A Study Of the First 1000 Shots in the EPL This Year

The first time we properly scrutinsed Differentgame’s expected goals model we had 3000 shots in the data bank. We were a little bit scared to make conclusions on what might have been considered a small sample size – a full Premier League season comes in at around 10000 shots. By the time those had been accounted for very little had changed. A full three seasons and over 30000 shots later, and we’d stopped worrying.

And then someone (we forget who exactly) in the soccer analytics community started going on about game state. They had a point. A team doesn’t go about the game the same way when 3-0 up in half an hour as it does when it’s 2-1 down with five minutes left. Or does it?

Being gluttons for punishment we decided to go back into the publicly available data and start over, recording all the things we should have recorded in the first place. It’s a laborious process, the kind of process where you continually set new targets to cheer your self up as you go along achieving them one by one. Well, once a 1000 shots were hit, we couldn’t resist delving into what we had. It turns out that on the original model a 1000 shots is still enough to get the same outcomes. Actually, it turns out that 500 was enough. Football can be predictable.

Anyway, the urge to write something came upon us. Here’s how the number of shots needed to score from each area changed depending on what the score was:

Game State Shots Needed

Ok, so a few points here…

First, even after a tenth of a season you’ve got no team yet scoring from wide inside the box when the game is tied at 1-1 or 2-2 etc, or indeed when 2 or more goals down. Also, no team has yet scored from outside the box when a goal up. Unbelievable, Jeff, even with a small sample size?

We can also see that the number of shots needed from wide inside the box, and outside the box are a little bit all over the place. We’ll deal with how desperate teams get with their shooting positions further down.

The two stand outs are in that central inside the box column. At 0-0, it’s taking 8 shots to score from there – possibly signalling two teams getting to grips with the game and being naturally cautious until they’re settled. At 2 or more down, it looks like a case of “heads gone”  – 39 shots to score? Again, sample size is an issue but it still looks capable of being a massive outlier going forward.

What’s not so different in that column are close game states once somebody has scored.

When soccer stattos start talking about game state they like to bring in Man Utd’s conversion rate – it’s ludicrously high year on year. Forgive us again technical bods, but we couldn’t resist testing the model out over just the 4 games Man Utd were involved in over this sample. The original model predicted them scoring 8 goals over these games. They actually scored 10. We plugged the new game state numbers into our formulas. It predicted well over 9. It remains to be seen whether we’ll get nearer for every team overall. Back to the data scraping for that…

We also wanted to see how ‘desperate’ teams get at differentgame (see what we did there) states depending on how long’s left. Here’s the numbers:

Desperation Time and Game StateWhen teams are drawing, it looks a lot like the average overall numbers of where shots are taken from. However, as we saw in the first graphic, teams are nearly twice as accurate from central areas inside the box once it’s a score draw. It seems again to reflect the fact that early game caution has gone. With the game more fluid we reckon the number of players committed to attack goes up. We’ll be collecting that data as much as possible too once we’ve been through all the MOTDs on our planner.

When a team goes a goal up it seems they suddenly start to get picky – like a baseball player waiting for the right pitch to come along. At one down, it seems teams work more chances in the box but much of them are in poorer wider areas. However, again the conversion rate is good as we see in the first graphic. Again we ponder on the number of attackers starting to pour forward.

The numbers at 2 or more goals up are really interesting. In the middle third of the game these teams look to start a turkey shoot. They look to be brimming with enthusiasm having recently gone two ahead. Once the game draws to a close they get sensible again working much better positions.

Teams at 2 or more goals down just look happy to shoot. They work plenty of chances in all areas. However, the conversion rates as shown in the first graphic are utterly dismal. Is the opposition simply defending in numbers or are attackers snatching at shots? Psychology could possibly have a greater part to play here? Has the belief simply gone?

Follow us on Twitter here.

Advertisements
This entry was posted in Sports and tagged , , . Bookmark the permalink.

4 Responses to A Study Of the First 1000 Shots in the EPL This Year

  1. Nick says:

    How about looking at game states in 18 5-minute blocks. For example, does a 1-0 up scenario change goal probability as game continues? Not sure if that makes sense.

  2. There’s all sorts of different combos to try out, nick. In the off season I’ll be mainly data collecting so there’ll be lots of material to play with and articles to write over the blog’s 2nd yr!

  3. Pingback: differentgame

  4. Pingback: Striking Questions – Part One | differentgame

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s