MacDonaldвЂ™s model appears to have been one of the first in hockey that really caught on вЂ“ or at least the idea.

He presented this at the 2012 SLOAN conference, where he expanded on the article and provided some important references and notes. This model was never made completely public, and is (as far weвЂ™re aware) not currently available publicly. These methods/approaches worked alongside the generally simpler concept of вЂњweighted shotsвЂќ, explained in Tom TangoвЂ™s write-up here.Д«rian MacDonald developed an xG model hereВ that expanded on this prior work. There are a few additional articles on Alan RyderвЂ™s website hereВ as well. Following this, he wrote about some of the issues with the (then new) RTSS data here, which then led to further work from Ken Kryzwicki hereВ and here, and Michael Schuckers hereВ and here. The first version of this method appears to have been developed by Alan Ryder in 2004 here. The earliest versions were often described as вЂњshot qualityвЂќ models, attempting to more or less вЂњweightвЂќ shot attempts using the data that was available. All of these models have their roots in work done over a decade ago. The models/metrics weвЂ™re familiar with today generally use multiple seasons of RTSS data (NHL play by play data) and often include over 20 variables as inputs. xG now sits neatly next to Corsi and Fenwick as a metric that is quite useful for both team and player evaluation.

Note: this article was originally published on Д®xpected Goals (xG) as a metric has become common-place in hockey analytics circles over the past few years.

Nhl 2004 rebuilt 2015 pbp