it's the only model that is used and used for each team at least that model is the common denominator and can be used as a useful comparison/ discussion point ?

Neckinger Eagle
12-10-2017, 05:27 PM
But if the model is biased or faulty then it is of no use as a discussion point.

And we don't know if the model being used is biased or faulty unless we evaluate the sample, which none of us have the time to do.

It's the old Charles Reep/Charles Hughes conundrum. They espoused the fact that most goals came from moves of fewest touches, which is quite true. However, their model was faulty and led them to hypothesise that this was the case wherever the move started. This led them to evangelise long ball football. The reason why most goals come from moves of few passes is that most goals come from moves that start close to the opposition goal, and that teams who suffer the most turnovers are more likely to lose games. Suffer a turnover near your own goal is likely to lead to a short move resulting in a goal.

I love a good football stat. Unfortunately all the good football stats have been monetised. The only free stat of value is the final score.

Chas
13-10-2017, 04:16 AM
Can we stay up on "Expected Points" if the actual ones aren't enough?

Bryan
13-10-2017, 05:49 AM
I think this is a candidate for the “most predictable replies to a post” award. Well done everyone!

victor43
13-10-2017, 06:45 AM
So your argument is that is a model is not 100% accurate then it is completely useless?

I realise I will have trouble converting you but the aim of it is to provide a better indication of a teams level than simply just using results. It is there for they same reason people look at number of shots, or shots on target, it gives an indication of how well each team played based on the chances they had. It shouldn't be used in isolation obviously, but it is a factor to consider. Over a large number of games, the teams who perform well regularly in terms of expected goals will pick up more points, more so than if you just looked at previous results or shot numbers.

All it does is try to put an estimate on the value of a chance, effectively exactly what we all do watching football anyway - 'he's got to score that' or 'we battered them today, we really should have won'

No one is saying it is perfect, but it certainly gives more information than just looking at our results this season and concluding 'yep, we are the worst prem team ever, we'll be relegated with less points than derby'.

sw16girl
13-10-2017, 07:14 AM
Pundits love a good stat as do betting companies - it makes punters feel that they are somehow more in control of the situation and therefore more likely to win and therefore more likely to bet. You still never see a poor bookmaker.

And as a side point the statistical analysis falls down because its sample is so small - you cannot make reliable conclusions on a sample of 3 (which is what the article does for Benteke) if it had been carried out over the whole of his career it might provide a slightly better indication but again it is likely to be inaccurate because of other factors affecting his performance over that period which would bring in further variables.

And as a further side point I doubt the raw data provided - I recall one definite header that Benteke has missed this season which I would have defined as "great" - there is no mention of it - how is Scott Dann's header against Burnley not down as "superb".

Interpretation is also subjective - of course Townsend has a large number of "poor" chances - it is a result of where he plays, I suspect a winger is never going to get many great or superb chances because of his position - there is no penny to drop - at least he is having a go and has actually scored as a result - the alternative might have been a pointless tippy tappy pass to a colleague which would have increased his pass percentage but done nothing.

Lego Knight
13-10-2017, 07:18 AM
As long as we keep creating chances, eventually we'll start scoring.

rambo1
13-10-2017, 07:40 AM
As long as we keep creating chances, eventually we'll start scoring.

Absolutely.
If chances weren't being Created,I'd be More Concerned.
Even the Top Strikers Go through Lean Spells,then 1 comes off they're Arse & they Go on a Scoring Streak.

Hector
13-10-2017, 08:42 AM
Benteke would get more chances if he ran a bit more

WorthingEagle
13-10-2017, 09:03 AM
Pundits love a good stat as do betting companies - it makes punters feel that they are somehow more in control of the situation and therefore more likely to win and therefore more likely to bet. You still never see a poor bookmaker.

And as a side point the statistical analysis falls down because its sample is so small - you cannot make reliable conclusions on a sample of 3 (which is what the article does for Benteke) if it had been carried out over the whole of his career it might provide a slightly better indication but again it is likely to be inaccurate because of other factors affecting his performance over that period which would bring in further variables.

And as a further side point I doubt the raw data provided - I recall one definite header that Benteke has missed this season which I would have defined as "great" - there is no mention of it - how is Scott Dann's header against Burnley not down as "superb".

Interpretation is also subjective - of course Townsend has a large number of "poor" chances - it is a result of where he plays, I suspect a winger is never going to get many great or superb chances because of his position - there is no penny to drop - at least he is having a go and has actually scored as a result - the alternative might have been a pointless tippy tappy pass to a colleague which would have increased his pass percentage but done nothing.

The key thing is not the sample size of the chances this season, it's the size of the data used to come up with the expected goals for each scenario. If they've used 20 seasons worth of data, with the position, type and outcome of every chance recorded, it's probably fairly accurate. If it's half a season, not so much.

All it gives you is the likelihood of an average player scoring against an average defence and average goalkeeper in a given situation. Top strikers will outperform the average in the long-run, but of course might go on spells where they can't convert the easiest of chances. Top goalkeepers would keep goals below expected goals over the course of their career.

As for interpreting what is a 'poor, good, great, superb' chance - I would hope they work backwards from the situation to calculate that rather than just saying 'Oh, that looks like a great chance to me'. If they know the position and type of chance, among other variables such as how many defenders are in the way, they can work out the expected goals value based on all the matching records in the data.

Assigning it to a category for the purposes of reporting is then just a case of whether you call 'Superb' anything between 0.8 and 1, or 0.7 and 1 etc.

The overall conclusion the data gives, that we've been particularly wasteful and unlucky in front of goal, and a club like Burnley have been particularly ruthless/lucky and therefore find themselves flying high, seems to ring true to me.

smoll
13-10-2017, 09:06 AM
It's a load of old bollocks. I'm more interested in actual goals

WorthingEagle
13-10-2017, 09:09 AM
It's a load of old bollocks. I'm more interested in actual goals

With Palace, that's very much in hope rather than expectation.

victor43
20-10-2017, 04:10 PM
New post - a look at the numbers behind our first 8 performances, and a look at the similarities with Big Sam's start for us last season

https://andyjohnsonsmagichat.wordpress.com/2017/10/20/no-longer-the-punchline/