November Deep Dive

If there is a hockey god, my prayer to him
or her would be for November to never end. Make no mistake about it, I want to
live in a world where Jannik Hansen can score a hat trick and Luca Sbisa can
run through Patrick Kane like bad food court sushi. I want to live in a world
where Ryan Miller is better than Cory Schneider, where Bo Horvat is the best
Canucks rookie since Trevor Linden, and the Canucks are better than the Sharks
and Blackhawks. November was fun. Long live November.

Of course, it was not all sunshine, rainbows,
and puppy dogs. Find out what I thought about the Canucks November after the
jump.

The
Competition

After cruising through an October schedule
where the league threw the Canucks a ton of softballs, many people, including myself,
expected the Canucks to stumble in November when facing stiffer competition
from the likes of the Kings, Blackhawks, Red Wings, Sharks, Ducks and
Predators. By no means were the Canucks flawless. They were handed losses by
the resurgent Predators, buoyed by future 2015 Calder winner Filip Forsberg and
Pekka Rinne, who seems determined to regain his mantle
as one of the league top goalies. They were outclassed against the Kings (yet
again), they failed to show up against a vastly inferior Coyotes team, and they
fell short in net and defensively against a strong Detroit team.

Other than that, things went almost
perfectly for the Canucks. The won a game they had no business winning against
the Sharks, they convincingly crushed a Blackhawks team which is still very
much the class of the NHL, they mounted a very impressive comeback versus the Avs,
they split points with the Anaheim Mighty Kesler’s, and Miller posted a pair of
shutouts against the Columbus and New Jersey. They collected 8 out of a
possible 10 points against this month’s lottery opponents (Edmonton x 2,
Ottawa, Columbus, and Arizona), which is exactly what you’d expect good teams
to do.   

They exited November with a 16-7-1 record,
which was 1 point back of top spot on the NHL. They enter December continuing
their 7 game road trip, which will see them play the Capitals, Penguins, Maple
Leafs, Senators, and Canadiens.  The
Capitals, Senators, and Canadiens will all be highly motivated to avenge their
losses to the Canucks from earlier in the year, so this should make for an
entertaining road trip, to say the least.

The
Goaltending

The graph below shows the save percentage
for Lack and Miller in the first two months of the year: 

nov goalie

On the positive side, in his three starts
and one relief effort, Eddie Lack was significantly better than he was in
October, posting a very respectable .922 SV% in the month of November.
Hopefully he can build off these efforts and earn some additional playing time.

Ryan Miller, on the other hand, ran either
red hot, or ice cold this month. On the one hand, he had two shutouts. On the
other, he played 4 games where he gave up 4 or more goals. He failed to stop
over 90% of the shots he faced in 5 of his 11 starts, but did well enough in
his other 6 that was able to bring his save percentage in line with the league
average. For a while now I’ve been claiming that Ryan Miller is average goalie at best, but I suppose a better way of describing him is predictably unpredictable, with an upside of average. Not exactly the value you’re hoping for at a $6M AAV, especially with the backdrop of a potentially stagnating salary cap next summer

The
Defense

defense

The table above shows even-strength CF% for
each defender by month, and year to date. On the positive side, the Edler and Tanev
pairing continue to play excellent hockey, and after a slow start the Bieksa
and Hamhuis pairing turned their possession numbers around in November, prior
to Hamhuis’ lower body injury. In my October report, I expressed a fair amount
of skepticism that Luca Sbisa was in fact as good as his corsi numbers
indicated, so it wasn’t a complete surprise that his possession numbers have
come down to earth in November, however Ryan Stanton’s play thus far has beenmiles away from what I expected from him.

Because of the Hamhuis injury, Willie Desjardins
has had to shuffle is pairings around a bit. Here is a look at defensemen
corsi-for percentage with each of their most common playing partners with the
current pairings per dailyfaceoff.com highlighted in green:  

d wowy

By the chart above, you have to think
someone within the Canucks’ coaching staff is a user of stats.hockeyanalysis.com.
The first thing that jumps off the table for me, is that Ryan Stanton has been
an absolute train wreck when paired with anyone other than Yannick Weber, so
without Hamhuis in the lineup, a Stanton/Weber tandem is a very logical choice
as the Canucks third pair.

At 53.5 CF%, the Edler/Tanev pairing is
right up there with the top defensive tandems in the league, so Desjardins is wise
to not make changes to this pairing as a result of the Hamhuis injury.
Similarly, the Bieksa/Sbisa pairing hasn’t played a ton together so far this
season, but when they did play together they did very well in terms of both
corsi for (55.5%) as well as goals for (60%), so it’s worth giving this line a
chance even after watching them struggle the way they did against Detroit onSunday.

Obviously, the Canucks can’t wait to get
Hamhuis back in the lineup, but if the current pairings can continue to produce
as they have in the recent past, the defensive core should be able to tread water
until Hamhuis returns.

The
Forwards 

forward corsi

Although their offensive production has
dipped slightly in November after a white hot October start, the Sedins and
Vrbata have continued to produce at elite possession levels.

The Bonino-Higgins-Burrows line has probably
been one of the most pleasant surprises of the season thus far, with Nick
Bonino leading the team in even strength points with 18. Granted, the line is
still benefiting from a high on-ice even-strength shooting percentage. Bonino, in particular, has a 10.33% on-ice shooting percentage, which ranks 44th
out of 336 forwards who have played more than 15 games so far this season (14th
percentile). While, I think some regression is likely for Bonino, based on what
I’ve seen from him so far, I don’t think it’s out of the question that he’s
able to maintain an above average (ie >8%) on-ice shooting percentage
through the season. If he does, that would provide the Canucks with a very solid top two lines. 

While Bonino has received much of the
attention so far this season, as a result of his strong play since arriving in
the Kesler trade, the unsung hero on this line is none other than Alex Burrows,
which became increasingly evident in the 5 games that Burrows was out of the
lineup. Here are Higgins and Bonino’s even strength possession numbers with and
without Burrows. 

2nd line

There has been a lot of talk about the
Canucks being a team with three third lines. However, the even strength
corsi-for numbers for the Burrows, Bonino, Higgins line when on the ice
together and their points/60 minutes figures are all consistent with first line
production. If they can keep up their current possession trend and avoid a
significant regression in on-ice shooting percentage, they will end the year in
the discussion as one of the better 2nd lines in the league.

The bottom two lines are a different matter.
Here is a look at the most common even-strength combinations, with current
lines highlighted in green from dailyfaceoff.com: 

3rd fourth lines

Right off the bat, we can see the impact of
Matthais and Richardson losing Kassian from their line, as they drop from a
~49% even-strength corsi-for, which is what you’d expect from an average middle-6 line, to 47%, which is below average for a third line. The Vey,
Matthais, Richardson group had a great game against Detroit, recording all three
goals for the team. Hopefully, they can continue this going forward, although I
do have some skepticism as Vey has been a possession anchor in general this
season.

The fourth line of Dorsett, Hansen, and
Horvat has been eaten alive so far this season from a possession standpoint,
but when you look a bit deeper an interesting trend emerges. According to
war-on-ice.com, Horvat’s corsi competition (the TOI-weighted Corsi% for a
player’s competition) was 51%, which aligns with first or second-line
competition from a possession standpoint. Combined with his 46.97% offensive
zone-starts (lowest 40% of NHL forwards), and it’s no surprise his overall
possession numbers are where they are.

Horvat is the exact opposite of a
typical sheltered rookie, although his possession numbers to date are far from acceptable. They should improve as he gets acclimatized to the NHL and as his usually effective linemates round in to form, however. In terms of tangible production, Horvat, Hansen, and Dorsett have been nothing short of exceptional.

Team Level                                     

The table below shows key statistics for the
Canucks when at even-strength, on the power play, or on the penalty kill, for
October, November, this year-to-date, and last year: 

special teams

At even strength it’s hard to say the
Canucks have made any improvements over last year, except of course being more
fortunate in the area of shooting percentage. They’re actually posting a worse
save percentage and CF% than the prior year, so there’s definitely reason for
concern that the Canucks are punching above their weight-class. That said, as I discuss
above, there is some reason to believe the overall even-strength CF% could
improve if the team is able to maintain the possession trends they have seen when deploying their optimal line combinations on the ice.

On the power play, the Canucks experienced
a major regression in November in terms of the number of shot attempts per 60
minutes of ice time (PP CF/60). With a lower frequency of shot attempts they’ve
seen their goals for per 60 minutes of ice time (PP GF/60) plummet from top ten in the league to the bottom ten. This should
be a major area of concern and focus for the coaching staff.

The penalty kill, led by the Tanev/Edler
pairing, and most commonly Richardson and Higgins up front, has been sublime this year,
allowing one of the lowest shot attempts against (PK CA/60) in the league.
Unfortunately, they’re PK save-percentage in November was one of the worst in
the league, so the overall numbers don’t give their penalty killers the credit
their due. That said, the ability of this PK group to limit shot attempts the
way they have is impressive, and if Ryan Miller can do his job in net, they’ll be one
of the better penalty killing teams in the league.  

Conclusion

So what is the big picture? The
underlying number clearly indicate that the Canucks aren’t as good as their
record indicates, but they also point to areas of optimism, in particular the top two lines, the penalty kill, and the play of long-time Canucks Army favorite, Bo Horvat.

At the 24 game
mark, the Canucks have accumulated 33 points, so assuming the low-water market to
get into the wild card spot is 90 points again this year, the Canucks will only
have to play .500 hockey the rest of the way to make it into the post season. There’s nothing I’m seeing in the underlying numbers that would convince me
they can’t play .500 hockey the rest of the way, so it may be time to start bracing ourselves Vancouver fans: we may actually be looking at a playoff team. 

  • money puck

    At the end of the day, I’m just happy that the Canucks are at least playing exciting hockey. Hopefully we see an uptick in SV% (maybe with Lack getting a few more games), so even when we see the SH% regression it all balances out.

  • Canucks even-strength corsi % is average, but they’re among the top teams in the NHL in FenwickClose(which is a better predictor of future success than EVCorsi%) – they’re 6th in the NHL at 53%.

    Depending on the possession metrics you look at, the Canucks range somewhere between “average” and “borderline-elite”. The truth is probably in between – this is a pretty good possession team and a pretty good team overall, but they’re still a key piece or two away from being a legitimate contender (boy it would be nice to have another $5 million forward or defenceman instead of that $6 million goalie right now…)

    • If the Canucks had not spent a disproportionate amount of time up and down by one goal, the possession numbers would be better…

      This is probably a borderline top 10 team with the caveat that injuries haven’t tested Vancouver’s depth too much.

      As much as I’m not a fan of Miller’s contract, a Lack/Markstrom tandem could have been as bad as what Edmonton has ran out all year.

      Hoping that two should-be backups can provide good goaltending is pretty risky for a non-elite team with playoff aspirations.

      Ribeiro would have been a welcomed cheap addition, though…

      • Not necessarily saying they should have run with Lack/Markstrom. There were a number of other options. Hiller signed in Calgary for less money and a shorter term and has been better than Miller so far. Reimer could probably have been pried away from Toronto for not too much. League-average goalies aren’t very hard to find and don’t cost very much money (and with a 24th-in-the-league .904 team save %, Canucks goaltending is pretty awful – the worst it’s been since 05/06).

        Ribeiro would have been nice. Ehrhoff back on the 2nd pairing would have been pretty swell, too.

        Can’t quibble at all with Vrbata though – he’s looking like the best UFA signing of the year so far.

        And really, the Canucks are in the top 5 in the NHL right now and barring insane injury troubles are probably in the playoffs. This is all fussing over details.

        • Burke’s relationship with Hiller is probably why he chose Calgary.

          I’m not sure if Toronto’s front office wants to deal with Vancouver’s even with a different GM.

          I do agree, though, that either Hiller or Reimer (or Halak) would have been better than 3/18 for Miller.

          Miller himself wasn’t a horrible option if his contract was closer to Hiller’s.

          I just worry that if his skills erode a little or the save percentage environment continues to rapidly improve that he will no longer be an average goalie and then the contract becomes an albatross.

          For now, we cross our fingers that he doesn’t turn into a pumpkin during this contract…

  • “Assuming the low-water market to get into the wild card spot is 90 points again this year”

    Last year was an unusual year in the Western Conference with five teams at 107 points or above.

    Based on previous full seasons, 95 points is probably around where the 8th place team in the West will finish.

    Great post…

  • JCDavies

    @HR Pufnstuf and Goon

    Recent work by Micah McCurdy calls into question the usefulness of score close possession metrics. He finds not only that adjusted measures are superior for predicting future results but also that raw Corsi and Fenwick are superior to score close measures at almost all sample sizes. He finds that score close measures exclude too much “good” data and magnify the smaller portion of data that they keep.

    Here is a link to McCurdy’s article if you would like to read it:

    http://hockey-graphs.com/2014/11/13/adjusted-possession-measures/

    For what its worth, puckon.net currently has the Canucks ranked 19th in score adjusted Corsi and 12th in score adjusted Fenwick.

    • “He finds not only that adjusted measures are superior for predicting future results”

      Rank all 30 teams last year based on wins. Boston ranked 1st Anaheim 2nd all the way down to Buffalo 30th. Then Rank all teams 1-30 based on Score Adjusted Corsi. Compare the rankings. Score adjust Corsi accurately predicted (within +/-2 spots) the win ranking of 8 teams.

      All this talk about corsi and how it relates to future results is sketchy at best. People are putting way to much stock into a statistic that can accurately predict results (within 13%) 26% of the time.

      FYI – Score adjusted Fenwick predicted 9 of 30, 5/5 Corsi close predicted 12 of 30, 5/5 Fenwick Close predicted 11 of 30.

      • JCDavies

        No one is saying that Corsi and Fenwick are perfect but they are currently two of the best predictors we have for future team performance. I see a lot of criticism of Corsi and Fenwick where the faultfinders fail to offer any better alternatives. If you read the article, you would see that the author’s findings come as a result of an effort to improve on the measures we are currently using. I don’t know about you but I find any attempt to improve on the current metrics and move the discussion forward to be worthwhile.

        McCurdy’s findings seem pretty clear – that score close statistics shouldn’t be used for predictive purposes. Only time will tell if his work will be repeated by others and where it will lead to next. However, if you have specific concerns about his model, I suggest you contact him. Your suggestions might help him improve the work he does next.

        • JCDavies

          I’m not saying corsi and fenwick don’t have their place. They give us valuable insight in to how a team is playing, where a team excels, and where a team has deficiencies. My point simply was that lately it seems that a lot of people are using corsi and fenwick -mainly corsi – as be-all end-all for determining how ‘good’ or ‘bad’ a team is and trying to use those numbers to support their own narrative or disprove a narrative they don’t agree with.

          Corsi and fenwick do not determine outcomes of hockey games, goals do. “The process” is important but it doesn’t mean !@#$ if a team can’t put the puck in the net or keep the puck out. That’s why last year teams like Colorado, Montreal, and Pittsburgh made the playoffs and teams like New Jersey and to some extend Nashville and Florida (both bubble teams possession-wise) missed.

          To better determine a teams’ future success we need to take the emphasis off how a team controls play and start looking at how effective teams are at goal scoring goals and how effective teams are at preventing goals. Last year 14 of the 16 teams that made the playoffs had a combined GF/60 and GA/60 inside the top 16 in the league. The previous season 15 teams were inside the top 16.

          Not shot attempts (corsi for) or un-blocked shot attempts (fenwick for) but actual pucks on net (sh/60) and putting them in the net (SH%) are the only real metrics that determine a teams’ goal scoring effectiveness. Suppressing shot attempts against and blocking shots (corsi against), and goaltending (save %) are the only real metrics for determining a teams’ goal prevention effectiveness.

          A good example of using advanced metrics in a constructive manner is the following:

          Rank teams 1-30 based on 5-on-5 CA/60 and rank teams 1-30 based on 5-on-5 SV%. Combine those two ranking numbers to create a ‘goal prevention effectiveness’ (GPE) stat. Now rank all teams 1-30 based on 5-on-5 GA/60 and compare that to the ‘GPE’ ranking and you’ll see that GPE predicted (within +/-2 spots) 20 of 30 teams last year and if you expand that to +/- 3 spots you get 26 of 30 teams. By this method you can use advanced metrics to prove that – for the most part – teams that have the best combination of suppressing shot attempts against, blocking shots, and save percentages generally are the best at suppressing goals. The benefit of this is you can look at the three combined metrics to identify an area of weakness that can be improved if a team is struggling to prevent goals.

          Same goes for offence. Combine the team rankings of 5-on-5 SH/60 and 5-on-5 SH% and compare that to the 5-on-5 GF/60 and you will see that they relate fairly closely. So now you can use advanced metrics to prove that a team that is the best at generating offence is the team that has the best combination of getting shots on net and scoring on those shots.

          Based on those two examples, one can see that advanced metrics can not be used in singularity or in a vacuum, they have to be viewed as pieces to a very large puzzle.

          • JCDavies

            You make some really good points and I would love to see you or someone else explore those stats (goal prevention effectiveness’ (GPE))in more detail.

            However, like I said to NM00 above, you are using your stats to explain what has already happened. To prove the ability of a stat to predict future outcomes, you would need to demonstrate that it can consistently and repeatedly predict things that haven’t happened yet.

            “My point simply was that lately it seems that a lot of people are using corsi and fenwick -mainly corsi – as be-all end-all for determining how ‘good’ or ‘bad’ a team is and trying to use those numbers to support their own narrative or disprove a narrative they don’t agree with.”

            I agree with this. But I also think that this will always be the case. If you replace Corsi with a new and better stat, there will always be people willing to twist the new stat to fit their own narrative/agenda. In my opinion, it is best to try not to worry about them too much.

      • JCDavies

        FYI – For 2013-2014, Of the top-16 teams in score adjusted Corsi and score adjusted Fenwick:

        – 75% made the playoffs

        – 75% were in the top-16 in wins

        For 2013-2014, Of the bottom-14 teams in score adjusted Corsi and score adjusted Fenwick:

        – 71% missed the playoffs

        – 71% were in the bottom-14 in wins

        • JCDavies

          FYI the top 15 teams in 2013-2014 by goal differential (i.e. all 15 teams with positive goal differentials) all made the playoffs…

          “I see a lot of criticism of Corsi and Fenwick where the faultfinders fail to offer any better alternatives.”

          Nowhere is Rob arguing that there are “better” alternatives.

          But there is no reason to be a slave to “the best we currently have”…

          • JCDavies

            “But there is no reason to be a slave to “the best we currently have”…”

            I never said this. In fact, I think I was advocating the efforts to find better alternatives…

            “Nowhere is Rob arguing that there are “better” alternatives.”

            And that’s usually how it goes, finding fault is the easy part…

            The author of the article certainly seems to recognize the limitations of what he is doing and is trying to improve on them.

          • JCDavies

            I’d say the easy part is saying, for example, the Canucks are ranked X in shot differential metric Y which is a mildly better predictor than shot differential metric Z…

            Personally, I find it interesting when the shot differentials don’t line up with the goal or win-loss differentials and people are compelled to ask “why” as opposed to simply chalking it up to “luck”.

            This is the second year in a row (coincidentally, the only years of the NM00 era) where this has been a pertinent question in regards to the Canucks’ hot or cold start to the season.

            Last year, I’d argue (and did argue) the early season shot differentials (particularly score close) were bolstered by

            A) Vancouver spending a disproportionate amount of time down by a goal

            B) Overplaying the better players

            C) A lack of injuries that did not test the (poor) depth

            I find Rob’s final standings illustration to be a useful reminder that these metrics are not gospel.

            I’d rather dig deeper into the “why” than throw up my hands and say “these are the best predictors we have” or, even worse, chalk it up to “luck” instead of exploring further…

          • orcasfan

            Great discussion, everyone! At this point in the development of advanced stats for hockey, it seems that using them as a predictive for future performance is a better fit for individual players, rather than for teams. This may be due to just the much larger number of factors that are involved for teams as opposed to a particular player.

          • JCDavies

            “I’d say the easy part is saying, for example, the Canucks are ranked X in shot differential metric Y which is a mildly better predictor than shot differential metric Z…”

            For the record, McCurdy isn’t saying that score adjusted stats are mildly better that score close stats…

            From his conclusion:

            “Finally, and least obviously, we see that score-close possession metrics are utterly indefensible for any purpose at any time. Raw measures are preferable for conceptual clarity and for predictivity at almost all sample sizes, and adjusted measures are superior for predictivity at all sample sizes. It is difficult to overstate how important it is that they be purged from the lexicon of all right-thinking people. They purport to distill the essence of possession when in fact they do great violence to data by censoring large tracts of meaningful information and magnifying a smallish portion. Adjusted measures, by contrast, apply small nudges to the raw data—their seeming complexity masks how much closer to raw data they are than ‘close’ measures.”

            “Personally, I find it interesting when the shot differentials don’t line up with the goal or win-loss differentials and people are compelled to ask “why” as opposed to simply chalking it up to “luck”.”

            Completely agree. I fully support anyone looking into this.

          • JCDavies

            “I find Rob’s final standings illustration to be a useful reminder that these metrics are not gospel.”

            I think it worth noting that Rob’s numbers (and also mine and yours) are not predicting anything. We all used end-of-season data to explain what had already happened.

            If Rob wanted to demonstrate a predictive ability, he could pick a point in the season (say mid-season, for example) and compare the ability of Corsi, Fenwick, goal differential or any other measure to predict the second half of the season. This is what McCurdy did. If Rob wanted to prove that there are better predictive measures, he would need to do something similar.

            “I’d rather dig deeper into the “why” than throw up my hands and say “these are the best predictors we have” or, even worse, chalk it up to “luck” instead of exploring further…”

            No one is doing this…

          • JCDavies

            I’m not saying there are more predictive measures, maybe there are. I certainly do not have the time to find out.

            My only point was that too many people are looking a teams’ 5v5 CF% in a vacuum and using it to determine if a team is ‘good’ or ‘bad’ and making the assumption that team ‘A’ is definitely going to finish higher in the standings than Team ‘B’ because team ‘A’ has a higher 5v5 CF%.

            Hockey has too many variables, you have to look at shot generation, shooting percentage, shot suppression, shot blocking, save percentage, luck.

            NM00 pointed out in a post on another article that there are other factors at play that can effect a teams’ possessions numbers like score effects and spending a disproportionate amount of time up or down by a goal at 5v5 can greatly decrease or increase corsi percentages respectively.

            Corsi numbers tell us how well a team is controlling the play but there are so many variables that come between controlling play and scoring goals and scoring more goals than allowing that using one advanced metric to make predictions doesn’t make sense. To me anyway.

          • JCDavies

            “If Rob wanted to demonstrate a predictive ability, he could pick a point in the season (say mid-season, for example) and compare the ability of Corsi, Fenwick, goal differential or any other measure to predict the second half of the season. This is what McCurdy did. If Rob wanted to prove that there are better predictive measures, he would need to do something similar.”

            “You make some really good points and I would love to see you or someone else explore those stats (goal prevention effectiveness’ (GPE))in more detail.”

            Looking back and seeing what has happened in the past is the only way to draw conclusions and try to apply it moving forward. We have to look at what we know has already happened, try to get to the bottom of how it happened, and apply those same priciples moving forward to try and predict the future. We have reliable advanced metrics going back 5-6 years which is a long enough sample period to draw conclusions from and try to apply them moving forward. I am certain that somewhere in all the data there is a better way to predict how a team is going to do moving forward and I would love to explore it deeper but I just do not have that kind of time.

    • Cale

      Thanks JC.

      I’ve been hearing about these new findings, but it’s never really been discussed, clarified or elaborated upon here.

      It would be nice that a number of metrics were discussed to give a broader picture of a shorter season.

      I guess it’s my way of saying I miss the awesome work of the past from Charron and Drance. Did they set the bar too high or have things been slipping alot here lately?

      • JCDavies

        Cheers.

        I do admit that I have had to go looking elsewhere for information in an attempt to keep up.

        There is still a lot of content on this blog that I like, though.

  • JCDavies

    Whether or not McCurdy is right about that, it’s fair to say that at least when you’re trying to put something as small as 20 games under the microscope, you opt for the larger sample size of ES corsi events over the score close metrics, which in a lot of cases cause the entire third period of each game to be thrown out the window.

    But far more important is this:

    “Unfortunately, they’re PK save-percentage in November was one of the worst in the league, so the overall numbers don’t give their penalty killers the credit their due”

    Jesus… Dude, you’re killing me here. That actually caused me physical pain to read. Please have mercy.

  • orcasfan

    Don’t get me wrong, I’m not one of those guys that just brush off advanced metrics. I am 100% on board with using individual corsi stats (CorsiRel, QoC, QoL, on-ice corsi etc etc) for players. It has been a huge break though in player analysis, especially when corsi numbers are combined with zone start data. They can paint a very clear picture on how players contribute to their team’s success or failure. I’m just still a little skeptical on using corsi in a vacuum to determine a teams future success or determine how good or bad a team is. There are just too many variables in the game.