NFL Week 3 | Coin vs. Machine
SEP 21 2017
Week 1 - Week 2 - Week 3 - Week 4 - Week 5 - Week 6
Week 7 - Week 8 - Week 9 - Week 10 - Week 11 - Week 12
Week 13 - Week 14 - Week 15 - Week 16 - Week 17
Wild Card - Divisional - Conference - Super Bowl
SEP 21 2017
Week 1 - Week 2 - Week 3 - Week 4 - Week 5 - Week 6
Week 7 - Week 8 - Week 9 - Week 10 - Week 11 - Week 12
Week 13 - Week 14 - Week 15 - Week 16 - Week 17
Wild Card - Divisional - Conference - Super Bowl
Welcome to Week 3 of a thus-far-very uninspiring 2017 NFL season. Maybe we’ll forget about it in a few weeks, but to this point we’ve seen the lowest average point totals through two weeks (40.29 points) since 2010 and the highest average margin of victory through two weeks (12.94 points) since 2005.
Fourteen of the NFL’s 31 games have been decided by 14 points or more, and the league is on pace for the fewest touchdowns per game since 1993. Saturday’s Southern Cal versus Texas game drew more people than Sunday’s L.A. Rams and L.A. Chargers games combined, tickets to the Chargers home opener were reportedly going for $6, and the cheapest tickets to tonight’s 49ers-Rams game in Santa Clara would cost you less than two Levi’s Stadium pretzels. (The 49ers’ official attendance numbers are cute.)
Fourteen of the NFL’s 31 games have been decided by 14 points or more, and the league is on pace for the fewest touchdowns per game since 1993. Saturday’s Southern Cal versus Texas game drew more people than Sunday’s L.A. Rams and L.A. Chargers games combined, tickets to the Chargers home opener were reportedly going for $6, and the cheapest tickets to tonight’s 49ers-Rams game in Santa Clara would cost you less than two Levi’s Stadium pretzels. (The 49ers’ official attendance numbers are cute.)
Before we get into Week 3 picks, I’d like to highlight a study published by the Harvard Sports Analysis Collective (HSAC) last year on Covers.com. HSAC observed that from 1985 to 2015, NFL teams that only played home games during the first two weeks of the season were 61-40 (60.4%) against the spread in Week 3, while teams that only played road games in the first two weeks were 37-69 (34.9%).
Happenstance or not, it’s a cool observation to have in mind headed into Week 3. The teams that fit that mold this season are the Bengals (0-2 at home), Broncos (2-0 at home), Rams (1-1 at home), and Jets (0-2 on the road). If you buy into HSAC’s theory, then you might be inclined in Week 3 toward the Bengals (+9) at Green Bay, Broncos (-3) at Buffalo, Rams (-3) at San Francisco, and Dolphins (-6) at the N.Y. Jets.
Happenstance or not, it’s a cool observation to have in mind headed into Week 3. The teams that fit that mold this season are the Bengals (0-2 at home), Broncos (2-0 at home), Rams (1-1 at home), and Jets (0-2 on the road). If you buy into HSAC’s theory, then you might be inclined in Week 3 toward the Bengals (+9) at Green Bay, Broncos (-3) at Buffalo, Rams (-3) at San Francisco, and Dolphins (-6) at the N.Y. Jets.
|
Tiers Methodology
In previous weeks, I’ve made reference to various “tiers” for each week’s model- or machine-based picks, including “best bets” and “toss-ups” and everything in between. I’ve since refined that approach to include five tiers – Tier I picks are those picks that the models agree on most, while Tier V picks are those that the models agree least. You can see these tiers at the bottom of each week’s picks table.
This column focuses on FiveThirtyEight’s “Elo,” Jeff Sagarin’s “Rating,” and Romeo Peričić’s “RP-Excel I.” But to produce these more nuanced tiers, I am also overlaying Jeff Sagarin’s “Offense-Defense Method Predictions” and David Wilson’s “Nutshell Sports” ratings. Like RP-Excel, Nutshell is lesser known but has been successful against the spread, ranking second (52.9%) to only RP-Excel (54.3%) among the 24 models that have picked every game since 2012 (once again putting aside the Computer Adjusted Line).
Each week’s Tier I picks – which I’ll also refer to as “best bets” – are the teams that all five prediction methods agree will cover the spread. They’re the unanimous picks. Tier II picks are the teams that four of the five models agree on. Tiers III, IV, and V all include teams that three of the five models agree on.[1]
To differentiate between Tiers III, IV, and V, I also look at the difference between the models’ point spreads and the actual betting line. FiveThirtyEight, for example, has Kansas City as an implied seven-point favorite at the L.A. Chargers this week. The Chiefs are only three-point favorites in Vegas, so there’s a four-point difference between the model and the line.
Sometimes, three models pick one team, but the other two models pick the other side by a bigger combined points margin, creating a "disagreement" between pick frequency and magnitude. When there is no such disagreement – that is, when a majority of models pick a team and the difference between the models’ point spreads and actual line favor the same team – those are Tier III picks. Picks with some degree of disagreement are Tier IV. Picks with total disagreement are Tier V “toss-ups.”[2]
In previous weeks, I’ve made reference to various “tiers” for each week’s model- or machine-based picks, including “best bets” and “toss-ups” and everything in between. I’ve since refined that approach to include five tiers – Tier I picks are those picks that the models agree on most, while Tier V picks are those that the models agree least. You can see these tiers at the bottom of each week’s picks table.
This column focuses on FiveThirtyEight’s “Elo,” Jeff Sagarin’s “Rating,” and Romeo Peričić’s “RP-Excel I.” But to produce these more nuanced tiers, I am also overlaying Jeff Sagarin’s “Offense-Defense Method Predictions” and David Wilson’s “Nutshell Sports” ratings. Like RP-Excel, Nutshell is lesser known but has been successful against the spread, ranking second (52.9%) to only RP-Excel (54.3%) among the 24 models that have picked every game since 2012 (once again putting aside the Computer Adjusted Line).
Each week’s Tier I picks – which I’ll also refer to as “best bets” – are the teams that all five prediction methods agree will cover the spread. They’re the unanimous picks. Tier II picks are the teams that four of the five models agree on. Tiers III, IV, and V all include teams that three of the five models agree on.[1]
To differentiate between Tiers III, IV, and V, I also look at the difference between the models’ point spreads and the actual betting line. FiveThirtyEight, for example, has Kansas City as an implied seven-point favorite at the L.A. Chargers this week. The Chiefs are only three-point favorites in Vegas, so there’s a four-point difference between the model and the line.
Sometimes, three models pick one team, but the other two models pick the other side by a bigger combined points margin, creating a "disagreement" between pick frequency and magnitude. When there is no such disagreement – that is, when a majority of models pick a team and the difference between the models’ point spreads and actual line favor the same team – those are Tier III picks. Picks with some degree of disagreement are Tier IV. Picks with total disagreement are Tier V “toss-ups.”[2]
|
Looking Back at Week 2
The models did very well last week, registering an aggregate record of 10 wins and six losses against the spread. FiveThirtyEight led the way at 12-4. Our silly little coin picked only five games correctly.
The three main models agreed on five picks in Week 2 and won them all (highlighted in dark blue in the table). We’ll keep an eye on those games over the next few weeks. Of those five picks, three were unanimously selected by all five models mentioned above – Washington (+2.5) at L.A. Rams, Baltimore (-8.0) versus Cleveland, and Tampa Bay (-7.0) versus Chicago. They, of course, were all wins as well.
Overall, the machines went 3-0 in Tier I “best bets,” 2-3 in Tier II, 2-0 in Tier III, 2-0 in Tier IV, and 1-3 in Tier V “toss-ups.” (In coming weeks, we’ll take stock of each tier’s record through the first few weeks.)
The models did very well last week, registering an aggregate record of 10 wins and six losses against the spread. FiveThirtyEight led the way at 12-4. Our silly little coin picked only five games correctly.
The three main models agreed on five picks in Week 2 and won them all (highlighted in dark blue in the table). We’ll keep an eye on those games over the next few weeks. Of those five picks, three were unanimously selected by all five models mentioned above – Washington (+2.5) at L.A. Rams, Baltimore (-8.0) versus Cleveland, and Tampa Bay (-7.0) versus Chicago. They, of course, were all wins as well.
Overall, the machines went 3-0 in Tier I “best bets,” 2-3 in Tier II, 2-0 in Tier III, 2-0 in Tier IV, and 1-3 in Tier V “toss-ups.” (In coming weeks, we’ll take stock of each tier’s record through the first few weeks.)
|
Looking Ahead to Week 3
Week 3 looks to be a little trickier, with only one unanimous selection (Tampa Bay) between the three models. There are no “best bets” this week, yet half of the Week 3 slate clocks in as Tier II. (Again, Tier II means that four of the aforementioned five prediction models pick the same team against the spread.) This week’s toss-ups are Houston (+13.5) at New England and Arizona (+3.0) at Dallas on Monday night.
Week 3 looks to be a little trickier, with only one unanimous selection (Tampa Bay) between the three models. There are no “best bets” this week, yet half of the Week 3 slate clocks in as Tier II. (Again, Tier II means that four of the aforementioned five prediction models pick the same team against the spread.) This week’s toss-ups are Houston (+13.5) at New England and Arizona (+3.0) at Dallas on Monday night.
Good luck and enjoy Week 3 – which, if Thursday night’s unexpectedly exciting 49ers-Rams game is any indication, might make for a fun weekend of games.
|
Footnotes
[1] When four models agree on a pick and one has "no opinion" – because its line matches the actual betting line – I am bucketing it in Tier I with the "unanimous" picks. When three models agree on a pick and one has "no opinion," I am bucketing it in Tier II.
[2] When four of the models are split on a pick and the fifth has "no opinion," I am bucketing it in Tier V as an outright "toss-up."
[1] When four models agree on a pick and one has "no opinion" – because its line matches the actual betting line – I am bucketing it in Tier I with the "unanimous" picks. When three models agree on a pick and one has "no opinion," I am bucketing it in Tier II.
[2] When four of the models are split on a pick and the fifth has "no opinion," I am bucketing it in Tier V as an outright "toss-up."
The models referenced are created by and sourced from FiveThirtyEight, Jeff Sagarin via USA Today, RP-Excel.com, and NutshellSports.com. Lines are sourced from FootballLocks.com as of Sep 21 at 4pm PT. Data was compiled and analyzed by ELDORADO. All charts and graphics herein were created by ELDORADO.
ELDORADO | Berkeley, CA | New York, NY
eldo.co | @eldo_co
ELDORADO | Berkeley, CA | New York, NY
eldo.co | @eldo_co