Y2YWCF: A Quick Note on Regression to the Mean

PART I:
"They gained Kevin Durant and they're going to win LESS games? No way!!"
-BETS FURIOUSLY THAT GOLDEN STATE WINS MORE THAN WHATEVER VEGAS HAD THEM AT-


This logic on its own is incorrect for many reasons. We will call it the Year to Year Win Correlation Fallacy (or Y2YWCF* for short).The most important reason this statement should be is because of is regression to the mean.

Single-year-win totals are subject to variance - for this example, we would statistically assume Warriors had some things go their way they probably won't have this season.

Here's the breakdown of year-to-year win totals, via basketball reference, since the 30-team ERA, (2011 adjusted for lockout):

(By prior year and following year, I just mean one year and the following year)

Using this chart, we can see how a typical team regresses.




In fact, a 73-win team would be expected to only win 61 games the following season!
And from a basic stats perspective**, adding Kevin Durant to a stacked team would not net an additional 12 wins.

In this dataset, 49% of all NBA teams regress towards 41.




PART II:
"How many games will the Miami Heat win this season"
Ok. Here's some projections:

Year to year correlation (from the formula used in Part I):
45.4

My initial projection (a blend):
35.15

Most rating systems obviously had the Heat lower after losing their best player of ever, Dwyane Wade, for pride reasons. His BPM rating put his wins above replacement at +4.32 last season.

Miami also will probably have zero playing time from Bosh for (mostly?) health reasons, who was +4.86 Wins Above Replacement:

Oh yeah also Luol Deng is a Laker now. Last season he had +6.21 Wins Above Replacement for Miami.
FiveThirtyEight's Projection after 1 Win:
35


So....35 is still looking pretty good. Unless you're a Heat fan.




*- I was hoping for something catchier GSBLEWA31LEAD but I couldn't find a good way to make it fit.

*- Also from an advanced stats perspective (NetRating). The Warriors' Net Rating last season was lower than we would expect from the 73-win GSW team last yr (or you could say, their Win% was higher than their NetRtg expects). This is another important factor that we expect to regress, which I kind of mentioned at the beginning.

2016 NCAA Tournament Bracket Cheat Sheet (UPDATED 3/17)

2016 NCAA Tournament Bracket Cheat Sheet (UPDATED GOOGLE DOC 3/17)

Each team is ranked by their expected # of wins.

-Odds come from 10,000 simulations of the tournament, taking into account injuries, Four Factors adjustments, pace, and if the teams have a very close rating - free throw%.  I have updated the sheet to include an adjustment for shortened lineups (i.e. best players playing).

-Sleeper Rating is a rating from 1 to 5 that measures total volatility above Expected Wins, with successive rounds being weighted twice as much (i.e. using ESPN scoring). It's a good measure for teams with statistical ability to go further than my Exp. Wins suggests.

LARGE POOL PICK VALUE = (a team's odds  -  the public's expected odds )x total point value
This highlights high-value picks versus the field (namely, ESPN's bracket challenge)

VALUE + EXPECTED POINTS = Total Point Value + Large Pool Pick Value

This is a slightly more conservative version of "Large Pool Pick Value"

All statistics from kenpom.com except injury estimates from fivethirtyeight.com

Enjoy responsibly!