I've spent the last couple of days playing around with football betting strategies and systems, making spreadsheets mostly for my own amusement. Geeky, I know.

Anyway, I realised I could apply it to more than just home/draw/away bets and I've taken a shot at a clean sheet prediction equation which I thought might be of interest to someone. Could maybe be useful for choosing between cheap defenders.

The standard theory goes that if you take a Home Teams Goal Form (add the goals scored minus those conceded in the last 5 games) and subtract the Away Teams Goal Form you get a Goal Superiority Rating. By taking historical data (I've used 05/06 to 09/10 seasons) you can correlate the percentage chance of a home win, draw or loss against this GSR. I've used the same method but correlated it against Home Clean Sheets and Away Clean Sheets.

In order to make the graphs below more accurate I trimmed any GSR's with less than 15 supporting historical matches in order to prevent a few results screwing the results due to lack of data. These were just the extremely mismatched games at either end of the spectrum.



As you would suspect, the stronger the home teams advantage the more likely a clean sheet, and vice versa.

The trendlines give us equations (bottom right) to calculate the chance of a particular team getting a clean sheet. These are historically 71% accurate for home teams and 73% accurate for away teams.

For example

Arsenal v Stoke, on Wednesday, 23 February 2011

Arsenal's last 5 home games: 2-0 2-1 3-0 0-0 3-1 (Prem only)
Goal Form: +8

Stokes's last 5 away games: 1-0 2-0 2-0 2-1 0-2
Goal Form: -4

GSR = +8 - -4 = 12

Chance of a Home Cleany (Arsenal):
0.0114 x 12 + 0.3144
= 45% compared to a standard 36%*

Chance of an Away Cleany (Stoke):
-0.0102 x 12 + 0.2916
=17% compared to a standard 25%*

* Over the last 5 seasons Home teams kept a clean sheet 36% of the time and Away teams managed 25%.

As you can see this would be a good match to play your Arsenal defenders and bench your Stoke players.

Any thoughts? Perhaps you think I should spend less time skiving?

When I get home from work I will take a closer look at this - bit of a stats man myself. Good job...

I particularly like this part:



Have more faith, I am sure more than one person will find this interesting...

1. Thank you
2. Cincirollers will be here shortly with his own statistical analysis to compare to yours
3. It's a good effort and interesting. The best part about it is that you used so many seasons for data and not just one or two.
4. The one things that bothers me is, that it seems as if there is massive flactuations of the data results over the trendline, which
reduces the equations' effectiveness. But it is exactly what makes footy so beautiful, giving strength to form and luck of players.

My first impressions are that it makes for interesting reading, but that your method is flawed, and you are introducing irrelevant data. For example Blackpool have never failed to score at home all season (No team has kept a clean sheet against them). Yet their Goal Difference (Supremacy) is -4 over the last 6 games. If you were to look at this you would have to think there wasn't much chance of an away team keeping a clean sheet, but your methodology would suggest that there was a fair chance. You may be better just comparing the goals conceded from one team against the goals scored for the other in some manner.

I'd certainly be interested if you altered your methodology as I currently use a crude method based on what I have described above. I certainly appreciate the time you've put in to this.

Let me know if you come up with something that then correlates to the recorded data in a better way!

edit:
I was just yesterday ruminating on the sad demise of these type of posts on this board, hence my Top 50 post from yesterday. Thanks again for sharing your efforts as it is this that makes (made) this board something I enjoy reading.

I'd be much more interested if you could come up with a system that could make me some £'s, rather than FPL points

For the Arsenal Stoke game Bet365 have priced up clean sheet chances as follows -

Arsenal clean sheet - 5to6 (54.5% chance of success)
Stoke clean sheet - 9to1 (10% chance of success)

comparing the bookies estimation to your system has immediately come up with some significantly different %'s

I'm giving you A for effort, C- for accuracy

I'm off to have a tenner on a Stoke clean sheet, and £10 on Arsenal not to keep a clean sheet (also priced at 5to6).

Interesting post and I will certainly take a closer look at this as I like stats.

Completely agree. Nail, head.

Thanks, for the input guys. Its far from a finished work, more of a proof of concept I put out for some feedback. The theory itself is fairly new to me and I'm still getting my head around it. Keep the comments coming and I'll see if I can turn it into a more accurate calculation.



I'm with you both on this one. Perhaps this method is only useful at working on FT result market probabilities rather than the more intricate clean sheet stats. Do you think the entire concept is flawed?

The way it works at the moment, it works out the probability of any Premiership team in a match with that GSR. I'll take a look at calculating it based on individual teams' historical results in that situation.



Haha, don't blame me when it goes wrong.

Its this bit I would question actually. It isn't so much "Arsenal" that were keeping the clean sheet in 2006, but Clichy, Gallas, Toure, Eboue & Lehman - only one of which is there now, so I'm not just sure how relevant (Clichy, Gallas, Toure, Eboue & Lehman)'s chances of keeping a cleanie then - are relevant now?

Perhaps a Bramernic style current form (last 6 games) may be better?

Or am I missing something?

The strength of the Goal Superiority system is also its weakness. The 'form' part comes from recent games and are individual to each team, however, the equation it is evaluated against is based on a generic question, 'How would an average Premiership team do in this situation?'. This allows for a greater amount of historical data to be used, because while those specific players are no longer at that specific club, players of that standard still exist in teams of that standard today.

However, as has been mentioned the broadness of the equation also means it doesn't cover individual situations (such as Blackpool's guaranteed goals at home this season).

From what I understand this system gives the same weighting to a scoredraw as a 0-0 draw so it's more a % chance of a draw rather than a clean sheet. Maybe include points from the last few home or away games as a weighting?
Good work though!!

I've been rethinking my whole approach in light of exactly this.

The system is setup to predict results not scores, and I've tried to force it into doing something else. I think my basic principle is flawed.

I'm thinking I should refocus onto 'Team A's goals conceded form less Team B's scoring form' and move away from the 'goal difference A less goal difference B' idea all together. That should shift the working towards clean sheets rather than win/draw/lose... at least I hope. I'll keep you updated.

As a side note, if anyone is interested in the equations success at predicting results I'm getting R squared figures (accuracies) of 87% for home win predictions and 88% for away win predictions. See below:

I agree. It seems to be pretty good at what it was meant for, but not as accurate as it could be in predicting clean sheets. Interesting work though nonetheless.

Although I appreciate the confidence, I've found this to be very difficult and so haven't put much effort into it.

My analysis is actually very similar to this but uses the data differently. I also try to avoid using past year's data because of the difference in defenders and have always thought the formation they were playing should also be a factor (but have never tried). And along these lines, I also like the hypothesis someone posted earlier in the year that the holding mid-fielder may play a major role.

The one thing that didn't occur to me until reading this is the self-fulfilling prophecy of the concept of goal superiority. If a team enters a match with a negative rating, they may choose to play a defensive set where they're trying to minimize rather than prevent a goal (while hoping for the clean sheet of course). This puts the team with the positive rating in an attacking position for most of the match and so increases both their chances of scoring and getting the clean sheet.

I think Blackpool actually support this theory because they have publically stated they will attack at all costs and are willing to concede goals in the process (which is why I like them, fantasy-wise).

Just to keep you updated with how I've moved on with this.

I've set the spreadsheet up to calculate the chance of a home clean sheet by taking the goals conceded by the home team in the last 5 home games and adding the goals scored by the away team in their last 5 away games.

By comparing this figure against data from the last 5 seasons I've got a formula that is 85% accurate.

The system works the same for away clean sheet predictions but is less accurate.

I'm planning to try and improve it by weighting it against the number of clean sheets in the last 5 games also but I'll let you know how I get on.

I'll include the current graphs below:

This looks much MUCH better. A far more 'definite' trendline and if it is anything like real life, then it looks to me relevant to the other discussion about Skill V Luck in FPL, only for the EPL teams.

But now it's your chance to take it just one step further and immortalise yourself in the same fashion Robin did with his 'Missing Fixtures' act. If once every GW, you could give us your predictions regarding CSs, you could influence many decision making processes (even if people won't admit to it). *whispering in a sinister voice* Imagine the power!

Anyhow, it looks like something many people here would love to sink their teeth in to. Thanks for the effort.

I think that your R-square is suspiciously high.

Any chance of seeing your spreadsheet with column headings and calculations? Can we download it somewhere?

Feel free to take a look: Clean Sheets Spread Sheet - EDIT: Link updated 3/March/2011

I'd appreciate any suggestions and you'll have to excuse the messy layout, it wasn't meant for anyone but me.

As you'll see this version is currently only based on 4 seasons and is designed to ignore results backed up by 5 or less historical games.

Very interesting thread, much appreciated input guys, thank you.

Is their a spreadsheet available for d/l with the results for 09/10 including players scores, remember seeing one last season for 08/09. Had each player and their corresponding scores against each team home and away..

Been looking all season but always come to a dead end.

I might have done something wrt Player Stats a couple of years ago - can't remember - but if you can describe a bit more what it contained/how it was laid out I should be able to do something as I have all the necessary data. Do you still have the one from 08/09? ... if so you could send me a copy for reference. (contact me via PM and I'll give you an email addy)

This. The graphs are pretty, but I have no idea how to translate them into predictions.

use the formula to give you a clean sheet percentage Shirley?

Hi Romyn, you may know the file as I originally sourced it from FISO



I've uploaded to http://www.tempfiles.net/

http://www.tempfiles.net/download/20110 ... _Full.html

I recognise that ... it's one of mine

I'll put something together for 09/10...

You can easily get clean sheet odds already, from online bookmakers.

dammit why didn't I think of this before...

incidentally does anyone know where I can get an updated 10/11 spreadsheet each week? trying to do something more accessible with all this work...

The next step now is to post predicted results & confidence in each pick. Historically your system looks sound, lets see what happens when it's tested over say 50 EPL games (5 (ish) game weeks)

Next 3 EPL Fixtures are as follows

West Brom V Wolves (Sun 20)
Blackpool V Spurs (Tues 22)
Arsenal V Stoke (Wed 23)

We would be looking for clean sheet percentage chance, predicted winners/draw, predicted goals in the game (for both teams).

From this data you could look at the following bets & see which throws up the best possibility of making a profit (long term)

Match result (H, A, X)
Match score
Clean sheet potential
Predicted goals per team
Goals over/under 2.5

Looking at the 3, I would predict

WBA V Wolves
Draw - I suspect the stats will suggest a home win
1-1 - the stats will lean towards a 2-1 or 2-0 scoreline
WBA 28%, Wolves 20%
WBA 2.3 Wolves 0.6
Over (marginal)

For me no standout bet

Blackpool V Spurs
Away
2-3 - Stats may go with a 2-2 or 2-1 to Blackpool
B'pool 15%, Spurs 15%
Blackpool 2.2 Spurs 1.7
Over

Bets - Over 2.5 goals

Arsenal V Stoke
Home
2-0 - Stats to go 2-0 or 2-1
Arsenal 50%, Stoke 20%
Arsenal 2.2 Stoke 1.2
Under (marginal)

No stand out bet

But if all you are looking for is a guide to which defenders to play in an FPL team on the basis of a likely clean sheet, you just need the relative clean sheet odds of your candidates' teams. For example, suppose we had the luxury of selecting defenders now for the three remaining games of GW27. One web-site (other on line bookmakers are available) lists the clean sheet odds in order from most likely to least likely as Arsenal/West Brom/Spurs/Wolves/Blackpool/Stoke. Of course, this only takes account of defensive performance, so while, for example, in this case Evatt's team is rated more likely to keep a clean sheet than Huth's, other fantasy considerations may lead you to select Huth.