First, let me quickly address the tragedy of my college
basketball picks last season. I’ll be brief. No one wants to hear excuses. I’m going to call it the “demo effect.” Anyone who has ever performed a demo that
utilizes technology knows, for whatever reason, the technology seems to stop
working when the lights are on and people are watching. Maybe I felt the pressure and subconsciously altered
the way I was picking. Maybe last year
was an anomaly. No one really cares what
happened at this point. The facts are 1)
the picks were free AND 2) there are many “experts” out there having losing
seasons who are selling their picks to
countless willing buyers. Be grateful I
spared you any expense!
Enough about that – Let’s compare the NET, RPI, and Kenpom
rankings for the 2018-2019 season through end of day 12/11/2018. At the bottom of this post there is a line chart showing the
top 60 Kenpom teams and their corresponding NET and RPI rankings. (I made it as big as I could within the limitations of blogspot. Please copy and paste it to Excel, Word, or another program and resize it for your viewing pleasure.) As you can see, the vast majority of NET data
points are correlating more closely with Kenpom than RPI. (Quick aside: You might ask why I’m including
Kenpom rankings in this comparison. The
Kenpom rankings are widely regarded by those in the know to be the very best predictive
rankings available. If one team has a
better Kenpom ranking than another, that team is favored in Vegas and is more
likely to win a game against the other.)
It’s still too early to be sure, but I think we’re headed in the right
direction folks!
With any data comparison there are going to be outliers, especially
with a limited sample size (Few if any teams have played more than 11 games
this year and by year’s end will play roughly three times that number of games.). Four teams I’d like to put under the microscope
are North Carolina State, Houston, Arizona State, and San Francisco.
NC State
|
Houston
|
Arizona State
|
San Francisco
|
|
Kenpom
|
27
|
36
|
43
|
51
|
NET
|
18
|
14
|
26
|
21
|
RPI
|
162
|
64
|
56
|
57
|
Why am I looking at these four teams? There’s two big reasons.
- All are either likely NCAA tournament teams or at least bubble-worthy based on their level of play so far this season.
- All have a Kenpom ranking that is better than their RPI ranking and worse than their NET ranking meaning there is a large disparity in the ranking from the system used last year (RPI) to the one in use this year (NET).
So let’s analyze the differences in the rankings. Doing so will help us understand how we can
expect the NET rankings to adjust the rest of the season. This gives us some idea about the chances of
our favorite teams making the tournament.
These four teams have two glaring commonalities:
- They all have high win percentages (at most one loss each).
- All of the losses for these four teams are to other teams with good NET rankings.
- Nevada = 8
- Wisconsin = 11
- Buffalo = 12
The enormous flaw in the RPI was it put far too much
emphasis on how many wins your opponents had (50% of the product) and how many
wins your opponents’ opponents had (25% of the product) while almost completely ignoring the strength of those wins.
So what is NET doing differently and better?
First, we look at the four factors that make up the NET formula.
- Game Results Algorithm = “set up to reward teams who beat other GOOD teams.”
- Net Efficiency = summation of how efficient you are on offense and defense
- Winning Percentage
- Adjusted Win Percentage = Giving more credit for road wins vs. home wins
I see no component in the formula that rewards a team for beating
lots of relatively crappy teams who end up with decent or good winning
percentages because they play in crappy conferences. I DO see two things I like: valuing beating truly good teams and measuring how efficient a team's offense and defense is performing. ONIONS!
One key question to consider from here on out as our four
spotlighted teams head into conference play is “How strong is the Game Results
Algorithm?” NC State is in a very good
conference. San Francisco is in a relatively
bad conference. Arizona State and
Houston fall somewhere in between. At
season’s end, if these four teams finish with fairly similar records we will
have a great understanding of how heavily weighted the Game Results Algorithm
really is by comparing their final NET rankings to the ones in the table above. If their records end up being fairly similar
AND their NET rankings do as well, there’s a strong inclination the Game Results
Algorithm isn’t very strong. If NC State’s
NET ranking improves relative to Arizona state and Houston relative to San
Francisco, we know the Game Results Algorithm is more heavily weighted.
All in all, no one can argue the RPI is a better system than
the NET Rankings. My personal opinion is
they should simply use the tried and true Kenpom Rankings rather than inventing
a brand new system, but I suppose admitting some guy came up with a great ranking
system 16 years ago that only now have you tried to emulate may not be in your
best interest. Progress is progress.
*Note
*Average difference in
Kenpom Ranking vs. NET Ranking for all 353 D1 teams = 26.2
*Average difference in
Kenpom Ranking vs. RPI Ranking for all 353 D1 teams = 43.5
