2016 NFL Game Predictions: Week 3

Welcome!  I intend this to be an ongoing project of predicting NFL game outcomes, point spreads, and final scores.  My hope is that my models will improve over time as they become more sophisticated and use better data.  I will try to regularly publish predictions to keep myself accountable, and will make observations along the way about what is working and what isn't.  See below for latest updates.  Enjoy!


Previous 2016 Predictions:

• Week 2: 11 - 5

------------------------------------------------------------------------------------------------------------------------

Recap: Week 3

9-7.  So the running total is 20 - 12 (62.5%).  With the probabilities I gave, the expected outcome was 9.44 correct predictions.  So this is as expected given my model performance and the probability predictions it output.

Did the ensemble do any better?  Yes.  If I go with the ensemble prediction with ties decided by outcome prediction, the prediction was 11-5.  If I let ties be decided by the final score, the prediction was still 11-5.  So I definitely should look into ensembling in a more robust and consistent way in the future.

_________________________________________________________________________________

Week 3:

My goal is to make some improvement in my models, however minor, once a week.  This week I made two changes.

The first was to normalize the distribution of final scores before predicting them.  The point of doing this is to reduce heteroscedasticity so that the regression errors of my model are more uniform.  In particular, the games in which a team scored 0 points cause the data to be skewed left.  I applied a power transform (.75) to bring the values above 0 closer to 0 so that the overall distribution is more normal.  This caused my final score model to perform better (r-squared = .146 to r-squared = .156).  It's still a terrible model, but now it is less terrible.

The second change was to start making my outcome, score difference, and final score models vote on the outcome.  The models often disagree in the final outcome, so this is a way of ensembling the models to produce a single outcome.  Right now, I am mostly curious about how this ensembling model performs compared to the outcome model.  So this is just exploration.  In the future, I hope to make a more serious attempt at ensembling models and using this ensemble to predict outcomes.  For now, my outcome predictor is my final prediction of the game.

Oddities and Bold Predictions:

• Broncos vs. Bengals:
• my outcome model gives the Bengals the win (barely at .504), the score difference predicts 0, and the final score model predicts 27-23 Broncos victory.  So this is essentially a tie.  Personally, I'd go with the Broncos against my outcome prediction.
• Chiefs vs. Jets:
• same as above.  Chiefs get the outcome win, score difference is 0, and Jets get the final score victory 27-25.  So another tie.  Still going with a Chief's victory.
• Giants vs. Redskins:
• my outcome says Giants, but the score difference and final score models both say Redskins.  I'd go with the Giants.
• Raiders vs. Titans:
• outcome says Titans, but score difference and score final say Raiders.  Both are 1-1 and I don't know enough about either team to make a judgement call.

Good luck in Week 3!

Here are the predictions for Week 3

Week	Date	Team	Home Away	Opponent	Prob. Win	Pred. Team Win	Actual Team Win	Pred. Team PD	Actual Team PD	Pred. Team Score	Pred. Opp. Score	Actual Team Score	Actual Opp. Score	Ensemble Prediction Win
3	9/22/2016	Patriots		Texans	0.523	1	1	3	27	25	20	27	0	1.00
3	9/22/2016	Texans	@	Patriots	0.477	0	0	-3	-27	20	25	0	27	0.00
3	9/25/2016	49ers	@	Seahawks	0.468	0	0	-2	-19	18	22	18	37	0.00
3	9/25/2016	Bears	@	Cowboys	0.422	0	0	-3	-14	21	25	17	31	0.00
3	9/25/2016	Bengals		Broncos	0.504	1	0	0	-12	23	27	17	29	0.50
3	9/25/2016	Bills		Cardinals	0.295	0	1	-5	15	21	26	33	18	0.00
3	9/25/2016	Broncos	@	Bengals	0.496	0	1	0	12	27	23	29	17	0.50
3	9/25/2016	Browns	@	Dolphins	0.388	0	0	-3	-6	20	23	24	30	0.00
3	9/25/2016	Buccaneers		Rams	0.745	1	0	7	-5	24	14	32	37	1.00
3	9/25/2016	Cardinals	@	Bills	0.705	1	0	5	-15	26	21	18	33	1.00
3	9/25/2016	Chargers	@	Colts	0.499	0	0	-1	-4	25	26	22	26	0.00
3	9/25/2016	Chiefs		Jets	0.541	1	1	0	21	25	27	24	3	0.50
3	9/25/2016	Colts		Chargers	0.501	1	1	1	4	26	25	26	22	1.00
3	9/25/2016	Cowboys		Bears	0.578	1	1	3	14	25	21	31	17	1.00
3	9/25/2016	Dolphins		Browns	0.612	1	1	3	6	23	20	30	24	1.00
3	9/25/2016	Eagles		Steelers	0.382	0	1	-5	31	19	24	34	3	0.00
3	9/25/2016	Giants		Redskins	0.509	1	0	-1	-2	24	27	27	29	0.33
3	9/25/2016	Jaguars		Ravens	0.277	0	0	-6	-2	21	25	17	19	0.00
3	9/25/2016	Jets	@	Chiefs	0.459	0	0	0	-21	27	25	3	24	0.50
3	9/25/2016	Lions	@	Packers	0.377	0	0	-3	-7	25	26	27	34	0.00
3	9/25/2016	Packers		Lions	0.623	1	1	3	7	26	25	34	27	1.00
3	9/25/2016	Panthers		Vikings	0.595	1	0	1	-12	23	21	10	22	1.00
3	9/25/2016	Raiders	@	Titans	0.49	0	1	2	7	30	27	17	10	0.67
3	9/25/2016	Rams	@	Buccaneers	0.255	0	1	-7	5	14	24	37	32	0.00
3	9/25/2016	Ravens	@	Jaguars	0.723	1	1	6	2	25	21	19	17	1.00
3	9/25/2016	Redskins	@	Giants	0.491	0	1	1	2	27	24	29	27	0.67
3	9/25/2016	Seahawks		49ers	0.532	1	1	2	19	22	18	37	18	1.00
3	9/25/2016	Steelers	@	Eagles	0.618	1	0	5	-31	24	19	3	34	1.00
3	9/25/2016	Titans		Raiders	0.51	1	0	-2	-7	27	30	10	17	0.33
3	9/25/2016	Vikings	@	Panthers	0.405	0	1	-1	12	21	23	22	10	0.00
3	9/26/2016	Falcons	@	Saints	0.628	1	1	5	13	29	27	45	32	1.00
3	9/26/2016	Saints		Falcons	0.372	0	0	-5	-13	27	29	32	45	0.00