Sep 27, 2021 11 tweets 7 min read
I'm seeing a lot of questions in @GameFlowxPG about the #RCTID and #RSL game, so I'll wade into the deep end here.

The @AnalysisEvolved xG model has this game 1.8 to 2.47 for RSL.

@FotMob has it 2.07 to 1.69 in favor of Portland.

The #WGCF model below is barely Portland.

1/
Think of xG models like judges in a boxing match. One may score a shot one way and another a different way due to their own perspectives (both strengths and blind spots).

Usually they agree on the final winner, but often it's a split or majority decision.

2/
Although one judge may see a shot differently, in aggregate, good xG models will usually agree about a team over a season, and typically even over 5 to 10 games.

Models typically improve over time, if there is acknowledgment about their blind spots and how to improve them.

3/
Now @GameFlowxPG (which conceptually @Tacticsplatform and I helped create, but practically is the work @etmckinley) was moved from using our earlier xPG method to @AnalysisEvolved's g+ last season.

g+ is not a shot model like xG. It is a *possession value model*.

4/
Cumulative xG and GameFlow are not the right tools for measuring the #PORvRSL scenario.

You need a Win Probability Model that understands game state for that.

For xG race charts and GameFlow, anything after 3-1 or 4-1 in the early second half, is basically 🤷‍♂️.

5/
.@GameFlowxPG shows the maximum *net* cumulative g+ value (i.e., maximum goal probability) of each minute based on the possession.

Each team may possess the ball multiple times in a minute, but the wicks are the net max probability for the stronger team *in that minute*.

6/
The best way to think of GameFlow is as a momentum meter. Teams usually have the best chance of scoring when they have a wick longer than 0.15 but the minute-by-minute wicks can do funny things.

Goals can belie current momentum, but not usually in that minute.

7/
Neither an xG race chart nor an Offensive g+ number have any idea about game state. "Goals change things."

Unfortunately for RSL, they gave up two goals first. Now they have to throw numbers forward. Now Portland can counter-attack against fewer defenders. Momentum lies.

8/
Personally, I like to count the longer wicks over 0.1 or 0.2 in @GameFlowxPG. They are the best indicator for me of danger created with or without a shot.

Revisiting #RCTID and #RSL, that goes 6-3, which is a decent indicator of the potential score given the better chances.
So effectively:
* @GameFlowxPG = minute-by-minute maximum possession momentum using g+ offensive values (with a comparison to @AnalysisEvolved xG included for fun). Count longer wicks for best chances.
* xG = shots x avg shot quality

More on g+ here: americansocceranalysis.com/what-are-goals…

9/
More on the Where Goals Come From xG model here: americansocceranalysis.com/home/2021/8/12…

/end

• • •

Missing some Tweet in this thread? You can try to force a refresh

This Thread may be Removed Anytime!

Twitter may remove this content at anytime! Save it as PDF for later use!