Wouldn't it be nice to know how your goalkeepers compare to the rest of the league at saving different kinds of shots? Are they better at dealing with long range efforts to their left, to to their right? Wouldn't it be interesting if you could figure out weaknesses in your opponent's goalkeeping arsenal?
I think it would be, so I've patched together something I think is quite neat - a new interactive viz that shows the frame of the goal broken down into eight 4' by 4' zones. We then can look at the result of every shot that found its way into each of these zones, giving us goalkeeper efficiency for each area of the goal. Let's take a look at an example after the jump:
Figure 1: Premier League average save percentage by zone, 2006-present. [Link to full applet]
Of course, you don't just need to look at every shot. You can filter by season, defending team, shot type, and even distance in order to figure out, say, that if you want to beat Edwin van der Sar with a long range shot you'd better be aiming it up and to his right (i.e. the top-left corner). For Pepe Reina, it's the other corner you want to be aiming at.
There are a couple of things that limit this as an analytical tool, however:
- Filter by teams, not individual goalkeepers. I could fix this, but you're looking at a few days of just poking excel with a stick, something I have very little intention of doing right now. For the most part, it doesn't matter thanks to regular goalkeepers getting most of the playing time, but it's a problem when you want to compare, say, Hilario to Petr Cech.
- No sorting by velocity. It would be nice to show that harder hit balls are difficult to save - we know it to be true, intuitively, but we don't know by how much. Again, a question of time on my part.
- No extrapolation of final shot location. If a shot is saved by the goalie right at the feet of the shooter that would have gone into the bottom right corner, it's marked as the location of the save rather than the location of the final shot. This may lead to slightly skewed results.
All in all, I don't think these are that problematic, and certainly this is a better tool than the absolutely nothing we were using before. I plan on developing this further and addressing the problems mentioned above... but for now, go nuts!