| |
Ice Time Estimation
In 1997 the NHL began using the Real Time
Scoring System (RTSS). Among the other assorted jewels of information the
system recorded, such as hits, blocked shots, giveaways, etc, it also, for
the first time, began logging a player's ice time: a very important event
in trying to understand this game more deeply.
Knowing how much icetime a player gets can greatly fill in the blanks as
to what kind of a player he is. By seeing power-play, short-handed,
even-strength minutes played, we learn where a player's skills lie and
where they don't. Knowing these things also allow us to make some
inferences about what has probably always been one of the great unknown
quantities: defensive skill.
The problem is, what about all those years BEFORE the introduction of RTSS
tracked ice time? Can we attempt to measure it? Basically, we can have a
good idea, actually.
Since 1968, when the league made its first big expansion since the 1920s,
the league has been measuring plus/minus. In order to do that, the league
has kept track of when a player was on the ice in situations of goals for,
goals against, and the number of those times when it was a power-play or
short-handed opportunity. By studying those instances we can estimate how
much time a player received during the season. It's easiest to proceed
through this step-by-step.
Team Minutes Played
Essentially, there are 3 versions of this formula, depending on the era of
hockey you're studying
Before 1984, the league didn't force over-time in an attempt to break
ties. So...
TMIN = MGP * 60
Where TMIN equals Team Minutes Played, and where MGP equals the Maximum
number of Games on the schedule. I know that MGP seems an odd wording, but
I named it such so as to reduce confusion between GP, which I only use in
reference to a player's own games played.
1984-1999
TMIN = (MGP * 60) + (T * 2.5)
Where T equal Tied Games, obviously. We don't know at all whether teams
played 1 or 5 minutes in overtime, on average, through the season. I aimed
to cut the error in half as best I could, so we give all teams an average
of two and a half minutes played per overtime game. Not pretty, not
perfect, but effective enough.
2000 -
TMIN = (MGP * 60) + ((T - OTL) * 5) + (OTL * 2.5)
Where OTL equals Overtime Losses. We at least know, now, how many games
went the full 5 minutes. Just about the only positive thing from the
cheesiest rule change since tied-down sweaters.
Minutes Played
By looking at how many times a player was on the ice when his team scored
or was scored against, we can get a good idea of how much he played. If a
player was on the ice for half of team's GF and GA, it's quite likely that
he was on the ice half the time, which of course is alot of ice time. Of
course, there are things that can fudge this up a bit. Very good offensive
players do not require as much time to score points as other players do.
So, immediately, the system will overrate their time. The opposite holds
true for players that score less.
MINS = (((PGOI / TGOI) * TMIN) * DCM)
Where MINS equals Minutes Played, GGOI equals the amount of goals the
player was on the ice for, TMIN (again) equals Team Minutes, and DCM
equals Depth Chart Modifier. A word on DCM:
By finding out the percentage of time that a player was on the ice when
his team scored or was scored against, we can multiply it by the number of
minutes the team played. We then modify that time, based on the line he
played on...
I created the DCM as a measure for correcting the various over and under
estimations of players based on the line would have played on. The ice
time system, unmodified, treats all players equally, which causes small
errors with players at each end of the ice time spectrum.
DCM Modifiers
1st line player: 0.9
2nd line player: 1.0
3rd line player: 1.0
4th line player: 1.1
During the testing phase of the formula, I discovered that the Ice Time
Estimation system overrated 1st liners by an average of 10% and underrated
4th liners by about the same degree. Some personal judgement has to be
made as to which players play on which lines, but when viewing the
unmodified minutes, it's VERY easy to tell.
NOTE: Some teams don't use four lines, especially before the 1980s. In
these cases, third line players receive the fourth line player DCM. In any
regard, the differences aren't generally staggering, but do go further
towards created a higher overall level of accuracy.
Situational Minutes
Because the league tracks whether or not goals are scored at
even-strength, power-play, or short-handed chances, we can also estimate
the amount of time a player received in each of these situations.
Power-Play
PPMIN = ((PGF / PGOI) * MIN)
Where PGF equals the number of times the player was on the ice when his
team scored on the power-play.
Short-Handed
SHMIN = ((PGA / PGOI) * MIN)
Where PGA equals the number of times the player was on the ice when his
team was scored on the killing penalties
Even-Strength
ESMIN = (MIN - (PPMIN + SHMIN))
Pretty straight-forward, right? His number of minutes, minus the amount of
minutes he played on special teams.
When comparing ITE estimates to actual 1999 and 2001 ice times, the system
is found to be a little over 98% accurate. Given that, there are
situations where a player's ice time has been off. In short, there are
moments that the system can't see it's way around. For example, my eternal
frustration with Mike Grier in 2001.
ITE prediction: 1145 MIN, 29 PPMIN, 279 SHMIN, 965 ESMIN
act. ice times: 944 MIN, 53 PPMIN, 234 SHMIN, 657 ESMIN
The special teams differences I can live with, but the even-strength
annoys me a bit. More than a bit, actually. The system, however, is
generally very accurate and cases like Grier's are very much in the
minority. Let's stick with the 2001 Oilers for another moment, and show a
happier moment from the testing process: Rem Murray.
ITE prediction: 1240 MIN, 85 PPMIN, 225 SHMIN, 930 ESMIN
act. ice times: 1259 MIN, 86 PPMIN, 226 SHMIN, 947 ESMIN
Now for some fun... Let's reach into my magic hat, and pick a year between
1968 and 1997...
Ice Time League Leaders, 1971
Minimum, 50 GP
Average Time on Ice
Defence
1. Bobby Orr, BOS, 30:41
2. Pat Stapleton, CHI, 29:36
3. JC Tremblay, MTL, 28:04
4. Dale Rolfe, NYR, 27:11
5. Dale Tallon, VAN, 26:39
6. Bill White, CHI, 26:20
7. Duane Rupp, PIT, 26:16
T8 Guy Lapointe, MTL, 25:59
T8 Ron Stackhouse, CAL, 25:59
10 Gary Bergman, DET, 25:40
Forwards
1. Phil Esposito, BOS, 25:35
2. Orland Kurtenbach, VAN, 22:37
T3 Jim Johnson, PHI, 22:22
T3 Jimmy Roberts, STL, 22:22
5. Bill Goldsworthy, MIN, 21:59
6. Gilbert Perreault, BUF, 21:38
7. Bobby Hull, CHI, 21:32
8. Norm Ullman, TOR, 21:28
9. Jude Drouin, MIN, 21:26
10 Frank Mahovlich, MTL, 21:26
Sheesh. Espo got WAY more ice time than the other forwards that year.
Ironic that he spent most of it standing in front of the net. Laaazeee!
Power-Play Minutes
1. Bobby Orr, 551
2. Phil Esposito, 544
3. JC Tremblay, 527
4. Fred Stanfield, 516
5. Johnny Bucyk, 488
6. Bobby Hull, 450
7. Stan Mikita, 439
8. Pat Stapleton, 404
9. John McKenzie, 403
10 Gilbert Perreault, 401
The '71 Bruins are one of, if not the greatest offensive teams in history.
It doesn't seem like a great shock to see them loaded up with power play
time.
Short-Handed Minutes
1. Charlie Burns, 503
2. Gary Doak, 421
3. Ted Hampson, 369
4. Murray Oliver, 342
5. Bill Collins, 337
6. Pat Quinn, 322
7. Glen Sather, 321
8. Bob Plager, 316
9. Andre Boudrias, 314
10 Lew Morrison, 306
I'm most interested in the short-handed minutes, as it's probably one of
the most overlooked aspects of the game. If two players of differing
offensive skill play the same amount of minutes, it seems very reasonable
to guess that the less offensively-talented player is a much better
defensive player. I'm working on something else, though, that can
hopefully help fill in that story.
I will also be updating my website with more ice times, hopefully on a
pretty regular basis.
NOTES:
I'm quite sure that many people have, over the years, independently made
ice time estimate systems using the same statistics base that I did. It
seems natural to me, since TGF, PGF, TGA and PGA are the only stats
available that would really be able to help fill in those blanks. I've
been working on this for quite a while now, as you can imagine that
filling in all players since 1968 is a rather large job. And I *DO* work,
have a family and eat dinner from time to time, so this is obviously not
the only way I spend my time. In my travels around the internet, I
discovered that another home hockey analyst has devised his own methods,
and yup, we use the same base stats as the basis of our work. Though, it
does depart aside from the original idea.
I regret to say that this system is bound to produce errors with certain
players. This is an unavoidable circumstance. After all, it's called Ice
Time Estimate for a reason. I firmly believe, though, that it's better
than the current system in place, which is: NO SYSTEM. Take this work for
what it's worth: it doesn't answer all of our questions definitively, but
it's another step in a long walk, which I believe to be in the right
direction.
Lastly, I'm sorry if this reading is too long for some. I try to be
relatively thorough when I discuss things like this, so I really can't
help myself.
Back
|
|
