Thread by @ChartingHockey

We're back at #OTTHAC18 -- @timoseppa, @bmacGTPM, @SchuckersM, and @IneffectiveMath are on a panel discussing hockey modelling. I assume they'll have a catwalk portion at some point.

Starting with a talk on WAR
@bmacGTPM: regression-based models that account for on-ice players, control for linemates, strength of opponents, shift adjustments.
@SchuckersM: goal is to isolate impact of specific players while accounting for impacts of others

Schuckers: the shift, rather than the event, is probably the right basic unit for study in WAR/GAR models.

#OTTHAC18

Seppa: WAR is a descriptive measure.

McCurdy: Statistics can be for reveling in the past (@DownGoesBrown styled look at stories that are interesting). This is valuable.
Predictive model is more for people looking to work in hockey -- to help with million-dollar decisions.

McCurdy: models all have the same elements to different extents but the goal changes what you look for --> descriptive model of the past or predictive model to guide future decision-making.

#OTTHAC18

MacDonald: You don't want to use only WAR models in the same way you don't want to use only watching games to make decisions.

Stats might help identify a player worth watching more.

Most of the time, smart analysts and smart eye testers are going to agree. #OTTHAC18

MacDonald: An interesting small subsection is the players where models and eye test disagree. Is the model wrong? Are the scouts wrong? This is where interesting/good discussions about players can start.

#OTTHAC18

MacDonald: Both things (models and eye test) can work together. That's the way it should be.

#OTTHAC18

MacDonald makes reference to HR "WAR" models -- where HR depts conduct WAR analysis to determine who the best employees are. For example, "revenue above replacement."
I.e. How good was a sales rep relative to their "leads"