Ideas for Future Studies

[Czech Your Math]

Traditional adjusted points adjust a player's totals to leaguewide scoring levels. But we know that changes in leaguewide scoring don't affect all players the same. For example, scoring soared leaguewide as the 70s turned into the 80s, but the point totals of first line players increased by a much smaller amount than league offense as a whole. In the late 90s, scoring decreased all-round, but decreased a little less for 1st line players than others.

I would like to see a comprehensive look at the scoring levels of different "types" of players? Separating "first liners" from other forwards is one thing. I'm also interested in the percentage of offense that has come from defensemen and how it has changed over time.

Creating a new kind of "adjusted points" for just first line forwards and another for defensemen (and more for other categories that might be worth studying) would be the final goal.

I know CYM has done some preliminary work with this

I thought someone had looked at scoring by defensemen?

I can post a table of scoring tiers for all players (including d-men), proportional to the number of teams in the league (e.g. 1-30, 31-60, ... 331-360). I would have posted this quite a while ago, but it looks better as a graph (and can't figure out how to post a graph). It uses simple adjusted points, so any fluctuations indicate that a certain tier is doing scoring a larger proportion of points.

 

[metalfoot]

I know this is probably impossible to implement, but I wonder if a study on linear vs non-linear adjusted point systems would be possible. I mean, adjusted points always seem to blow up on outliers (cue yet another discussion on why Stamkos scoring 60 is way better than Gretzky scoring 92) but it seems to me that typical adjustment systems might fall down precisely because they are linear, and maybe there is a curve (of what form, I don't know) which would better fit the data? This would take a lot of brain power, spreadsheet/database wizardry, and imagination, but it might be worthwhile.

 

[Czech Your Math]

I know this is probably impossible to implement, but I wonder if a study on linear vs non-linear adjusted point systems would be possible. I mean, adjusted points always seem to blow up on outliers (cue yet another discussion on why Stamkos scoring 60 is way better than Gretzky scoring 92) but it seems to me that typical adjustment systems might fall down precisely because they are linear, and maybe there is a curve (of what form, I don't know) which would better fit the data? This would take a lot of brain power, spreadsheet/database wizardry, and imagination, but it might be worthwhile.

There's no evidence that adjusted points are inaccurate for outliers (individual players). Adjusted points seem to be less accurate when the talent pool becomes particularly thin or dense, which is also related to times when the league has extreme parity or disparity. They are also less accurate when conditions change and favor/disfavor certain types of players more than others (increased PPs, increased roster sizes, etc.).

There are two main periods that had "inflated" adjusted points: the decade or so after WWII and the decade or so after expansion. The main period that had "deflated" adjusted points was the decade or so after the WHA merger. Besides those periods, adjusted points work extremely well and the differences are relatively minor.

 

[WarriorOfGandhi]

despite the fact that I haven't been good at math since they started putting letters in it, I think this is an excellent idea for a forum

 

[seventieslord]

Something I'd like to look at someday (but honestly, it's not goaltender-related and therefore it always falls to the bottom of my list) is a comparison of the actual value of NHL draft picks (measured by some metric) versus the perceived value of NHL draft picks (measured by the value that NHL GMs impute when they trade those picks on or before draft day).

I'm throwing it up for grabs, and would love to be involved if things get moving.

I have a big sheet full of data that I put together in 2006 and 2007, that just might answer this exact question. I would just have to organize the data in it a bit.

I abandoned the draft project when I became as interested as history as I now am, but the information compiled in my sheet has come in handy on a few occasions to answer questions and settle disputes.

Related to expansion, I would be very interested in a study of how much the numbers of a player were helped by playing for an O6 team after expansion and how much playing for an expansion team hurt

Edit: TCG did a little of this in his "Questioning Ed Giacomin" post

http://brodeurisafraud.blogspot.com/2010/10/questioning-ed-giacomin.html?m=1

If we’re talking about scoring stats, it could easily be the other way too.

Look at all the guys who were top scoring players on the expansion teams in the fisrt few years, and what they had done prior to expansion. In a smaller league with deeper teams, they were secondary/depth players, or even minor leaguers. A few examples are Lou Angotti, Leon Rochefort, Gary Dornhoefer, Eddie Joyal, Lowell MacDonald, Red Berenson, Wayne Connelly, Ray Cullen, Andre Boudrias, Earl Ingarfield, Ken Schinkel, Gerry Ehman, and Bill Hicke. Being on an expansion team helped their stats.

 

[Buttonwood]

The problem with that is that as fans, our opinions are very biased.

I think an underlying, quantifiable 'dividend' needs to be present to keep opinions in line. I would suggest player's share of ice-time per point earned multiplied by a set magnitude. Then there's no way Phaneuf would be more valuable than Lidstrom. People would still have opinions and 'buy' speculatively, but I think people would be smart to try to buy the best performing players, which would take literally hundreds of opinions and combine them, which can be more valuable than one or two 'expert' opinions IMO. There might be a better mechanism to achieve this goal, but I think it could be more valuable than some 'garbage-in garbage-out' statistics.

 

[GKJ]

I'd like to see something around board battles. Which players are good at keeping/extending possession on supposed 50/50 pucks? Which guys who are regarded as gritty energy players who do the dirty work are overrated in their role by not winning as many pucks as we think?

 

[Lonewolfe2015]

Now that I think about it, a really useful project for the improvement of hobby sports statistics people would be for someone (or a group of people) to develop an opensource data miner for sites like nhl.com

It's not as bad copying/pasting data like in the past, but it's still tedious due to the 30 results per page format they use.

Unless anyone else ever came up with a good solution to this?

 

[pnep]

This is what Pnep has done with HHOF Monitor Points. I'm guessing he is using arbitrary, intuitive values. I think it could be fine-tuned using linear regression.

regression is used already

 

[pnep]

Now that I think about it, a really useful project for the improvement of hobby sports statistics people would be for someone (or a group of people) to develop an opensource data miner for sites like nhl.com

It's not as bad copying/pasting data like in the past, but it's still tedious due to the 30 results per page format they use.

Unless anyone else ever came up with a good solution to this?

AutoPager

 

[Guest]

I really like the idea of this forum subcategory.

This kind of reminds me of something I wanted to research. What the impact was on bench placement with the 2nd period being the long change versus the 1st & 3rd being a shorter length. I've had a theory that some teams with mobile defense, better skaters, or just overall younger would do better statistically in the long change 2nd periods. It could also lend itself to a defensive advantage for a team playing a trap or other suffocating systems.

 

[Czech Your Math]

regression is used already

That appears to be a reasoned calculation, which is great, but it's not exactly regression. The main difference is that regression simultaneously measures multiple variables, so if any are correlated, they will automatically be discounted to compensate for that. I see that Hockey Outsider mentioned regression in that thread.

I'm not suggesting you change your system, as it may not be a significant improvement. You might consider adding Pearson 1st/2nd, and third place for each major trophy (Hart, Pearson, Vezina, Norris) at some point. You've done great work!

 

[Hivemind]

I would love something like WAR (wins over average replacement) for hockey. There's just so many variables that make that difficult. Doing it with points only wouldn't be hard, but the defensive part would be hard to add in.

Isn't this what GVT (Goals Versus Threshold) is, or at least attempts to be?

GVT (Individual)

Goals Versus Threshold. Developed by Tom Awad of Hockey Prospectus, GVT measures a player's worth in comparison to a typical fringe NHL player. GVT has two major advantages over most metrics: it's measured in goals, which are easily equated to wins, and it is capable of comparing players across multiple positions and multiple eras. GVT is the summation of OGVT, GGVT, DGVT, and SGVT.

http://www.puckprospectus.com/glossary/

 

[Thirty One]

Coaches often pull their goalie at ~1:00 mark of the third period when trailing by a goal. I always wondered if that specific time could be supported by evidence. But I never had the general goal scoring rates (6 on 4, 4 on 6, 5 on 5) to come up with a time that would optimize the chance of tying the game.

 

[Czech Your Math]

Coaches often pull their goalie at ~1:00 mark of the third period when trailing by a goal. I always wondered if that specific time could be supported by evidence. But I never had the general goal scoring rates (6 on 4, 4 on 6, 5 on 5) to come up with a time that would optimize the chance of tying the game.

Interesting idea. There would be a lot of variables and assumptions involved to come up with a number mathematically. The various scoring rates and chances of each team receiving a penalty in each situation would be the main variables. However, teams are playing differently with a short time left in the game and one team ahead by one goal, so many of the general rates may not apply.

My instinct is that conventional wisdom may be wrong in this case. I think the goalie should probably be pulled much sooner. First, the defending team can take a penalty and the attacking team cannot fully utilize that power play. Second, the defending team is almost fully devoted to defending rather than attacking (and much of that is to keep possession as a form of defense).

There are many instances when the trailing team doesn't even get much of an opportunity with a minute left. Down one goal, I might pull the goalie when there is clear possession in the offensive zone with less than ~2.5-3 minutes left.

 

[Hivemind]

Coaches often pull their goalie at ~1:00 mark of the third period when trailing by a goal. I always wondered if that specific time could be supported by evidence. But I never had the general goal scoring rates (6 on 4, 4 on 6, 5 on 5) to come up with a time that would optimize the chance of tying the game.

This paper may interest you. Not exactly what you're talking about, but related and it talks about several other papers on the issue (haven't been able to find public links to most of them). Most famously there was also a paper in 1986 that produced the formula:
t=-{ln[(-L1*LB+L2*LA)/LA(L1+L2-LA-LB)]}/(LA+LB)
where L1 and L2 are the ES scoring rates of the two teams, L1 is the scoring rate of team 1 with a pulled goalie, and L2 is the scoring rate of team 2 when team 1 has pulled its goalie.
This paper reviews and expands upon that 1986 paper, and finds that (depending on the scoring rates of the two teams), the ideal time to pull is between 1.57 and 2.82 min remaining in the game. Obviously, it does not factor in game situations in that analysis (or factors like exhaustion, line changes, etc.). I would be interested in something expanding upon that to look at line matching and scoring rates of particular lines and defensive combos.


Somewhat related, I'd be curious to see what's a better strategy when on a 5-on-4 powerplay and your opponent commits a delayed penalty. Immediately turn the puck over to start the 5-on-3 or try to prolong the 6-on-4 as long as possible (thus extending the total time with a 2 man advantage)?

 

[hella rights]

I was wondering if anybody knows if there exists a validated measure for assessing player characteristics and abilities. There seems to be lots of quantitative data out there for players (points, time on ice, save %, etc.) but I'd be interested in combining that with some qualitative data and see, provided that the measure had sufficient reliability and validity, if there are certain characteristics and abilities (hockey sense, speed, agility, grit, etc.) or combination thereof that predict overall value (how to operationalize "value" I don't know yet, perhaps GVT or some other stat). Obviously how the characteristics are defined is crucial, was just wondering if something like this is already out there.

 

[maj]

One thing I think should be considered is "Win% from SV%".

Meaning we calculate a goalies Win% based on how many games he's won, with a SV% above the average(which I think is .910).

I did that for a few goalies, and some were real surprising, such as from Bruins and Thomas and Rangers and Lundqvist. Some goalies(like Luongo) are actually real similar to those guys.

 

[Bear of Bad News]

One thing I think should be considered is "Win% from SV%".

Meaning we calculate a goalies Win% based on how many games he's won, with a SV% above the average(which I think is .910).

I did that for a few goalies, and some were real surprising, such as from Bruins and Thomas and Rangers and Lundqvist. Some goalies(like Luongo) are actually real similar to those guys.

I calculate a support-neutral win-loss record based on save percentage (which I think is what you're describing). It has the same limitations as save percentage does.

 

[Czech Your Math]

It seems that weaker teams often receive more power plays than their opponents, at least in some seasons. Does anyone dispute this? Has anyone done a correlation study between team quality and net power play opportunities? I would probably use ES (GF-GA) as an indicator of team quality, to filter out the effects of the power plays themselves.

It seems to defy common sense that a weaker team would get more power plays than their opponents, because A) stronger teams should create more mismatches/advantages which necessitate penalties by the opposing team, and B) having more net power plays should make a team stronger by improving their GF/GA ratio and therefore win%.

My theory is that the NHL helps weaker teams to stay competitive by giving them more power plays, although I'm not sure when this started and how true it still is. I would guess as the league has become more balanced, that the disparity in power plays has largely or perhaps completely disappeared. Does anyone have any other theories as to why this would be? The only other one I can come up with is that weaker teams are less of a threat on the power play, so their opponents may have less incentive to avoid penalties. However, I don't believe this would explain the massive disparity in net power plays in some years.

 

[Lonewolfe2015]

It seems that weaker teams often receive more power plays than their opponents, at least in some seasons. Does anyone dispute this?

Colorado had a 77 PPO disparity between Philadelphia in both team's home games.
65 disparity on the road.

Overall? 7 of the top 10 PPO teams were playoff teams. And 11 of the top 15.

I'd highly dispute that actually.

A more interesting point would be to look into the trend of Carolina leading the league (or close to) in opportunities per year. This year was the only time they were not top 3 in the league since 01-02.

 

[r1234]

I had an idea of a measure allocating points to a player for his team's wins/losses based on his time on ice.

These are two ways to allocate the points based on TOI:
- TOI pts share: TOI/60 × (2 if win, -2 if loss, 1 if goes to shootout)
- Exact TOI pts share: TOI/(Total team TOI) × (2/-2/1)

Example of TOI pts share with Sidney Crosby (2011-2012):
- Nov. 21 game = 0.53 = 15.9/60 × 2
- Nov. 23 game = -0.62 = 18.68/60 × -2
- etc.
- Total TOI pts share = 3.13 = 0.53-0.62+...
- TOI pts share/game (×100) = 14.21

These are some issues with the measure:
- treatment of OT/shootout
- offensive players usually have lower TOI in wins than in losses
- doesn't consider magnitude of wins/losses (by how many goals). Could add a factor
- can't be quickly calculated. Could approximate with average TOI

I am not sure if this measure already exists or if it would give meaningful results.

 

[Thirty One]

This paper may interest you. Not exactly what you're talking about, but related and it talks about several other papers on the issue (haven't been able to find public links to most of them). Most famously there was also a paper in 1986 that produced the formula:
t=-{ln[(-L1*LB+L2*LA)/LA(L1+L2-LA-LB)]}/(LA+LB)
where L1 and L2 are the ES scoring rates of the two teams, L1 is the scoring rate of team 1 with a pulled goalie, and L2 is the scoring rate of team 2 when team 1 has pulled its goalie.
This paper reviews and expands upon that 1986 paper, and finds that (depending on the scoring rates of the two teams), the ideal time to pull is between 1.57 and 2.82 min remaining in the game. Obviously, it does not factor in game situations in that analysis (or factors like exhaustion, line changes, etc.). I would be interested in something expanding upon that to look at line matching and scoring rates of particular lines and defensive combos.


Somewhat related, I'd be curious to see what's a better strategy when on a 5-on-4 powerplay and your opponent commits a delayed penalty. Immediately turn the puck over to start the 5-on-3 or try to prolong the 6-on-4 as long as possible (thus extending the total time with a 2 man advantage)?

Thanks a lot. Good reads.

 

[Chootoi]

I've always wanted to see the statistics or percentages of a) how often the team that scored the last goal takes the next penalty (and how quickly), and b) the team that's leading taking the next penalty (or penalties) and how goal differentials play into it. I would be curious to see by the numbers how consciously or subconsciously it may be affecting a referees bias in making even-up calls, especially in 1 or 2 goal games vs. blowout games, etc.

 

[Hivemind]

A more interesting point would be to look into the trend of Carolina leading the league (or close to) in opportunities per year. This year was the only time they were not top 3 in the league since 01-02.

Divers! sarcasm: ?)