In the first part one of the series, we looked Flames forwards shot rates, mostly with a view to putting the top guys rates in context, both from a team and league perspective. It took me three tries and some helpful comments to ultimately get the chart right, but I blame excel for that.
The follow-up builds on that base to illustrate the effect of possession on individual shot (and goal rates) at even strength as well as gross on-ice totals and ultimately goal differential. This exercise puts some flesh on the bones of corsi/possession theory for those who wonder about the practical applications of that sort of advanced analysis.
The following table includes much of the data we had last week, but adds in some additional information to fill in some gaps.
|NAME||GP||TOI/60||G||SH||TS||ES ICE||shots/game||shots/60||SF/60||SA/60||total SF||total SA||shots %||Individ SH%|
After shots/60 (individual shots per 60/minutes of ice), SF and SA/60 shows the team’s rate of shots for and against with that player. Shots % is the players ratio of shots/for at even strength. Individual shot % is the percentage of total shots that player accounted for. For example, the Flames managed 187 shots at 5on5 with David Moss on the ice last year and he personally took 76 of them for an individual SH% of (78/187) 40.6%.
Finding Expected Goals
Shots% is a subset of corsi (which involves all shots at the net including block and misses) but for simplicity we’ll consider it a proxy for possession. As you can see, only Backlund and Moss we above water amongst those listed by this metric – meaning the Flames only outshot the bad guys when those two were on the ice last year in aggregate (although Comeau’s NYI results are mashed into his Flames outcomes, so take his number with a grain of salt. Ditto Cammalleri and Montreal).
The Flames big guns led the team in gross shots and goals for at ES and weren’t bad at personally generating shots/60. However, they were clearly on the wrong end of possession, suggesting a couple of things:
a.) with better possession, they could have spent more time in the offensive zone and therefore increased their shots on net/goal rates
b.) They gave up a lot of shots/goals against, limiting the value of their own totals
To model how an improved possession rate would affect those issues, I took the exisiting shot totals, assumed constant invidual shots % (ratio of personal shots to total shots for), a possession rate of 55% and applied league average shooting and save percentages to the results.
Here’s how things shake out:
|Player||Individ SH%||Shots %||total SF||delta total SF||expected TS||delta TS||expected goals||expected GF delta||total SA||delta SA||expected GA delta||expected GD delta|
The invidiual SH% and total shots while the player was on the ice last year remain constant. The shots % of 55% is a hypothetical possession rate for each guy. The total shots for is how many shots that team would have had with each player on the ice given a 55% posssesion rate. The delta SF shows the difference between this total and his actual SF total last year.
Expected TS and delta TS are similar, except they are for each player’s personal shots on net totals and are arrived at by multiplying his indvidual SH% with the new total SF. “expected goals” shows the additional goals each player would have scored given a 55% possession rate and league average ES shooting percentage. “Expected GF delta” shows the additional goals the team would have scored with the player on the ice given the 55% possession rate.
The process was repeated for shots and goals against. The expected GD delta shows the difference in expected even strength goal differential between existing shots ratios and hypothetical shots ratios, assuming league average SH% and SV%.
Lots of numbers here, but the important columns are the delta’s (or “difference”) which show how many more shots/goals each player could be expected given last year’s shot totals and if his possession rate had been 55% (and stable, league average percentages). In the case of Jarome Iginla and Olli Jokinen, for instance, they could have personally generated 40 and 36 more shots on net respectively had they managed possession rates of 55% (rather than 45%). In terms of total shots, the team could have accrued 154 more shots with Jarome on the ice and 116 more shots with Jokinen skating last year as opposed to what they actually managed.
As a result, Jokinen and Iginla would be expected to score 4.3 and 3.8 more even strength goals each with a 10% shift in possession. In terms of total goals for when they were on the ice, the expected goal difference is 12.6 and 9.5.
In contrast, the team would have given up 119 and 116 less shots against with each guy on the ice last year, saving the club 9.49 and 9.28 goals against at 5on5. All told, not only would both guys have scored about 8 more ES goals personally, but the team would be expected to improve their goal differential by more than 22 in just Iginla’s case (which is worth about 3.5 expected wins or 7 points in the standings).
The effect is obviously less pronounced for guys with less ice time. The cnage for Backlund and Moss is especially muted because both players didn’t play much and were close to a 55% possession rate anyways.
– Keep in mind an indvidual 55% possession rate is excellent-to-elite in the NHL (but certainly not impossible).
– I’ve assumed personal percentage of total shots on net as a constant from the season sample, but it likely varies year-to-year somewhat. That said, Iginla is usually at 30% or above, while the stingy Tanguay is usally below 20%. Moss’ number is skewed by a small sample size.
– This inquiry used league average even strength personal and on-ice ES SH% and SV% to control for assumed randomness in the actual results. So that is 10.53% SH% at even strength for forwards last year, and 0.08 on-ice SH% and 0.92 on-ice SV%.
This could be done for each player using his real 2011-12 on-ice SH% and SV% however. For example, Iginla personally shot at 12% at ES last year, while his on-ice ES SH% was 10.10 and on-ice SV% was .913. That expands his expected personal goal delta to 4.9, his expected goals for delta to 15.5 and expected goal differential delta to 25.83 (!!) – or 4.31 wins/about 8.5 points in the standings. .
If people have interest in this sort of comparision, I can re-run the results and post them later.