Am I the only who thinks that looking at each teams performance over the first 8 games and the next spell of 8 games is actually pretty cool? Just me then.
8 game splits is a tiny sample to work with, so any information is merely helpful rather than definitive. Hell, even 19 game splits would be an insufficient sample to work with.
What will follow is a collection of stat tables and the repeatability of each stat from the first 8 games of the season to the second 8 games of the season. Each group of 8 games is separate and none of the information overlaps. I must also state that this article is merely foreplay to a main event which will look at which stats are the best predictors of future goal difference or points totals.
It's all well and good knowing which stats are repeatable, but what we really really want to know are which stats are predictive of future outcomes in terms of the aforementioned goal difference and points. But that's for next time.
OK, we'll start.
|Points||Games 1-8||Games 9-16|
Southampton, West Brom and Fulham were the teams who dropped off to the largest degree in the first 8 games v the next 8 games. Palace improve.
Pretty meh relationship between the two sets of results. I'd include goal difference at this juncture but the results are too similar.
|Shots +/-||Weeks 1-8||Weeks 9-16|
Palace improved, Liverpool blew the doors off. Tottenham, Chelsea and Man United regressed. Quality of opposition may well be a big factor here.
R2 of 0.55 isn't too bad at all.
Shots On Target +/-
|SoT +/-||Weeks 1-8||Weeks 9-16|
Huh, Cardiff, Palace and Sunderland all improved in the second chunk of 8 games.
Tottenham, Swansea and Man United regressed.
Slight improvement on the predictability when using shots on target instead of shots. Still early in the season, though.
Scoring=goals for/shots on target for
|SC%||Weeks 1-8||Weeks 9-16|
Scoring% is pretty much all over the place. Here's how the first bucket of scoring% predicts the second bucket of scoring%:
A correlation of 0.01 between two sets of variables is really poor. Essentially there is no relationship.
Save= 100-(goals against/shots on target against)
|SV%||Weeks 1-8||Weeks 9-16|
Again, this is pretty much all over the place.
A negative slope to this chart. A negative correlation is probably best explained like this:
The more time I spend at the mall, the less money I have. The more time I spend at the office, the more I question the purpose of existence.
The higher a team save% was in the first 8 games the lower it is likely to be in the next 8 games. But that effect isn't huge, the correlation is just r2=0.22.
Save% and Scoring% numbers in the first 8 games were shitty predictors of Save% and Scoring% in the next 8 games. What happens if we put both of these stats together and call it PDO?
|PDO||Games 1-8||Games 9-15|
In most cases PDO shows little or no relationship from the first group of 8 games to the next group of 8 games.. The chart points this out:
A very slight negative correlation. But it's a wash really. Many people will point to our advanced understanding of shots location and their importance in separating teams from their basic numbers. Maybe they have a point, but the early season splits, and the r2 valuations in scoring%, save% point to those stats having very little repeatability league wide.
I think Mike Goodman put it best:
Putting all this information together we end up with a table like this:
|SoT For %||0.42|
I could have included around 15 or 20 other stat categories, but I think we've all seen enough for now.