Warm-up
11 Oct 2021Warm-up
Question 1
What issues do you notice by using this metric (‘SV%’) to rank goalies? What could be done to deal with this?
As we can see on the dataframe below (which is the top 10 goalies ranked by ‘SV%’ for season 2018-2019), some of the goalies have only played a few games. Since they faced way less shots against them than most of the other goalies, they can reach a very high SV% in case they stop these. Let’s take a look at Jack Campbell as an example. He is one of the top goalies according to this metric (‘SV%’ = 1.0) with only 5 shots against and 5 shots saved. While for the goalies which played most games of the season, it is way more difficult to reach a ratio close to 1 since they face way more shots. Thus, it’s almost impossible for them to reach such a ratio.
| Player | Tm | GP | SA | SV | SV% | |
|---|---|---|---|---|---|---|
| 63 | Alex Nedeljkovic | CAR | 1 | 17 | 17 | 1.000 |
| 85 | Dustin Tokarski | ANA | 1 | 5 | 5 | 1.000 |
| 11 | Jack Campbell | LAK | 1 | 5 | 5 | 1.000 |
| 29 | Kristers Gudlevskis | TBL | 1 | 3 | 3 | 1.000 |
| 26 | Jon Gillies | CGY | 1 | 28 | 27 | .964 |
| 51 | Charlie Lindgren | MTL | 2 | 59 | 56 | .949 |
| 81 | Alex Stalock | MIN | 2 | 54 | 51 | .944 |
| 17 | Aaron Dell | SJS | 20 | 533 | 496 | .931 |
| 9 | Sergei Bobrovsky | CBJ | 63 | 1854 | 1727 | .931 |
| 33 | Magnus Hellberg | NYR | 2 | 28 | 26 | .929 |
In the other case, if the goalies who played only a few games didn’t play well, their score is pretty low and they could not increase it, they had less opportunity than a goalie who played many games. The dataframe below is 10 goalies with the lowest SV% score.
| Player | Tm | GP | SA | SV | SV% | |
|---|---|---|---|---|---|---|
| 50 | Michael Leighton | CAR | 4 | 92 | 80 | .870 |
| 55 | Spencer Martin | COL | 3 | 96 | 83 | .865 |
| 58 | Zane McIntyre | BOS | 8 | 155 | 133 | .858 |
| 24 | Anton Forsberg | CBJ | 1 | 27 | 23 | .852 |
| 57 | Marek Mazanec | NSH | 4 | 87 | 73 | .839 |
| 32 | Andrew Hammond | OTT | 6 | 86 | 72 | .837 |
| 13 | Pheonix Copley | STL | 1 | 29 | 24 | .828 |
| 83 | Malcolm Subban | BOS | 1 | 16 | 13 | .813 |
| 19 | Chris Driedger | OTT | 1 | 15 | 11 | .733 |
| 1 | Jorge Alves | CAR | 1 | 0 | 0 | NaN |
So this scenario might be in favor or against the goalies while ranking them using SV%.
To face this issue, we could filter out the goalies which have played less than X games or which have faced less than Z shots. The NHL website has set this threshold X to 20 games. It make sense to us to use this value.
Question 2
Filter out the goalies using your proposed approach above, and produce a bar plot with player names on the y-axis and save percentage (‘SV%’) on the x-axis. You can keep the top 20 goalies.
Question 3
Save percentage is obviously not a very comprehensive feature. Discuss what other features could potentially be useful in determining a goalie’s performance.
Save percentage is a bit simplistic, we would need to take into account the game winning streak of the goalie, the number of shots faced per game, the average distance of the shots and the save percentage in 5v5 situations, since a goalie which spends a lot of game time in penalty kill is bound to let a couple of goals in more often.