Goal Difference!

[Snowball]

WW misunderstood the basic thrust of my opening post, but never mind.

As an exercise I thought I WOULD test the facts and ask

Is GD in January a good or bad predictor of FINAL Position?

It seems (although I WASN'T saying this) that it IS.


Here is a table showing the GD position (ranking) after the first league game in January
and whether that team made the top 6.

1 9 of 9 times
2 9 of 9 times
3 7 of 9 times
4 5 of 9 times
5 4 of 9 times
6 5 of 9 times

Sides in the top 2 GDs in January (1 game) made the play-offs 100% of the time.

Sides in the top 3 GDs in January (1 game) made the play-offs 93% of the time.
Sides in the top 4 GDs in January (1 game) made the play-offs 83% of the time
Sides in the top 5 GDs in January (1 game) made the play-offs 76% of the time
Sides in the top 6 GDs in January (1 game) made the play-offs 72% of the time

Here are the teams January GD ranking with their final position.

So the top GD early January finished top 6/9 times, second twice, 4th once.

The 2nd team on GD in Jan finished 4-2-2-1-2-4-6-1-2 (average position 2.67)

The 3rd team on GD in Jan finished 3-6-8-1-5-10-2-3-3 (average position 4.5 distorted by that one 10th place, otherwise 3.9)

The 4th team on GD in Jan finished 5-8-3-7-6-8-8-8-2-4 (average position 5.7)

The reason these average positions are lower is because sides from 7th and 8th, and in Sunderland's case nowhere, occasionally come with a late run and push X clubs down a place.

Also, with the top 3 teams there is often a decent GD gap. Teams in 4-5-6-7-8-9-10th are often very close on GD in January

27-15-13-12-10-9-9-8-7-6 at the moment, for example. QPR "clearly best", Watford a little clear of Forest-Reading (idenitcal on GD ppg) but then 10-9-9-8-7 are close to equal.

2001-02 5 of 6 1-4-3-5-7-2
2002-03 5 of 6 1-2-6-8-5-3
2003-04 4 of 6 1-2-8-3-4-15
2004-05 4 of 6 2-3-1-7-10-6
2005-06 6 of 6 1-2-5-6-3-4
2006-07 3 of 6 2-4-10-8-7-6
2007-08 3 of 6 1-6-2-8-10-11-5-4-3
2008-09 4 of 6 4-1-3-2-7-15-9-6-5
2009-10 5 of 6 1-2-3-4-6-11


Last season Blackpool were in 9th in January, but their GD was 5th, I'd say "therefore probably worth higher".
They finished 6th, reflecting their GD with joint 5th GD

[Snowball]

Sides in the top 3 GDs in January (1 game) made the play-offs 93% of the time.
Sides in the top 4 GDs in January (1 game) made the play-offs 83% of the time

That predicts (based on historical data, but the margins are small) that RFC have approximately an 88% chance of making the play-offs

[weybridgewanderer]

We are getting there now! You are beginning to take some basics stats and do some meaningful extrapolition and modelling.

The ones we are particularly interested in are :

The 3rd team on GD in Jan finished 3-6-8-1-5-10-2-3-3 (average position 4.5 distorted by that one 10th place, otherwise 3.9)

The 4th team on GD in Jan finished 5-8-3-7-6-8-8-8-2-4 (average position 5.7)

As our goal difference is 3rd or 4th depending on the measure you take.

Do we know what posotion these teams were in when they were on the third highest

So for example on the "3rd" place, were the teams that finished 8th and 10th also outside the playoff places in january?

Of the teams that finished in the play off places, how many of them were outside the place off places in january?

If the answer to the above is both and none respectively that suggests points at this stage is are a stronger indicator than goal difference

if the answer is both and none, then it suggests that goal difference is a stronger inidcator.

I suspect the answer will be somewhere between the 2, I am genuinely interested to see where. The Blackpool stat is interesting, 9th position with 5th highest goal difference and certainally supports your theory. It may be that just looking at 3rd and 4th yields some level of probability one way or the other, 1st thru 6th completely may give a more accurate understanding.

[Snowball]

I don't understand what you're saying here.

I haven't just looked at 3rd/4th. I've looked at the top 6-10 GDs (in January) and where the ended up

[Snowball]

Here's 4 year's worth


Were 2nd on points (5th on GD) ended up 7th 2001-02 Burnley (GD best predictor)
Were 7th on points (4th on GD) ended up 5th 2001-02 Birmingham (GD best predictor)



Were 05th on points (joint 4/5th GD) ended up 8th 2002-03 Norwich (No winner)
Were 10th on points (joint 4/5th GD) ended up 5th 2002-03 Wolves (GD best predictor)



Were 3rd on points (3rd on GD) ended up 08th 2003-04 Sheffield United (No winner)
Were 5th on points (6th on GD) ended up 15th 2003-04 Preston (GD best predictor)(slightly)
Were 6th on points (7th on GD) ended up 05th 2003-04 (Points edges it slightly)

Crystal Palace went from 18th on points to 6th!! (Madness!!)




Were 04th on Points (04th on GD) ended up 07th 2004-05 Reading (No winner)
Were 06th on Points (05th on GD) ended up 10th 2004-05 Millwall (Points edges it slightly)
Were 09th on points (08th on GD) ended up 04th 2004-05 Derby (GD edges it)
Were 10th on points (12th on GD) ended up 05th 2004-05 Preston (Neither way suggested it)

[Rev Algenon Stickleback H]

Sides in the top 3 GDs in January (1 game) made the play-offs 93% of the time.
Sides in the top 4 GDs in January (1 game) made the play-offs 83% of the time

That predicts (based on historical data, but the margins are small) that RFC have approximately an 88% chance of making the play-offs

If you'd made that the opening post then this would have been a much shorter thread.

The 3rd team on GD in Jan finished 3-6-8-1-5-10-2-3-3 (average position 4.5 distorted by that one 10th place, otherwise 3.9).

you can't really go around rejecting data because it doesn't fit. It stops being true statistics if you do that.

The mode, rather than the median is also more important. The "average position" is pretty meaningless as the more samples you take, the closer you'll invariably get to a perfect correlation.

What you need to find out is how likely a team is to finish in each individual position, working out the probability of each. And you need a lot for than 9 samples to do that with any degree of confidence. The stats above indicate a team is as likely to finish 10th as 2nd, which is unlikely, and has no chance of finishing 4th, which is absurd.

It also counters the detail that the average will be over three because there are only two possible positions above three, and 21 below it.

[Snowball]

Rev there are loads of problems with simple answers like "Cardiff finished 7th"

Sometimes there are wide spreads of data, big points differences, big GD differences, but often the differences are tiny, and really to say RFC have a better GD and a better chance when (say) they are a GD of 10 but there are six teams on 9, is silly.

In most of the cases where the predictabilty has faltered it's been either that the GDs in January were (a) small and (b) close to others, or sides have missed by inches after 46 games.


As for why didn't I post that info at the start? That was NOT my point at the start!

For the 126th time. GD is a very good measure of a team's worth and IMO BETTER than points or position until the last 3rd of the season. Because of that fact I considered RFC's level of quality to be higher than 11th or 8th or even 7th. I consider them to be about 3rd/4th (today) and therefore in with a better shout of the POs than 7th suggests.

[Snowball]

The 3rd team on GD in Jan finished 3-6-8-1-5-10-2-3-3 (average position 4.5 distorted by that one 10th place, otherwise 3.9).

you can't really go around rejecting data because it doesn't fit. It stops being true statistics if you do that.

I know that and I didn't reject the data. But by the same token, if wearing yellow boots meant the team finished top 9 seasons out of 10 and 24th in one season, the average position of 3.3 would be highly misleading

The mode, rather than the median is also more important.

1-2-3-3-3-5-6-8-10

and the mode (most common occurrence) for 3rd best GD early Jan >>> final position is .... 3rd
the median is also 3
the mean 4.55
the mean removing outliers at each extremity is 4.3

The figures are pretty much as expected when latecomers break into the top six

[koran]

1-2-3-3-3-5-6-8-10

and the mode (most common occurrence) for 3rd best GD early Jan >>> final position is .... 3rd
the median is also 3
the mean 4.55
the mean removing outliers at each extremity is 4.3

Removing outliers is good statistical practice but the accepted definition of an outlier is a value that lies more than 1.5 X the inter-quartile range above the upper quartile or below the lower quartile. (My apologies in advance to those who are bored by this!). For this set of data (small though it is) the LQ is 2.5 and the UQ is 7 so the IQR is 7 - 2.5 = 4.5. This gives 1.5 X IQR = 6.75. So outliers are numbers at least 6.75 above the UQ - ie at least 12.75 which does not include 10 - or at least 6.75 below the LQ which gives a negative number and so does not include 1.

My point is that, statistically, removing any results as outliers for this set of data is not justified.

[Snowball]

Not trying to PROVE anything by these stats

but one thing, this being football, every now and then we get totally freak results

Palace were 18th (-7 GD) after 22 or so games and went crazy, finishing 6th

Reading were good relegation candidates, had a fantastic run, looked on for an amazing PO place before finally "blowing"... then there was Sunderland who went from rock-bottom to Champions.

Wolves went from 10th to 5th


But what has surprised me is that "rockets" are quite rare (much rarer than I thought) whereas clubs blowing a great position and dropping like a stone happen more often

Palace went 7th to 14th the same season Coventry dropped from 8 to 20

There was

2nd to 7th
3rd to 10th,
5th to 10th,
6th to 11th,
6th to 15th
8th to 20th
7th to 14th
5th to 8th

[weybridgewanderer]

Crystal Palace went from 18th on points to 6th!! (Madness!!)

Sat 20 Dec 03 LEAGUE DIV 1 CRYSTAL PALACE (H) L: 0-3

At that time Palace were 18th at the time at rank rotten, we were 9th.

That result transformed their season and they went on to finish 6th. We finished 9th.

[weybridgewanderer]

Rev there are loads of problems with simple answers like "Cardiff finished 7th"

Sometimes there are wide spreads of data, big points differences, big GD differences, but often the differences are tiny, and really to say RFC have a better GD and a better chance when (say) they are a GD of 10 but there are six teams on 9, is silly.

In most of the cases where the predictabilty has faltered it's been either that the GDs in January were (a) small and (b) close to others, or sides have missed by inches after 46 games.


As for why didn't I post that info at the start? That was NOT my point at the start!

For the 126th time. GD is a very good measure of a team's worth and IMO BETTER than points or position until the last 3rd of the season. Because of that fact I considered RFC's level of quality to be higher than 11th or 8th or even 7th. I consider them to be about 3rd/4th (today) and therefore in with a better shout of the POs than 7th suggests.

You have now presented theory and some stats that support that theory. this makes some interesting reading and you appear to have quite a good case

Much better than your first post, well done!

personally I still think we will finish 7th, +/- 2 places

[Rev Algenon Stickleback H]

I know that and I didn't reject the data. But by the same token, if wearing yellow boots meant the team finished top 9 seasons out of 10 and 24th in one season, the average position of 3.3 would be highly misleading

No. It would just show one reason why a mean isn't of much use. If 10% of the time you get a "freak" result then that result is perfectly valid. It's where taking a meaningful sample comes in as it's show that either that result was a bizarre exception, or 10% of the time a team would finish 24th. With just 10 samples it's hard to tell. Take 30 or 100 and it'll be rather clearer.

1-2-3-3-3-5-6-8-10

and the mode (most common occurrence) for 3rd best GD early Jan >>> final position is .... 3rd
the median is also 3
the mean 4.55
the mean removing outliers at each extremity is 4.3

The figures are pretty much as expected when latecomers break into the top six

As I said above, you wouldn't expect the mean and median to be anything other than "about three" if you took enough samples. That would be true of pretty much any "successful" stat you can think of, be it goal differnce, goal average, goals scored, goals conceded, possession, number of corners, number of penalties awarded.... They'd all tend towards 3 given enough samples.

Three would obviously be the most likely outcome. The key is how likely, which is where mode comes in.

Working out the mode is 3 doesn't tell you much, but if you can say the odds of finishing in each position is...

1st 5%
2nd 12%
3rd 33%
4th 20%
5th 13%
6th 8%
7th or lower 9%

...then you have something meaningful. You'll need a much better sample than 10 to achieve that though.

You'd need to do that for both end of season and January though.

[Snowball]

Or, possibly, we'll end up in 7th with the 3rd best GD.

If we finish the season with the 3rd best GD we'll finish 3rd or 4th, probably 3rd

If we finish the season with the 3rd best GD we'll finish 3rd or 4th, probably 3rd

[Snowball]

GD is a very good predictor of final position and our current GD OF +10 is twice that of last season's finish.

1 22 +26 +1.18 per game Queens Park Rangers
2 22 +15 +0.68 per game Cardiff City
3 22 +10 +0.45 per game Reading
4 20 +07 +0.35 per game Nottingham Forest
5 21 +07 +0.33 per game Burnley

We had the third-best GD, we finished with the third-best GD

Final GDs

1 46 +39 +0.85 QPR
2 46 +27 +0.59 Swansea
3 46 +26 +0.57 Reading
4 46 +25 +0.54 Norwich
5 46 +22 +0.48 Cardiff
6 46 +19 +0.41 Forest

Note that the differences in GD for 2-3-4-5th are tiny

[Snowball]

And I'm saying, if we continue to improve our GD by .5 of a goal per game we will improve our points-per-game

Not necessarily - "we will more than likely" would have been a better sentence. Example:

Currently 40 points from 26 games (1.538 ppg) with a GD of +13

20 more games to go so by your example we'd be on a GD of +23 at the end of the season.

What happens if we win 5 games 5-0 and lose the other 15 games 1-0?

Our GD is +23 at the end of the season and our points per game is (40+15)/46 = 1.196.


Just saying.

Yup, I got it wrong. We finished on 26, not 23

[Snowball]

Relationship between Goal Difference and League Position


Championship 2010-11


The correlation is .968 or to put it another way 97% predictive of position

The top six teams have the top six GDs. The bottom six teams have the six worst GDs

88 39
84 25
80 27
80 22
77 26
75 19

72 11
68 4
67 14
67 5
65 1
62 0
62 -6
61 6
60 -3
58 -7
56 -11
55 -4
49 -13
48 -25
48 -26
42 -25
42 -35
42 -44

[T.R.O.L.I.]

And I'm saying, if we continue to improve our GD by .5 of a goal per game we will improve our points-per-game

Not necessarily - "we will more than likely" would have been a better sentence. Example:

Currently 40 points from 26 games (1.538 ppg) with a GD of +13

20 more games to go so by your example we'd be on a GD of +23 at the end of the season.

What happens if we win 5 games 5-0 and lose the other 15 games 1-0?

Our GD is +23 at the end of the season and our points per game is (40+15)/46 = 1.196.


Just saying.

Yup, I got it wrong. We finished on 26, not 23

LOL - not sure why you decided to quote that (other than paranoia perhaps) - my post was merely pointing out an example of how getting to a GD +23 wouldn't necessarily improve our ppg.

The fact that we have improved our GD & ppg is irrelevant to my post.

[3 veesinarow]

The top six teams have the top six GDs. The bottom six teams have the six worst GDs

Teams-at-the-top-in-winning-more-games-than-they-lose-and-teams-at-the-bottom-in-losing-more-games-than-they-win-producing-wholly-predictable-goal-differences shocker...I am not one of your critics as your work on stats does produce some useful information now and again, but this certainly isn't one of those useful occasions. Once in a very blue moon, a team at the top will win, say, 25 games by a combined slim margin and lose, say, 12 games by a wider combined margin to give a negative goal difference, but that would be a rare blip to combat the status quo.

[Hoop Blah]

You probably could've got some kind of research grant to cover this amazing bit of insight Snowball!