In my fixtures and team ratings article (link here), I explained my method behind team ratings and how these can be used to produced expected goals in the fixture list. Based on the expected goals I colour coded the fixtures. The slight problem with colour codes is the difficulty in knowing when to switch between one team and another via transfers. This article is based on defence of teams and the likelihood of keeping a clean sheet.
The spread firm I use have produced their view on goals scored in the opening matches and these compare to my estimates as follows:
I would trust a bookie’s estimate over mine (as they know more detail than myself). I hope that as the season goes on our estimates converge. For the sake of the analysis here, I use the bookie values for week 1 and mine thereafter.
Estimating the probability of a clean sheet
For those that didn’t read my article from last year, the number of goals scored in a match follows a Poisson distribution. In the above table, Arsenal have an expect number of goals of 1.4, therefore the distribution of goals they are expected to score is:
So if Arsenal score zero 25% of the time, this means Man City keep a clean sheet 25% of the time. The table below gives the probability of a clean sheet using the Poisson and expected goals v the bookie’s percentage derived from the odds (note Arsenal expected goal is 1.35 so figure slightly different to above):
You’ll note that the bookie has slightly higher percentages than that from using the Poisson and this is because they need to build in a profit margin.
Using the Poisson, we can plot the probability of a clean sheet based on the expected number of goals conceded as:
The chart demonstrates that a small difference in conceding a goal when expected goals conceded is low has a bigger difference on clean sheet probability.
Using the Poisson with the expected goals for each game from my ratings system, we can calculate the probability of a clean sheet for each team through the season. One final check on the accuracy is to calculate the expected number of clean sheet per team for the season and compare them with last season, which is as follows:
Though the individual teams each look reasonable, the total number of clean sheets this coming season is projected nearly 16 higher than last season. I have checked the previous few seasons and the highest number of clean sheets is 232, which leads me to assume this over-estimates the number of clean sheets.
Clean sheets over the season
The problem with looking at match by match data is it can become volatile (home away matches), so the easiest method of looking at the data is to do a rolling average (I have chosen four match average). The graphs below are:
Apart from Man City who is expected to achieve just over 2 clean sheets in their first four games, there isn’t much difference between the other top teams. For Sky players this takes you up to the over-haul. The graph also highlights Liverpool’s tough fixtures starting in game week 4. The graph also suggests that Liverpool and Chelsea never have any huge peaks.
The graphs for other teams can be found in the spreadsheet link below:
To view the graphs go to page “Graphs” and change cell A3 to a number 1, 2, 3 or 4.
This can be used to easily see information about when teams have a good run of fixtures. For example select cell A3 = 2 and you’ll see teams that peak during different periods. Newcastle first peak is point 9 (matches 9 to 12 which are home to Brighton, away to Southampton and home to Watford and Bournemouth), they also peak at 15 and 27. Outside the big five, Everton have the highest expected number of clean sheets in the opening four games (good to Sky players).
In the spreadsheet, the underlying data for the graphs are in rows 34 to 53 and I have highlighted any figures that have an expected number of clean sheets greater than 1.5 in a four match period in green.
There is another page called “correlation”, this has my ratings probability of a clean sheet for each match. I have coloured those with a greater than 0.35 chance in yellow and if greater than 0.4 in green. To the right of these figures I have produced a correlation matrix and highlighted in green any teams that have a negative correlation of less than minus 0.4. This indicates a possible rotation between the teams.
Conclusions for your fantasy team
There is no right or wrong way to use the data as it depends on your strategy to play various games.
My main strategy is to invest heavily in defence. Looking at the charts in terms of clean sheets, I should be after Man City players and indifferent between the other big teams, though am likely to lean towards Man United as their team sheet will be known before the first round of matches.
Another strategy is to go budget on defence to spend money elsewhere. Looking at the second tier of teams Everton and Burnley have good starts and Cardiff from tier 3 teams is another option.
At the overhaul, it would be a time to evaluate.
I believe in going for more premium defenders, but there are options to reduce costs.
If you don’t want to play 5 at the back choose two defenders that rotate well. In my previous article, I gave teams that were paired (eg one always played at home and one away each week). However, there may be room for improvement by looking at teams with high negative correlation, for example, Newcastle (xp 11.15 cs) are paired with Brighton (xp 10.41 cs), though rotating the players for best matches could expect 14.23 clean sheets. However, Crystal Palace (xp 9.99 cs) are negatively correlated with Newcastle and when rotated with them have a total clean sheet expectancy of 14.40
These cs figures can be generated on the “correlation” page and selecting various teams in cells B26:B27
Another option is to use the graphs to identify when teams have very favourable fixtures and then transfer a cheap defender between those clubs. For example, Burnley have an average set of fixtures to start and have a good run starting at game week 5-8 (away to two promoted clubs and home to Bournemouth and Huddersfield). Newcastle start a great run weeks 9 to 12 (three home games to bottom half teams and away to Southampton), etc.
I notice that going through the peaks for the first half of the season, you can jump from Burnley to Newcastle. I checked the correlations and these teams are high negatively correlated. Meaning you could use either a policy of rotation between two players or have one player and transfer between the two clubs to achieve a favourable run of fixtures.
The above is just an example of how to look at the data. The first question to ask yourself is do you want one defender to use transfers and hop between teams or two defenders to rotate.
One option I considered was having a Burnley defender with a Crystal Palace defender (to catch their favourable run from week 3) and then transfer the Place defender to Newcastle before week 9.
If you are considering premium defenders then an option I’m currently considering is 2 Man U, 2 Man C and a Chelsea defender. At match week 4, the Chelsea line crosses with the Spurs indicating a transfer of the Chelsea player to a Spurs defender (looking at the data the transfer should happen in week 6 or 7). At around game week 9, I see that Man U and Spurs are hitting a tough run of fixtures (so will be looking to sell) before buying back in again to catch their next peak.
When viewing the data to aid any decisions always remember, that these are based on my rating system which doesn’t align exactly with the bookie.
The projection of clean sheets is on the high side and tends to over-estimate the number of clean sheets for the teams near the bottom of the table.
Finally, the projections take no account of changes to the club during the summer – either manager or players. Bear this in mind when considering the probabilities.
Sky Sports Fantasy Football
Free league (link): 8366876
£5 League (link): 8366879 Password: FFH5