Visualising & analysing possession in hockey
This blog post focuses on possession in hockey, and is divided into 2 parts. The first presents 3 ways in which possession could be visualised, and the second investigates Team GB’s possession during their Olympic group matches.
Part 1
This post is in some ways an extension of an MSc assessment I recently had to produce. For this assessment, I decided to role-play as the Netherlands hockey teams analyst at this summers Olympics and produce an opposition analysis video of Team GB ahead of the two teams semi-final clash. In this role, I wanted to gain a thorough understanding of how Team GB played. One way I did this was by producing a social network graph (see one of my earlier posts). To produce one of these graphs I had to record all passes made by Team GB’s players, and in chronological order. Another way I did this was to expand upon the data by recording, for both teams, whether a defender, midfielder or forward made a pass, and how frequently these occurred per 5 minute period of the match. Pooling this passing data from each of Team GB’s 5 group stage matches, and calculating the averages, I was able to produce a graph to show the average number of time a Team GB or their opposition made a pass per 5 min period of the match, and split this according to position. For this assessment I chose to represent the averages, rather than overload the movie with graphs of each game, however for this post I will present one of these graphs that shows possession for just one specific match. I will then present 2 other graphs, and give my views on each of them. To help comparisons, the 3 graphs are all for the Team GB v Australia match.
Graph 1: Passes made by team, time, and position.
This is the same style of graph I used for my presentation. Presenting the passing data in relation to position may help, for instance, show if a team struggles to get the ball to their forwards, as well as revealing at which stages of the game teams dominate possession. In this graph it would appear that, for both teams, defenders make the most passes, and that whilst Team GB’s midfield seem to make more passes than Australia’s, Australia’s forwards make more passes than their Team GB counterparts. We can also see that GB made lots of passes early, made very few between the 21-25 minute mark, and then made more until the end of the first half. In the second half it appears that each of their 3 goals (yellow circle) came during a 5min period where they made more passes than Australia.
Graph 2: Cumulative Passes
This graph shows the running total, at 5min intervals, of passes made (goals are the big circles). We can see that the teams were relatively similar in the total and rate at which they made passes until the 16-20min mark, shortly after they took a 2-0 lead. It wasn’t until the 46-50min period that Team GB’s total passes overtook Australia’s, and it was at this point they scored their first goal. I like that after the last data point it shows the total number of passes made, as I was unable to tell who made more from viewing the first graph.
Graph 3: Possession %
This graph is the possession % per 5min period. Possession was calculated by dividing a teams number of passes in a 5min period from the total passes made by both teams in that same 5min period and multiplying by 100. From this graph we can see the wild fluctuations that occur as the match develops, and is particularly useful as the overall possession % were 50.7% for GB and 49.3% for AUS. Although it looks very similar to the first graph, there are differences. For instance in graph 1 the greatest peak in the data (most passes made in a 5min period) for Team GB is at the 31-35min mark. However, when we look at graph 3, we can see that, for this 5min period relative to the opposition, they only had just under 80% possession and actually had a greater % of possession during the 46-50min mark.
Conclusions
I like that, using the same data, it has been possible to present 3 graphs which all tell slightly different stories. Im not sure if I could say whether one graph is better than the other two though, and I think the greatest insights into the performance will come after assessing each one. I have presented these graphs not to analyse how Team GB performed against Australia, but to analyse the merit of each graph type in analysing possession.
Part 2
This section I have looked at the total number of passes Team GB and their opposition made, and examine if match half affected the number of passes made.
The graph below shows the number of passes made per game by Team GB and their opposition. We can see that in 4 of 5 games Team GB made more passes than their opponents.
Then, I examined Team GB’s passes relative to match half (graph below).
We can see that for every game Team GB made more passes in the first half than the second half. I wanted to examine this further, so performed a paired samples t-test, finding that Team GB make significantly more passes in the first than second half (t=3.06, p=0.03, p<0.05). Why this occurs – fatigue, tactics or other factors – will remain unknown without further analysis. What is interesting, is that Team GB scored 64% of their goals in the second half, despite making significantly fewer passes in this half.
I then wanted to see if the same was true for Team GB’s opposition (graph below).
3 times Team GB’s opposition made more passes in the second half than the first. Again, I ran a paired samples t-test, however I found no significant difference between the number of passes made in the first and second halves (t=0.73, p=0.50, p>0.05). An issue here is that combining the data for the 5 teams may mask any individual team effects, though it still gives some indication as to how teams were able to perform against Team GB.
Chris nice post. Some interstion points about posession. When looking at passes per position how did you class this? Was it by physical location on the pitch or who the player is? I ask because Middleton often comes very deep (sometimes deeper than the centre half to contribute to build up play). Also one of GBs strengths is overload from midfield.
One idea I had about an explanation for the passing in half is that often quicker more direct attacks lead to more decisive chances once a game is opened up because of tiring legs.
Cheers. I classed it on playing position, as I when I was analysing the matches for my coursework I was already recording where possession was won/lost on the pitch, which sides GB and their opp entered the 25/circle into and so wanted some slightly different data that reflected the players interactions to compliment the pitch location stuff. Only problem was whilst I had no problem identifying the GB players there was a bit (lot depending on team!) of guess work but tried to take notes pre-game of squad and deduce as I was analysing who was who.
Yeah definitely – little doubt fitness a major factor. On this point did some analysis over summer looking at frequency of substitutions in the group stages of Olympics and tournament standing; essentially the more subs you make the greater your final standing (will have to tidy it up and write blog post some time). Will need to check it all but found Germany averaged 73 per game and India just 36 – correlation was 0.81 (strong!). Medalists were the top 3 teams for number fo subs per match.
I think also, though didnt write about it in my post, in 4 of the 5 games analysed GB were leading at half time. Perhaps, and similar to some research in football (http://www.altorendimiento.com/es/congresos/entrenamiento/4362-stabilty-of-ball-possession-strategies-in-soccer-the-influence-of-situational-variables – specifically Bloomfield et al. (2005a), Jones et al. (2004), Lago and Martin (2007) and Sasaki et al. (1999)) teams in hockey that are ahead have less possession than the team that is behind.