goal differential

[flyingnote38]

the western conference standings coincide quite well with rankings based on goal differential. There are 8 western conf teams with positive goal differentials and they are the 8 currently occupying playoff spots. In fact the top 3 teams in each division rank exactly the same based on goal differential. The only slight blip is Minnesota edging Dallas for the 7th seed despite a slightly worse GD (+2 vs +6)

then........ there's the east.

Boston has the second best GD in the league despite but only has the second best point total in the east and 6th best in the league. The Bolts rank as the East's third best team based on GD (+16) but currently sit 4th in their division behind Boston, Montreal(E) and Toronto (-5). Ottawa is still in the playoff hunt, 4 points behind the 8th spot, despite a -27 GD 'good' for 13th best in the East and nearly a third of a goal per game worse than tenth place NJ.

The Bolts are chronic GD outliers. Last season, they finished with the third worst record in the league despite a goal differential of just -2 which was better than 13 teams.

Goal differential did a fair job of predicting playoff seeding last season (ignoring certain division winners' promotion to 3rd seeds) and even reflected the closeness of the races decided on the season's final day. In the West, the Jackets +1 should have bested the Wild -5 for the last spot but were eliminated instead on the last day. The Blues +14, Kings +15, and Sharks +12 entered the final day with almost any combination of 4th, 5th and 6th seeds on the table reflecting how comparable they really were. In the East, things followed better last season than this with the 6th seed Rangers +18 and 5th seed Leafs +12 being flip-flopped based on GD as the only failure of standings not to coincide with GD.

[cardsfan04]

I think goal differential is a pretty good measuring stick of a team's true ability, possibly moreso than record. In baseball, there is a stat called pythagorean record. It is (runs scored^1.81)/((runs scored^1.81)+(runs allowed^1.81)). Basically, doing that gives you the predicted winning percentage for a team. It's fairly accurate, but definitely has a number of outliers.

I just used that to calculate the Blues' pythagorean record and got 83.58 points. That seems like an unusually large discrepancy between their actual record and pythagorean record (most baseball teams are within 5 games over a 162 game season, so 8.42 points (or 4.12 wins) seems pretty far off in 63 games. My guess is that the possibility of 3 point games makes 1.81 the wrong exponent to use for hockey, but I don't know how to come up with the correct one.

[cardsfan04]

Apparently, the exponent is slightly above 2 for hockey.

http://en.wikipedia.org/wiki/Pythagorean_expectation

2.1 puts Blues at 86.58. Still a bit of an overachievement according to that, but a bit closer.

[flyingnote38]

I think the issue is that it predists wins rather than points. Using 2.1 as a stand in for 'slightly above 2' yields an expectation of 43.95 wins for the Blues in 64 games. Both the the Ducks and Hawks project to 41 wins in their 64 and 65 games respectively. So the Ducks have won a couple more than they should have and the Hawks have lost a couple more than they should have.

Tampa projects to 35 wins in 64 (pretty close to their 34). Toronto to 31.6 in 65 (so they have relatively overperformed at 34)

Interesting. The article your link cited was mystifying garbledygook but the formula does seem to function adaquately.

win prob = 1 / (1+(ga/gf)*2.1)

the point totals are obviously heavily influenced by how the teams achieved their loses.