One of the challenges when trying to make sense of the course a soccer game will take is to account for the changing dynamics that occur during the match.Unlike more established statistically investigated sports such as baseball and American Football,soccer has a partial dearth of numbers to crunch.Therefore,it's a useful exercise to look at the methods used to model more data rich sports and see if comparisons can be made with situations that arise in soccer.
The NFL has a much more structured game dynamic compared to the more fluid situations that exist in soccer.Every NFL play is recorded in the play by play data.Down,distance to go,field position,time remaining and score is available for each game played over the entire duration of the modern NFL,so it's a fairly simple matter to build up an extensive database for most of the available field position combinations.For example,the most common position NFL teams have found themselves in over the seasons is a first and ten from their own 20 yard line (it's the re start position from a touchback).The average outcome from a team starting a drive from their own twenty yard line nets them just over three tenths of a point,they've got about a 70% chance of getting another set of downs and around a quarter of the time they'll score points,either through a touchdown or a field goal.That's an impressive set of conclusions and undoubtedly of interest.However the outcomes are the result of a whole range of diverse matchups from poor defenses taking on elite offenses,to the best verses the best and any combination in between.In short the average outcome is not team specific.You can use this type of data to say what happens on average,but that average is likely to underestimate by some distance the likely outcome when a New England team at their very best takes on a Tampa Bay side vying for the number one draft pick.
Therefore if we use purely historical data as a way of predicting the likely course a sport of any kind will take we are almost certainly going to present a broad representation that inevitably lacks detail.So the first thing we can learn from looking at attempts to model other sporting contest is the importance of the need to account for the differences in overall team strength that exists between the competing teams.Knowing that all NFL teams who found themselves facing a first and ten from their own 20 averaged one touchdown in ever seven such drives tells you little about the number of touchdowns a fully strength Indianapolis side could expect to gain against the worse pass defense in the league.Similarly knowing that EPL sides who lead by a goal at half time won in 70% of games doesn't help you too much when deciding how Wigan would fare if they were fortunate enough to find themselves in that situation at Old Trafford.
Lesson one,match ups matter.
The second lesson we can learn from looking at attempts to model the NFL requires a little more understanding of the intricacies of American Football.The NFL is currently a passing league.Rule changes to make the game more attractive to TV audiences have made defending the pass much more difficult and the average gain made by passing the ball is around 50% greater than the gain to be made by running it.This discrepancy more than makes up for the increased risk of turning the football over to you opponent by going aerial.It also means that there is a reasonably well defined dynamic to the game.Scoring is more likely when you throw the ball,but the risks that go with this approach in terms of incompletions and interceptions also increase,so when a team trails they adopt a strategy of passing more.Alternatively,if a team is protecting a lead they take a more cautious approach by running more frequently than they do if they are trailing.They are less likely to score as freely as they would have done if they'd thrown the ball,but running as well as keeping the clock ticking and hastening the end of the game,also reduces the risk of turning the ball over.In short,teams are more adventurous when behind and more cautious when in front.This shift in strategy is readily observed in the play by play data,but because no comparable information exists in soccer and because of the relatively low scoring nature of the game,it's difficult to observe a similar tactical shift in soccer.
The most potent reason why game strategy is less extreme in soccer than other sports is the elevated likelihood that the game will currently be tied or will end up being tied.Scoring events in the NFL are more frequent compared to soccer,an average game of gridiron has 10 scoring events,while soccer has less than 3.Also a goal in soccer is always worth the same one goal,but a score in the NFL can add 2,3,6,7 or 8 points to the scoreboard.An NFL team is much more likely to be either behind or ahead and that game state will significantly dictate how adventurous or otherwise their play becomes.By contrast soccer teams are frequently tied for significant periods of a match and that can be a game position that both teams can find find themselves reasonably happy with.We may intuitively guess that one team is putting more effort into scoring than they previously had done,but we may simply be seeing the perfectly normal ebb and flow of a game.
If we want to see whether teams adjust game strategy and become more adventurous when behind and more cautious and defensive when in front we really need a game where the draw is of no value to one of sides and then we should expect to see a polarisation of alternative strategies being played out for the entire 90 minutes dependent upon the current scoreline.The trends will be easier to spot.Which brings us neatly to Stoke's Thursday night's Europa League game five encounter with Dynamo Kyiv.A draw or win guaranteed Stoke qualification for the knockout stages,but Kyiv almost certainly needed a win to keep qualification in their own hands.
Stoke City 1 Dynamo Kyiv 1.
Stoke with homefield advantage were the better team pre game,but they sent out a team that was more
equipped to defend by including two holding midfielders and they also included both of their longthrow experts at the expense of attacking width.Pre game radio quotes that Stoke would not play for a draw also alerted seasoned Pulis watchers to the likelihood that that was exactly what they would do.
For 30 minutes Stoke were happy to sit deep and hit Kyiv on the break and the Ukrainian's did most of the pressing,finally opening the scoring with a deflected shot.They briefly threatened a second killer goal,but as the game wore on they were happy to absorb pressure from a now more attack minded home side.Stoke's fullbacks pushed on,they withdrew one of the holding midfielders to be replaced by an attacker and Kyiv withdrew their star attacking turn Shevchenko.The dynamic of the game was evidently changing and the uselessness of the draw to Kyiv had made the change much more apparent compared to a midseason,run of the mill league fixture.Stoke were becoming more attacking and Kyiv were withdrawing into a defensive shell. When Stoke equalised seven minutes from time they already had five attack minded players on the pitch and were preparing to add more.
With Jones' equalising goal the game's micro climate shifted for a third time,Crouch's intended arrival was aborted and instead a midfielder replaced Jones,Stoke's goalscorer.Walters an attacker with a willingness to help out in midfield was also introduced to help stem Dynamo's all out assault.Had the game been played under an English referee,Stoke would have had to endure at least 5 minutes of injury time,but German officials reputation for adding little or no injury time is well earned and Begovic's world class save in the second minute of added time was the game's last action.
If teams adjust their strategy and thus their likelihood of scoring or conceding goals in matches where game situations are stark such as the Stoke/Kyiv game,the question now becomes does the same thing happen to a lesser extent in more normal games.Are teams becoming more defensive when they lead,if so by how much and is the trade off they presumably receive in fewer goals conceded worth the tactical alterations.We should be able to find evidence for it within the match data and this knowledge will also be necessary if we are to more accurately trace the likely course of a match.
I'll present that data in a future post.