Back in 2009, Tyler Dellow first elaborated on the idea of what we now call “score effects,” or how teams with a lead will go into a “defensive shell” and purposely withdraw from the possession battle to preserve their score. Score effects are the primary reason the go-to possession stat is “Fenwick Close” today – the “close” implies the importance of looking at possession measures when teams still have a reason to engage. The limits of historical shot recording, and the possibility of score effects, are precisely why I’ve advocated the use of 2pS% (shot-differential percentage from the first two periods) as an historical possession measure.

The one thing I never completely took for granted was that score effects had always existed in the NHL. To test this, I broke down each game into individual period shot battles, and looked separately at the correlation* of 1st, 2nd, or 3rd period shots-for percentages to final goals-for percentages. The result above clearly shows that the 3rd period SF% begins to drop away drastically after 1977 or so, after a quarter-century of running pretty close to the others. It does seem possible, then, that the re-introduction of overtime in 1983-84 (gone since 1943-44) had an impact on the growth of score effects (although I’m not sure how); on the other hand, the introduction of the “loser point” in 1999-2000 doesn’t seem to have had any effect. We can also do a similar graph of correlations to goals-for percentage to validate the use of 2pS%:

As you can see, score effects have essentially become the norm, much to the detriment of overall shot differential. At any rate, whomever put two-and-two together back in the 1970s probably had the right idea; I’d forward the hypothesis that the 1970s NHL was ripe for change and innovation (a lot of competition; growth of league = increase in decision-makers and opportunities to exploit market inefficiencies). In that kind of environment, protecting the lead quickly became a best practice, and it steadily grew to a league-wide practice by the mid-1990s or so.

* Or a -1.0 to +1.0 relationship of the variance in one variable to the variance in another; positive means as one goes up, the other tends to go up, suggesting a positive relationship or correlation. A negative correlation suggests that, as one goes up, the other tends to go down. The closer to 0.0, the less likely the variables have any relationship at all.