I spend a lot of time making lists and obscure statistical arguments about playoff-bound teams these days, but I rarely look at what I like to call "basic research". Well, here's one for my more statistically-inclined readers.
Since the early days of the "shot quality" concept, people have been aware that rebounds are more dangerous than other shots taken from the same location. In particular, shots taken 1, 2 or sometimes 3 seconds after the last shot are proportionally more dangerous than all other shots. You can see that in this table, which shows the expected number of goals and shooting percentage – based on shot location – vs the actual number of goals. The ratio of the two is the "relative danger" of a given even-strength shot:
Time | # Sh | Expt Goals | Actl Goals | Ex S% | Act S% | Rel S% |
0 | 3665 | 293 | 580 | 7.99 | 15.83 | 1.98 |
1 | 6292 | 636 | 1105 | 10.11 | 17.56 | 1.74 |
2 | 7174 | 681 | 1335 | 9.49 | 18.61 | 1.96 |
3 | 5995 | 443 | 662 | 7.40 | 11.04 | 1.49 |
4 | 6568 | 411 | 450 | 6.26 | 6.85 | 1.09 |
5 | 7301 | 433 | 351 | 5.93 | 4.81 | 0.81 |
6 | 7604 | 441 | 377 | 5.80 | 4.96 | 0.86 |
7 | 7656 | 442 | 390 | 5.78 | 5.09 | 0.88 |
8 | 7519 | 431 | 365 | 5.73 | 4.85 | 0.85 |
9 | 7422 | 417 | 329 | 5.61 | 4.43 | 0.79 |
10 | 6743 | 386 | 355 | 5.73 | 5.26 | 0.92 |
11 | 6630 | 378 | 365 | 5.70 | 5.51 | 0.97 |
12 | 6178 | 344 | 339 | 5.57 | 5.49 | 0.99 |
13 | 5817 | 337 | 304 | 5.79 | 5.23 | 0.90 |
14 | 5625 | 314 | 282 | 5.59 | 5.01 | 0.90 |
15 | 5495 | 299 | 320 | 5.45 | 5.82 | 1.07 |
16 | 5116 | 285 | 287 | 5.57 | 5.61 | 1.01 |
17 | 5118 | 284 | 284 | 5.54 | 5.55 | 1.00 |
18 | 4940 | 276 | 259 | 5.58 | 5.24 | 0.94 |
19 | 4881 | 268 | 237 | 5.50 | 4.86 | 0.88 |
20 | 4666 | 260 | 268 | 5.57 | 5.74 | 1.03 |
There are a couple of interesting phenomena here. First, I was surprised to see that shooting percentage is significantly lower than expected for shots taken 5-9 seconds after the last shot, while for 10+ seconds there's no difference. I don't see a good reason why shots would be less dangerous in that time frame – while they're not rebounds, they still imply that the attacking team recovered the puck and got an opportunity to shoot on net. I don't doubt that the collective wisdom of my readers will present an obvious explanation for what's baffling me at the moment.
The second thing that surprised me is the "zero-second" rebound. In other words, two shots taken in such quick succession that no time elapsed on the clock. We've seen bias in pretty much every other subjective measure recorded by NHL scorers, so it's no surprise that there's a 3.5-1 variation in "zero-second" rebounds across the league:
fla | 250 | dal | 121 | buf | 87 |
nyi | 222 | col | 118 | pho | 86 |
det | 213 | nj | 116 | cls | 82 |
sj | 187 | nyr | 107 | edm | 79 |
van | 171 | ott | 104 | la | 79 |
tor | 170 | bos | 101 | min | 79 |
atl | 160 | ana | 100 | mon | 78 |
chi | 156 | was | 99 | nsh | 75 |
cgy | 152 | phi | 97 | car | 74 |
tb | 141 | stl | 89 | pit | 72 |
Tom Awad had previously pointed out the irregularities in Florida. There's much less variation in other recorded shots:
Time | 0 | 1 | 2 | 3 | 4 | 5 |
Max/Min | 3.5 | 2.4 | 2.4 | 2.0 | 1.9 | 1.9 |
The ratio of the highest-shooting rink to the lowest-shooting rink on non-rebound shots is approximately 1.3, which is slightly larger than what we'd expect given how many shots are taken in games played by the highest-pace teams vs the lowest-pace teams, but it doesn't indicate a massive recording bias. But there is a huge amount of variation in the total number of rebounds recorded in each rink:
fla | 1299 |
car | 1172 |
col | 1152 |
mon | 1126 |
edm | 1116 |
nyi | 1115 |
tor | 1071 |
sj | 1069 |
nyr | 1062 |
ott | 1061 |
tb | 1052 |
nsh | 1010 |
la | 994 |
was | 992 |
atl | 985 |
pho | 978 |
buf | 976 |
phi | 974 |
cgy | 955 |
pit | 955 |
ana | 928 |
min | 915 |
stl | 893 |
van | 889 |
dal | 862 |
det | 852 |
nj | 832 |
bos | 816 |
cls | 815 |
chi | 778 |
Rebounds were considered to be all shots that were taken 0-4 seconds after a previous shot. The difference between Florida and Chicago is about 125 shots per season (I used 2005-09 in my database) which amounts to something like extra goals expected goals per season. It's possible that the rebound total variations aren't due to scorer bias, but I have my doubts. I'll check into that at some future date.