By Eliot McKinley (@etmckinley)
Much has been written and studied about set pieces in soccer. Penalty kicks have been Bayesed multiple times, I’ve analyzed free kicks in MLS and at the World Cup, corner kicks have been rigorously studied. But what about the humble throw-in? Aside from when teams develop a long throw-in program (see Delap, Rory) they are largely ignored or even ridiculed, in the case of Liverpool hiring a throw-in coach (see the first comment here).
Throw-ins are an integral part of every soccer game. In MLS regular season play since 2015 almost 64,000 throw-ins were taken (all data is based on 2015-2018). They occur 44 times a game on average, accounting for almost 5% of all passes. Twice during the 2018 MLS season throw-ins accounted for over 10% of passes in a game. However, a search of MLS’ website brings up only the definition of “Throw-In”, a defunct regular column (with some words about early use of analytics in MLS), and some discussion of long throw-ins as scoring threats. Nobody, at least publicly, seems to be talking about normal run-of-the-mill throw-ins.
I got interested in looking at throw-ins for a couple of reasons. First, while sitting on the outfield berm at First Tennessee Park for a Nashville SC game wrangling my 3-year old son, I noticed that Nashville seemingly kicked the ball out of bounds from the opening kick off. While this is not a common tactic, it is well established. This forced New York Red Bulls II to take a throw deep in their own half that was ultimately won by Nashville. This play ultimately didn’t lead to a shot or a goal, but it piqued my interest. A few weeks later, the Crew suffered an embarrassing loss to Orlando after giving up a penalty in second half stoppage time. Just before the penalty, the Crew’s Hector Jimenez allowed an errant cross to run out of bounds when it appeared to he could have recovered the ball and retained possession (I wrote about this at Massive Report). The Crew ended up losing possession from the ensuing throw-in and Orlando hit to crossbar with a shot a few seconds later. This was the beginning of a flurry of shots from Orlando that could have possibly been prevented had the Crew not allowed the ball to go out of touch. Both of these led me to think about how throw-ins can be utilized to gain an advantage over one’s opponent.
Let’s start with some definitions. A successful throw-in is one that “goes to a teammate directly without a touch from an opposition player”. This doesn’t mean that the team retains possession from a throw-in, just that the thrower hit one of his or her teammates with the throw. While a successful throw-in is generally a good thing and occurs 81% of the time, if a throw-in is too hard, to the head, or to a player under pressure it may not be retained by the throwing team. To quantify this, I took the possession definitions Cheuk Hei Ho initially derived for expected possession goals and looked seven seconds after the throw-in was taken to see if there were any possession changes. If the team kept possession for these seven seconds, the throw-in was retained, which occurred for 60% of throw-ins. The choice of seven seconds was somewhat arbitrary, but watching some video leads me to believe that this is in the right ballpark. Thus there are four types of throw-in results:
Successful with possession retained (51% of throw-ins; example video)
Unsuccessful with possession retained (9% of throw-ins; example video)
Successful with possession lost (30% of throw-ins; example video)
Unsuccessful with possession lost (10% of throw-ins; example video)
xThrow and xRetain Models
I created expected throw-in (xThrow) and expected throw-in possession retention (xRetain) models to better understand throw-ins. Both models are almost identical to the expected passing (xPass) model developed by Matthias Kullowatz & Jared Young here at American Soccer Analysis, but with one extra variable, the amount of time since the last action. The time since the last action turned out to be the fourth most influential variable in xThrow (after angle, whether or not it was a long throw, and the x-location) and the third most influential for xRetain (after angle and x-location). Go read Matthias’ article for specifics on how these types of models models work.
As you can surmise from the raw numbers above, retaining possession after a throw-in is much less likely than successfully throwing the ball to a teammate. Much like xPass, the angle that a throw-in is taken matters a lot. Throw-ins taken vertically down the field are less likely to be successful or retained, whereas throw-ins taken backwards are more likely for both.
Moving on to the position a throw-in is taken. In short, the closer a throw-in is taken to a team’s goal the less likely it is to be successful or retained. In fact, for throw-ins taken within 20 yards of a team’s own goal, possession is more likely to be lost than retained. However, only 7% of throw-ins are taken that deep in a team’s own half. As you can see by the marginal distribution at the bottom of the plot, most throw-ins are taken between 30 and 90 yards from a team’s goal where both the probability of success and possession retention are near their maximum. Within about 30 yards of the opponent’s goal the probability of success and possession retention begin to fall off again, but both remain more likely than throw-ins taken deep in a team’s own half.
Time Since Previous Action
The time between the ball goes out of touch and a throw-in is taken has a surprisingly large effect on the probability of success and possession retention. Most throw-ins are taken within 7-15 seconds after the ball goes out of touch. The optimal time to take a throw-in is about five seconds after the ball goes out of bounds to retain possession. Presumably, if a throw-in is taken prior to this, the throwers team mates may not yet be prepared to receive the ball, and after five seconds allows the opposing team to get set to defend. As I mentioned previously, I added a “time since previous action” variable to the xThrow and xRetain models that was not present in the xPass model which is apparent when looking at the plot. The solid black like of actual throw-in results tracks much closer to the xThrow model than the xPass model due to the addition of this variable.
Similar to passes in general, short throw-ins are more successful than long throw-ins. Throw-ins fewer than 10 yards are completed over 90% of the time and the rate of success drops off pretty linearly through 40 yards. Throw-ins with a distance between seven and 25 yards make up 70% of all throw-ins, and 13% of throw-ins travel more than 30 yards. As the xPass and xThrow models don’t take into account the distances directly, but rather use a “long ball” flag, neither are as well matched as the other variables discussed earlier. Surprisingly, the distance a throw-in travels does not appear to have an effect on the probability that a team will retain possession within seven seconds of the throw. I would have expected that, like success, possession retention would also decrease with increasing throw-in distance, but a spot check of about 100 throw-ins suggested that the algorithm was reporting retention accurately.
Throw-in End Locations
Finally, let’s look at some maps of the end locations of throw-ins. I binned each 5x5 yard square on the field into zones and filtered out those that had fewer than 50 throw-ins ending in that zone. The size of the squares indicate the number of throw-ins terminating in that zone and the color denotes the success or retention probabilities. Gray Arrows indicate the median position on the touchline where throw-ins for each zone originated from. First, the end location of most throw-ins are within 10 yards of the sidelines (see the marginal distributions on the left side). Second, as you’d expect, throw-in retention is lower everywhere on the field except for throw-ins into the attacking 18 yard box. Presumably this is due to the defensive team clearing the ball from their box and the subsequently being recovered by the team that took the throw-in. Third, throw-ins received in a players own half are (attacking direction is left-to-right) are more likely to be coming from a backwards throw than those in the attacking half.
Finally, you may be asking yourself: “You said that throws taken in the final third were less likely to be successful or retained, but those maps make it look like they have higher rates. Why did you lie to me?” But the maps above are looking at where throw-ins ended, not where they started. If you take the average throw-in end location every five yards on each side of the field it tells a different story. Throw-ins originating in the defensive third are much more likely to be more vertical than those taken anywhere else on the field. As such, they are less likely to be successful or retained by the throwing team.
If I were a coach, here are three actionable steps that I would have my team make to gain an advantage in throw-in situations:
Don’t let the ball go out of bounds deep in our own half if you can prevent it. It is basically a coin-flip whether or not your team will retain possession after a throw-in in your defensive third. So if there is not a lot of pressure from an opponent, hustling a bit to keep the ball in bounds is a smart decision.
Use when our opponent has a throw-in in their defensive third as a pressing trigger. The likelihood of a turnover following a throw-in is high so be prepared to take advantage of it.
Take your throw-ins 5-10 seconds after the ball goes out of play. I would have my players immediately get into positions to receive a throw-in before my opponents get set. I’d also employ some gamesmanship (at least in CCL), perhaps having my players try to slow down an opponent from taking a throw-in to allow my defense to get into shape. When playing at home I may instruct the ball kids to expeditiously get the balls back to my team while being a bit slow and inaccurate with getting balls to the away side (but never go full Swansea ball boy).
There are many lessons that can be learned by studying throw-ins, the biggest is that we should probably be paying more attention to them. There is potentially much to be gained with relatively little effort. I’ll be following up shortly about how teams and players are taking their throw-ins during MLS play, so stay tuned.