Same Scooby Snacks ... New Location, Blog Entries are soooooo 2009 Options

[scoobyliscious]

Home GF	Home GA	Home Team	Exp. Score	Exp. Score	Away Team	Away GA	Away GF
3.58	0.74	Chelsea	3.75	0.87	West Brom	1.95	1.16
2.74	0.63	Man Utd	2.65	0.50	Newcastle	1.68	0.89
2.26	0.79	Liverpool	1.86	1.60	Arsenal	1.37	1.84
2.11	0.63	Tottenham	1.65	1.29	Man City	1.32	1.68
1.68	1.00	Sunderland	1.70	0.97	Birmingham	1.79	1.00
1.58	2.00	Wigan	2.12	1.60	Blackpool	2.32	0.63
1.53	0.84	Aston Villa	1.70	0.71	West Ham	1.95	0.89
1.47	0.95	Blackburn	1.17	1.24	Everton	1.47	1.32
1.37	1.63	Bolton	1.22	1.24	Fulham	1.63	0.63
0.68	1.16	Wolverhampton	0.33	0.66	Stoke	1.42	0.53

Welcome to Scooby's Snacks, a semi-regular "feature" that is moving from my Blog to a (hopefully) weekly pinned Topic. The Snacks are a way to use the most recent table as a way of predicting the results for the week.

What I've decided to do is show you this week's games based upon last seasons' "stats" and to explain the "methodology" again. In the table above, each row represents 1 game in the upcoming week. For example, the first row represents West Brom's visit to Stamford Bridge. The table has 8 columns:
1.Home GF: this column lists the average goals per game scored by the Home Team (Home Goals For)
2.Home GA: this column lists the average goals per game conceded by the Home Team (Home Goals Against)
3.Home Team: this column lists the Home Team for this particular game / row
4.Exp. Score: this column is the expected number of goals for the Home Team (more on how I derive this number later)
5.Exp. Score: this column is the expected number of goals for the Away Team (more on how I derive this number later)
6.Away Team: this column lists the Away Team for this particular game / row (so, the middle 4 columns are Home Team: Score: Score: Away Team, allowing you to quickly see what I predict is the score)
7.Away GA: this column lists the average goals per game conceded by the Away Team (Away Goals Against)
8.Away GF: this column lists the average goals per game scored by the Away Team (Away Goals For)

Furthermore, the first row represents the most goals scored by a home team since that is likeliest to be the best attacking team in any given week. Results are then sorted by expected goals scored for the home team and further by expected goals conceded by the home team.

Now that you know how to read the table, I'm sure you are wondering "but how do you predict these scores?" That's a very good question. First, I calculate a theoretical league-wide "average score" for a particular venue (home or away). Essentially, it's the total number of goals scored (either at home or away) divided by the total number of games played. Amazingly enough, these numbers have remained fairly consistent until last season, when the number of goals scored increased. Usually, with home teams averaging about 1.5 goals per game and away teams averaging about 1.0 goals per game. Then, I calculate each team's difference from that average.

For example, last season the "average" number of goals scored by a home team was 1.78 goals per game. Tottenham, however, scored 40 or 2.11 goals per game, so their "difference from the average" is 0.33 goals per game better than average. Meanwhile, Man City conceded 1.32 goals per game on their travels, meaning they are 0.46 goals better than average (note that the number of "average" goals for home teams scored and away teams conceded is the same value). Now, I combine those 2 "differences" into a single figure, meaning that, combined, Tottenham at home to Man City are a total 0.13 goals per game to the traveling team away from the "average". I then add this calculated difference to the average number of goals scored by the home team (remember that? It's 1.78) to arrive at a total value for expected goals: 1.65.

In other words: I figure out how much being at home helps (or hurts) a team and combine that with how much being away helps (or hurts) their opposition. Then, I add that number to the theoretical average and call it a prediction.

Sure, it's a complicated method, but it's what I came up with. The most important element to the "methodology" is that, while it isn't a true statistic (I don't normalize, calculate variance, or provide a confidence interval), it does manage to show you what happens when a good offense plays a bad defense. It also helps to identify surprising predictions, like Wolves terrible home attack matched up against a slightly-above-average Stoke away defense resulting in the week's best prediction for defenders in Stoke only conceding 0.33 goals (and that on the road).

A couple of quick caveats:

1.Remember that these values are an average of several averages, without any regard for variance. In other words: they are just some reasonable numbers, but shouldn't be thought of as actual accurate predictions. They are pretty good, but by no means perfect. Take them with a grain (or a few thousand grains) of salt ...
2.Because I am not performing any actual analysis of variance, outcomes that are close should be considered the same. So, when Wigan is expected to score 1.53 goals home to West Ham and Blackburn are expected to score 1.47 goals home to Everton, those calculations are virtually the same.
3.Speaking of variance, I would be remiss if I didn't point out that, early in the season, we don't have a lot of data to go by. However, as the weeks go by and the data is updated (I will hopefully be posting the table every week, give or take), the accuracy increases.
4.These values refer to the TEAM and not to a particular player. So, it might be nice to say that Chelsea look like they will score many goals against West Brom, but that doesn't imply that Anelka will be the one scoring them. For all you know, he could not have a shot on target and see a yellow card. In other words, these tables can help you identify which matchups to take advantage of, but they won't tell you which players from those matchups are likeliest to perform.
5.This table does not calculate form, either. I am trying to track the increase or decrease of the "rank" values in this table over the course of the season (so we can all see when Liverpool's 16th ranked away defense has begun to climb back up to the top half of the table, and take advantage of that improvement by selecting Reina on the road to Blackburn) and may publish those results (I just figured out tables, and now we want to move onto charts?) at a later time. Still, beware of putting too much stock in these values if a team has hit a rich vein of form, like Portsmouth at the end of the '05-'06 season. Their numbers were horrible (they finished the season as the number 19 home offense), but were playing so well that you really should have had a Pompey player in your lineup for many of those games.
6. Speaking of West Brom and season averages and all of that: the table I have used contains "projected" per game values for the promoted teams, based upon a comparison of previously promoted teams goals scored in their promoted season to their Premier Leagus season and, of course, how each team performed last season.

And, lastly, please remember these are last season's numbers. Teams have changed personnel. Teams have changed managers. Teams have changed tactics. Surprises happen.
Good luck for the season!

[Elyzrin]

Quite possibly one of the dizziest posts I've seen on S11....

Can I just clarify though, you say you've based the table on the whole of last season's stats. As the weeks go by, from what I understand, the new week's results will be averaged onto the table. When that happens, will your base change as well, i.e. the total number of matches the average is derived from increases as the season goes on? Or will you be removing the oldest stats from your calculations?

Or maybe you've mentioned this somewhere already and I just didn't see it

[BRFC]

It's happened. I got my wish. I can now die a happy man. The "Scooby Snacks for Sticky" campaign I have been running since '97 or so is over.


Now what to do with my life?

Excellent stuff Scooby. This is the tool I use most often to balance out the gut with the brain. Many thanks for all the time that goes into the stats

[scoobyliscious]

Quite possibly one of the dizziest posts I've seen on S11....

Can I just clarify though, you say you've based the table on the whole of last season's stats. As the weeks go by, from what I understand, the new week's results will be averaged onto the table. When that happens, will your base change as well, i.e. the total number of matches the average is derived from increases as the season goes on? Or will you be removing the oldest stats from your calculations?

Or maybe you've mentioned this somewhere already and I just didn't see it

I will use last season's stats for the first few weeks, usually until I have multiple games at each venue per team (i.e. until all teams have played at least 2 home games and 2 away games), at which point I will drop last season and only use this season.

So, the table I currently have will determine the first few weeks until we have enough data from this season to begin to make projections. Which means that early-season projections have one of two faults: they are based on data that may not be relevant (last season) or they are based on a small sample size (this season).

[especialc]

Now that you spent so much time coming up with this time consuming oracle, it would be interesting for you to post a prediction list for the first games based on your number crunch and see the percentage of right answers. Can you do the lottery numbers plz Scoobs?

[especialc]

Great piece of work. Thanks

[straycat]

Scoob, you rock. I always look forward to your analysis. Keep up the good work!