Monday, January 5, 2009

Playing to the Score

The outshooting and scoring numbers for the 1980s Oilers were so interesting I was inspired to gather similar data for other top goalies of the modern era. Using the same method as I did for Grant Fuhr, and restricting my sample to the playoffs only (because that is when teams really play to win), I collected shot and goal data for the first two periods combined, and then for the third period and overtime (if any). I then compared the results based on game score to see how teams played to the score. The data are not complete, because the Hockey Summary Project only has playoff data from 1994 to 2008 (with 1997 missing), so the numbers will not exactly match the goalies' career totals.

The goalies in my sample were Patrick Roy, Dominik Hasek, Ed Belfour, Curtis Joseph and Martin Brodeur, the same goalies I had included in a previous study of playoff performance, so I was hoping to revisit that analysis with some updated information as well as to test the effectiveness of some of my estimates. I will get to the individual numbers in another post shortly, but first I'd like to present the aggregates because they have value in addressing some fundamental hockey questions about how teams play to the score and the effect of strategy on goaltender play.

1st two period stats:

28 SF, 27 SA, 2.52 GF/60, 2.16 GA/60, 9.0% SH%, .920 SV%

All of the 5 goaltenders had a lot of wins in their careers, and this line shows why. Some of the goalies played on outshooting teams while others were outshot, and some of the teams were better at scoring than others, but overall the percentages worked out to an average edge of a 1% higher scoring rate, which accounts for 0.27 goals per game. The rest of goal differential comes from the slight advantage in outshooting.

3rd period stats:

These third period stats are for the entire third period that began with a specific score. Ideally we would like to count based on actual game score throughout, rather than just at the start of the period, but that requires play-by-play data. My "box score method" is much simpler, but achieves it through a loss of accuracy. However, there were relatively few lead changes in the third period, especially since most of the sample was during a low period for goalscoring league-wide. The strength of the observed effect, even in this limited sample, is strong enough that I feel confident that these results do approximate the real effect. I would certainly encourage someone with the programming chops to break it down in exact detail for some recent playoff seasons to see if my results are typical.

When leading by one goal after 2 periods:

21 SF, 29 SA, 2.36 GF/60, 2.03 GA/60, 11.8% SH%, .928 SV%

When up a goal late in the game, the teams drastically cut back on offence. Similarly, the trailing team started putting more shots on net. Interestingly, however, save percentage went up and so did shooting percentage, implying that the trailing team is taking more shots of lower quality, whereas the leading team is generating higher-than normal shot quality. Another possible explanation for this effect is that the leading goalies simply played better (by focusing more, bearing down more, etc.) in this important situation. However, I would imagine that the goalies who were down 1 would also be highly-focused, and yet their save percentages went down almost 3%. Whatever the explanation, this resulted in a huge percentage gap (almost 5%) in scoring rates, which made the leading team actually slightly more likely to outscore its opposition despite the one-sided shot differential.

I doubt that is generally the case (remember, we are only looking at strong goalies here, and any save percentage below .919 would result in the leading team getting outscored), however it certainly helps to have a great goalie when you are trying to hang on to a one goal lead late in the game.

When tied after 2 periods:

26 SF, 25 SA, 2.34 GF/60, 2.08 GA/60, 9.2% SH%, .916 SV%

When tied, the outshooting results are similar, with some evidence of teams playing more cautiously. Many of these periods started out tied but then had one of the teams take the lead, so could be quite different than what the picture would look like if we looked at just the time when the game was actually tied. We will, however, see that situation that when we look at the overtime results.

When trailing by one goal after 2 periods:

30 SF, 22 SA, 2.64 GF/60, 2.01 GA/60, 9.0% SH%, .916 SV%

When trailing, the teams started throwing pucks on net and taking greater risks offensively. The scoring rate was identical to the rate in the first two periods, although save percentage was slightly lower. We would need to look at a larger sample to see if the shot quality for the trailing teams is generally lower. I would suspect that it might be, but I am not sure. The evidence does appear to be there that the leading team will probably have higher than average shot quality for. The goalies' save percentage was .004 lower than during the first two periods combined, and, and earlier we saw how the shooting percentage went up almost 3% when the teams were up by one.

This shows again why these teams were winning a lot of games - their scoring rates were better than the opposition's in all three situations. Part of this is probably that the teams sampled were more likely than usual to tie the game up and therefore some of this time would have actually been played with the score tied. However, the team leading after two periods had a very good record (86% of the time the team that was leading after 2 periods ended up winning the hockey game).

Overtime stats:

28 SF, 28 SA, 2.03 GF/60, 2.14 GA/60, 6.7% SH%, .926 SV%, 14 W, 14 L

In overtime, save percentage rose and shooting percentage went way down. It was interesting that these goalies' teams were likely to get outscored in overtime. Even though all the teams had a good goalie in net they were just as likely to lose as to win once the game went into the fourth period. This suggests that overtime is quite random. Note: Patrick Roy was a major outlier in terms of overtime results through the course of his career, but most of his overtime outperformance came in Montreal pre-1994 and is not included here.

Outshooting results:

Given how shooting results are apparently strongly dictated by the scoreboard, shooting results for an entire game are not particularly useful. If a team dominates the first two periods and goes up a goal or two entering the third, they are likely to sit back while the other team takes a lot of long-range shots in the third. The shots might have been something like 25-15 through two periods, but then get reversed 13-3 in the third, and both teams end up with 28 on the night even though Team A was clearly the better team. I observed many games with that type of distribution when going through the box scores.

These effects were less strong in the first two periods, however, so we can look at how outshooting determined the results over those two frames. I calculated a "winning percentage" based on the score after 2 periods, giving 2 points for being ahead and 1 point for a tie game.

When outshooting the opposition, the teams had a .635 win percentage. When outshot by any amount they were just .481. Given that we are only looking at the first two-thirds of a hockey game here, that is a lot of evidence in favour of outshooting driving results. Even with a Hall of Famer in your net (Joseph is debatable, but let's just assume he is for the sake of a more punchy sentence), evidence suggests that your team is more likely to be trailing than leading if you get outshot over the first two periods of a playoff game. TV analysts take note: The best way to predict a playoff series is not to simply pick the team with the better goalie, but to pick the team more likely to have the edge in puck possession and shots. Certainly teams can have the percentages go their way and ride a combination of a hot goalie and goals against the run of play for a few games or even a series or two, but past history suggests that the the inevitable hand of regression nearly always shows up to end the fairy tale short of the final prize.

11 comments:

JLikens said...

Highly interesting.

One question: did you filter out empty net goals when doing the analysis? If not, this might (partially) explain why the team leading at the end of the 2nd period experiences an increase in its shooting/save percentage.

In any event, to the extent that the effect is real (which I believe it is), it appears that it's easier to have higher save percentage playing for a good team, as a stronger team will tend to be leading much more often over the course of a season than a weak team will.

The Contrarian Goaltender said...

I eliminated empty net goals allowed, as they were easy to check against the goalies' stats, but empty net goals scored are still in there. You are right, that would have had an effect on the shooting percentage for the leading team.

I'm still not sure about the overall save percentage effects of a strong team (if any). It apparently helps you in the third period, but I'm not sure about earlier in the game. One thing I plan to look at in more detail with this sample is the relationship between number of shots against and save percentage during the first two periods, when the score isn't likely to be dictating the style of play as much.

N said...

I really must thank Matt from the Battle of Alberta blog for linking to you. Your stuff on Brodeur is terribly amusing; the hockey equivalent of the film Loose Change. Keep it up!

Anonymous said...

Just a suggestion. Since it seems you have editing powers as the owner of this blog, it wouldnt hurt the credibility of your research to do a better job editing and perhaps cleaning up this site. There is a lot of excellent research(albeit in some cases biased) yet this site appears to be more of a troll harbor than anything. I have no idea what your purpose for doing all of your research would be, but in any event i dont see how it makes sense to do hours of writing and analysis, only to have a few childish trolls constantly quarreling with each other, or other users who would like to engage in adult conversations.

Anonymous said...

Just a suggestion. Since it seems you have editing powers as the owner of this blog, it wouldnt hurt the credibility of your research to do a better job editing and perhaps cleaning up this site. There is a lot of excellent research(albeit in some cases biased) yet this site appears to be more of a troll harbor than anything. I have no idea what your purpose for doing all of your research would be, but in any event i dont see how it makes sense to do hours of writing and analysis, only to have a few childish trolls constantly quarreling with each other, or other users who would like to engage in adult conversations.

Bill Morran said...

I don't know how you have Joseph listed as "debateable". This guy carried the Leafs during their best years since 1967 was one of the top goalies in the NHL for more than 10 years. His numbers are incredibly compareable to the other four, but he played for worse teams.

If you're listing him as "debateable" because of how you feel his chances of getting in are, and not what the statistics tell you, then you'd still most likely be wrong. Even if he was a border line HOFer statistically, East Coast bias and Leaf love from Hall of Fame voters will get him in.

The Contrarian Goaltender said...

I listed him as "debateable" because people who know more than I do about HHOF procedures seem to think he is a borderline candidate and may not be voted in. I think Cujo was a very good goalie and one of the 5 best goalies of his era, so he'd probably get my vote.

The Contrarian Goaltender said...

Anonymous: Thanks for the suggestion, I'll keep a closer eye on some of the comments I'm getting, especially on older posts.

James Benesh said...

Great info. The only comment I was going to make, that I caught right away, is the empty net goal factor. This is definitely having a large effect on the shot percentage of the team leading going into the 3rd.

The Contrarian Goaltender said...

I'm going back through the data to remove all empty net goals before I post the individual goalie stats.

Bruce said...

CG: I commend you for interesting and important work.

When up a goal late in the game, the teams drastically cut back on offence. Similarly, the trailing team started putting more shots on net. Interestingly, however, save percentage went up and so did shooting percentage, implying that the trailing team is taking more shots of lower quality, whereas the leading team is generating higher-than normal shot quality.

I think there is a great deal to be learned here about both shot quantity and quality, a subject I have raised many times in this and other fora.

Your methodology -- score after 40 minutes -- is crude but nonetheless should capture the essence of the relationship between game state and shots.

My multiple studies on the subject have shown there is only a mild relationship between outshooting and winning, something under 55% on a league-wide basis. The correlation is actually negative for some teams, notably my Edmonton Oilers throughout most of their history, from the storied dynasty of the 1980s to the battling-for-8th Oilers of the Lowe-MacTavish era.

At one point the discussion with the IOF types -- who have deep insights into the issue even as they may not (yet) have all the answers :) -- carried into the realm of game states. MC79 among others spoke of the importance of special teams shots, while I countered that I thought game score was even more important. Namely that the trailing team generates more shots, often way more, while allowing higher quality chances against.

Vic Ferrari stripped out shots by game score from the play-by-play, but the only numbers that he posted (that I saw) were shots when the game was tied. As I recall that was pretty reflective of shots overall. Where I would expect a big effect would be shots when the score is NOT tied, and your data addresses this. Imperfectly -- esp. until those empty-netters are accounted for, which expressed as a percentage of goals scored in the third period by the team leading by one after the second, would be significant -- but the effect is clear. The trailing team carries the play, but isn't necessarily rewarded for doing so.

All of which (still) draws into question the relevance of such shot-based measurements as Save Percentage and Corsi numbers. Does a team like Columbus Blue Jackets have great Corsi numbers because they're a great team, because they've got a system-oriented coach, or because they're always trying to come from behind? (Give the link a minute to load, but it's an eye-popper when you get there.)

If you're listing him as "debateable" because of how you feel his chances of getting in are, and not what the statistics tell you, then you'd still most likely be wrong.

Bill Morran: On this we entirely agree. Whatever the "cognoscenti of the cord cottage" might think about Wins, 450 of them is enough to impress the voters. In terms of rarity it's roughly the equivalent of 700 goals, no mean feat in any era. I would myself vote for Joseph, but not on the first ballot (which I would reserve for truly extraordinary players, which he is not).