[IBL] Comprehensive Playoff Odds - October 7, 2016

[Chris Hartjes]

I'm back!

So I took a look at my code and found the extremely subtle bug involving using
head-to-head matchups. But upon reflection I felt like we are dealing with sample
sizes that are too small to be accurate enough. Here's the results when using the
fixed code and head-to-head results:

10000 simulated remaining seasons using head-to-head matchups


AC East
COU 98.8% Div + 1.2% WC = 100.0%
NJR 0.0% Div + 0.0% WC = 0.0%
TRI 1.2% Div + 75.2% WC = 76.4%
KAT 0.0% Div + 0.0% WC = 0.0%

AC Central
PHI 99.7% Div + 0.3% WC = 100.0%
PAD 0.0% Div + 0.0% WC = 0.0%
WMS 0.3% Div + 99.7% WC = 100.0%
BOW 0.0% Div + 0.0% WC = 0.0%

AC West
MCM 0.6% Div + 2.4% WC = 3.0%
POR 12.6% Div + 6.5% WC = 19.1%
BAL 23.4% Div + 7.7% WC = 31.0%
MAD 63.4% Div + 7.0% WC = 70.4%

NC East
MNM 99.9% Div + 0.0% WC = 99.9%
SKY 0.1% Div + 0.0% WC = 0.1%
CSG 0.0% Div + 0.0% WC = 0.0%
SDQ 0.0% Div + 0.0% WC = 0.0%

NC Central
CAN 21.1% Div + 5.9% WC = 27.0%
SFP 78.8% Div + 1.8% WC = 80.6%
HAV 0.0% Div + 0.0% WC = 0.0%
NYK 0.0% Div + 0.0% WC = 0.0%

NC West
ODM 16.1% Div + 77.8% WC = 94.0%
SEA 18.2% Div + 80.3% WC = 98.6%
GTY 65.6% Div + 34.2% WC = 99.8%
EDG 0.0% Div + 0.0% WC = 0.0%

As much as I would like to think I am brilliant manager (winning games against TRI
even though I used my RH lineup against SEA Elias) even I don't think I have a 63.4%
chance of winning my division while only leading by a game with 18 to go. That and I
am -8 on our pythag projections.

So now I also can generate results based on flipping a coin:

10000 simulated remaining seasons using coin flips for outcomes


AC East
COU 99.7% Div + 0.2% WC = 100.0%
NJR 0.0% Div + 0.0% WC = 0.0%
TRI 0.3% Div + 52.7% WC = 52.9%
KAT 0.0% Div + 0.0% WC = 0.0%

AC Central
PHI 99.1% Div + 0.9% WC = 100.0%
PAD 0.0% Div + 0.0% WC = 0.0%
WMS 0.9% Div + 99.1% WC = 100.0%
BOW 0.0% Div + 0.0% WC = 0.0%

AC West
MCM 0.7% Div + 4.1% WC = 4.8%
POR 24.9% Div + 14.9% WC = 39.8%
BAL 32.5% Div + 13.4% WC = 46.0%
MAD 41.9% Div + 14.6% WC = 56.5%

NC East
MNM 98.8% Div + 0.0% WC = 98.8%
SKY 1.2% Div + 0.0% WC = 1.2%
CSG 0.0% Div + 0.0% WC = 0.0%
SDQ 0.0% Div + 0.0% WC = 0.0%

NC Central
CAN 12.2% Div + 7.3% WC = 19.5%
SFP 87.8% Div + 1.9% WC = 89.7%
HAV 0.0% Div + 0.0% WC = 0.0%
NYK 0.0% Div + 0.0% WC = 0.0%

NC West
ODM 38.7% Div + 59.6% WC = 98.2%
SEA 8.4% Div + 84.5% WC = 92.9%
GTY 52.9% Div + 46.7% WC = 99.6%
EDG 0.0% Div + 0.0% WC = 0.0%

Finally, I can generate results using the log5 method as explained here
http://tangotiger.net/wiki_archive/Log5.html

10000 simulated remaining seasons using log5 matchups


AC East
COU 99.6% Div + 0.4% WC = 100.0%
NJR 0.0% Div + 0.0% WC = 0.0%
TRI 0.4% Div + 42.7% WC = 43.1%
KAT 0.0% Div + 0.0% WC = 0.0%

AC Central
PHI 98.4% Div + 1.6% WC = 100.0%
PAD 0.0% Div + 0.0% WC = 0.0%
WMS 1.6% Div + 98.4% WC = 100.0%
BOW 0.0% Div + 0.0% WC = 0.0%

AC West
MCM 1.1% Div + 5.7% WC = 6.7%
POR 28.9% Div + 17.9% WC = 46.9%
BAL 30.4% Div + 16.0% WC = 46.4%
MAD 39.6% Div + 17.4% WC = 57.0%

NC East
MNM 98.8% Div + 0.0% WC = 98.8%
SKY 1.2% Div + 0.0% WC = 1.2%
CSG 0.0% Div + 0.0% WC = 0.0%
SDQ 0.0% Div + 0.0% WC = 0.0%

NC Central
CAN 14.8% Div + 7.2% WC = 22.0%
SFP 85.2% Div + 2.1% WC = 87.2%
HAV 0.0% Div + 0.0% WC = 0.0%
NYK 0.0% Div + 0.0% WC = 0.0%

NC West
ODM 55.3% Div + 44.0% WC = 99.4%
SEA 7.2% Div + 84.8% WC = 92.0%
GTY 37.4% Div + 62.0% WC = 99.4%
EDG 0.0% Div + 0.0% WC = 0.0%

Going forward I will be using the log5 method for the predictions -- it seems to take
into account the "talent" level of teams better than the other methods.

There are other ways to do this like a "Monte Carlo" method involving actual runs
scored by teams but I only have so much time to fiddle around with this stuff.

--
Chris