Thursday 20 February 2014

Splitting Data At Home and Away.

Home advantage or the tendency of a team to produce consistent and significantly improved results at home compared to on their travels, is a recognized fact of virtually every professional sport. The existence of it in football is relatively easy to see. Depending upon the league, a team can expect to have an average goal difference per match at home that is around seven tenths of a goal superior to that that it records away, over the long term.

If its existence is a given, the reasons why a team performs less well away is more contentious. It's steady decline over time in leagues, such as the English top flight, implies that improvement in travel arrangements, that may have been major contributors in the past are having less of an effect currently. Also more detailed analysis has implicated potential factors, such as cowed referees and a desire to protect your home environment as a more primeval cause of home advantage. A factor that was particularly well recognized by home and visiting football fans, especially in the 1970's.

Another widely held cause is the partly synchronized approach where teams are naturally more adventurous at home causing their visiting opponents into a more measured defensive stance than they would usually employ. Even supporters of moderately successful sides eventually tire if their team appears to lack attacking intent on home turf.

If sides find themselves forced into adopting different tactical approaches at home compared to away, we should be able to see this effect in the game by game statistics, in much the same way as goal difference for home sides averages out at around 0.35 of a goal less than that recorded by the visitors.

To avoid one abnormal match skewing averages, for example if a tactically limited team tried to retrieve a situation by attempting cross after cross, I paired each Premiership teams matches by opponent from 2011/12. Therefore, in addition to looking at averages, we are also looking at a consistent tendency under relatively consistent conditions to produce above or below numbers of game events at home compared to away matches.

Stoke in 2011/12 had an inferior goal difference in away matches in 15 of their 19 paired matches compared to the corresponding home match. Three others were tied. For example, the Potters defeated Fulham 2-0 at home, but lost 2-1 in the reverse fixture. Turning a goal difference of +2 in the former into -1 in the latter. That the pattern continued for the great majority of their games makes it unlikely that Stoke were as adept at home as they were away, but the consistently under performing splits occurred through simple random fluctuations over the sample run.

In short, it was likely that the differences were real.

We can further break down the components of City's home field advantage in their last season on their legal, but extremely narrow playing surface. They scored significantly fewer goals away from the Britannia Stadium when comparing paired matches against the same opponent and they allowed significantly more scores on their travels.

For example, they conceded three goals at the Etihad, but just one when Manchester City visited the Brit and scored twice at home to Swansea, but failed to find the net at The Liberty.

Of more interest is if we apply this approach to each teams in game events data. Passes and other recorded events are the means a team attempts to use to create the ratio of goals scored and conceded and these may better describe their tactical approaches during matches. Be it through choice or because their course is partly dictated to them by their opponents.

For example, in 2011/12 (the only year I have this data), Arsenal, based on raw averages passed more frequently at home than away. 590 per match at home compared to 572 on the road. However, they passed more frequently against Norwich at Carrow Road than they had done at the Emirates and this wasn't an isolated event. It also occurred in paired matches against six other opponents.

So, although Arsenal may prefer to pass more often at home, that preference isn't so strong that we can say they do with a high degree of certainty based on this single season. An average of 18 more passes at home may be indicative of such a preference, but combined with nearly 40% of paired games when they attempted more passes in the away fixture, we could also put these splits down to random fluctuation.

We are on much more solid ground in citing a more pass happy home approach for the likes of Villa, Chelsea, both Manchester sides, Newcastle and Norwich. To cite an example, Newcastle played more passes in home matches compared to the away return in 14 out of the 19 paired league games in 2011/12. Averaging 460 home passes against 380 away passes.

So there is perhaps stronger evidence for a different tactical approach in these cases.

Team Longballs. Touches in Opposing Box. Big Chances Allowed. Fouls Won in Danger Area. Cards Shown.
Arsenal. More away. More home. More away. More home. -
Man. U. More away. More home. - More home. -
Everton. More away. - - More home. More away.
Chelsea. More away. - - More home. -
Man. C. More away. More home. More away. - More away.
Liverpool. More away. More home. - - -
Stoke. - More home. - - -
Newcastle. - More home. - More home. -
QPR. - More home. More away. More home. -
Bolton. More home. More home. - - -

This approach takes a season long view of the frequency at which sides attempt to do things on the pitch. By comparing matches against the same opponent, many variables remain relatively constant. The manger, playing staff, make up of the opposition, ageing curves of the players, all should remain fairly stable and the only major change involves the swapping of the venue. So, hopefully any significant change in the data, seen by an inflated or depressed average, occurring over most of the paired matches can be attributed, at least partly, to a different tactical approach being pursued on the road compared at home.

In the two tables, I've summarized significant tendencies for a group of teams and a variety of on field events from the 2011/12 season. A dash represents a difference in home and away data that didn't appear extreme enough for a preference to be attached to that particular side. Their tactical approach at home was broadly similar to the one they employed away for the stats under review.  
Team Total Passes. Forward Passes. Final 3rd Passes. Through Balls. Key Passes.
Arsenal. - - More home. More home. More home.
Man. U. More home. - More home. More home. More home.
Everton. - - - - -
Chelsea. More home. - More home. - More home.
Man. C. More home. - More home. - -
Liverpool. - More home. More home. - More home.
Stoke. More away! - - - More home.
Newcastle. More home. More home. More home. - More home.
QPR. - - More home. - -
Bolton. - - More home. - -

Where tendencies possibly exist, they are generally shared by Premiership teams. A team that creates more touches in their opponents box, always does so at home (a sign of offensive pressure) and increased volume of long balls (to relive pressure) most often happens on away trips.

Inevitably, even with the uniqueness draining from their veins, Stoke managed one more surprise. They played significantly more passes away from the Brit than they did at home. A Stoke side with the ball, being considered by opponents less dangerous than one without it, perhaps.

I haven't listed crosses, but Manchester United supporters may be relived to know that every team, except their neighbours, City cross more frequently at home. Although their side's recent crossing obsession against Fulham was somewhat excessive. Of more concern may be United's apparent ability to diversify their passing options, by venue at least, under SAF, compared to Everton's much less varied repertoire, shown in the table above, under David Moyes,

1 comment:

  1. Why are away teams starting to do better than they did (at least in terms of goal expectancy)?

    The post suggests two answers: 1) better travel arrangements means visiting players are less tired and play better; and 2) teams increasingly set up defensively to avoid conceding, sometimes looking to draw or nick a win on the counter or through a set-piece. This strategy does better on average than playing a more open game.

    There may be value for punters or pundits in considering some other possible answers. For example, 3) the increased visibility of officials' decision-making could impose a greater pressure to be fair, that is to ride out influence from the crowd and players in turning down appeals from the home side. One variable here could be the degrees to which officials progress (get big games, cup finals etc., stay refereeing the same division) depending on whether their judgments are right) and another could be the degree to which associations enforce exactly common standards in all the matches. Punters could look at how good different officials are at withstanding pressure from home fans.

    Another explanation 4) for the improved expectancy differential between home and away games is that teams have come increasingly to discard losing players i.e. players who have shown they're not good enough by repeatedly appearing in losing sides. Whereas previously these players would be unable to improve on their previous results at an away venue (thus their side would be unable to get a better result), now sides are made up of new players without a history of losses against certain opposition. One very suggestive way forward in betting is to forget about team results and to look instead at the results in certain contexts (at certain grounds, at a certain level, against a certain type of opposition e.g. old central defenders) of given players and player combinations. If examining and making predictions on the basis of goal expectancy is a way of turning down the 'noise' associated with random match results, then forgetting team and looking instead at player performance may be another such technique. My judgement is that I have been able to identify pricing anomalies for my own betting using forms of this technique.