04-27-2018, 05:12 PM
(This post was last modified: 05-05-2018, 11:31 PM by Supersquare04.)
Code:
1582 words
There are many strategies for completing NSFL predictions. Some get the highlight from their GM and copy the GM's predictions, some find someone they consider knowledgeable and copy paste - hi timeconsumer, some roll with the home teams except in obvious situations, some go with who they want to win, and still others just pick more or less at random. Which of these strategies are best?
I don't know.
But what I do know is that the predictions and results of the 56 games of Season 6 indicate that it's usually but not always smart to go with the crowd.
I went through the archived predictions topics for each sim last season and recorded how many predictions there were for each team (this is how I discovered the treachery of Payton by the way) and compared that data to the actual result of the game.
There were 6 games where every single person predicted the same team:
Week 2 - San Jose (Home) over New Orleans
Week 8 - San Jose (Away) over Colorado
Week 9 - Orange County (Away) over Colorado
Week 12 - Orange County (Home) over Arizona
Week 12 - Baltimore (Home) over Colorado
Week 13 - Yellowknife (Home) over Colorado
In all 6 of those games, the unanimous choice prevailed.
There were 25 games where more than 90% of people predicted the same team but there was at least 1 contrarian. I won't list them all but the prediction favorite won "just" 21 of those games. If the predictions were, well, predictive you would expect that bucket of games to have around a win percentage around 95%. If we break down this bucket further we find that teams with 95%+ of predictions went 12-3 (80%, counting Yellowknife's tie at New Orleans as a loss for both teams since neither prediction got TPE) and those with 90-95% went 9-1 (90%). With valuable TPE on the line, picking your spots among these near-unanimous games could give you some valuable differentials to get extra TPE compared to your opponents.
Digging a bit deeper here at the risk of splintering an already small sample size further, these are the teams who won with <10% support:
Week 4 - Arizona (Away) over Baltimore
Week 7 - San Jose (Away) over Yellowknife
Week 7 - Baltimore (Away) over Orange County
All three of these games are very similar: two well-regarded teams play each other and people bandwagon the home team figuring that's a reasonable tie-breaker. In games between the top six teams (so excluding New Orleans and Colorado), the home team went 19-11 (63.33%) and were picked, on average, 72.50% of the time. This suggests that the playerbase likely puts too much emphasis on home teams when predicting winners. Indeed, the home team went just 31-24-1 (56.25%) overall last season despite being picked 62.57% of the time.
The results of the other deciles are as follows:
80%-89.999%: 6-2
70%-79.999%: 2-3
60%-69.999%: 4-2
50%-59.999%: 3-4 (only one game was exactly 50% - Week 11 Orange County at Philadelphia - so it's 2-3 if you discount that one)
If instead we break these down by quintiles we get the following:
80%-100%: 33-6
60%-79.999%: 6-5
40%-59.999%: 6-6
20%-49.999%: 5-6
0%-19.999%: 5-34
And by quartiles:
75%-100%: 35-7
50%-74.999%: 7-8
25%-49.999%: 7-6
0%-24.999%: 6-36
As you can see from these various breakdowns, our bandwagons often get carried away. If you spent the time and effort (which you absolutely should not, trust me it's not worth it) to calculate each team's pick rate and submit yours last with the following heuristics:
1. If one team has at least 2/3 of the predictions, pick that team
2. In a game where both teams are between 1/3 and 2/3 in terms of predictions, pick the away team
You would have finished last season with a predictions record of 43-13, quite a bit better than the 31-25 record you'd have if you simply picked the home team but not much better than the 41-14 (plus however you dealt with the tie) record you'd have if you simply picked the favorites.
So now that I've written 600+ words talking about the crazy, mind-blowing revelation that the team people think will win usually wins (or at least did last season) here are some fun facts about people's predictions:
1. Predictions participation drops fairly significantly over time. Week 1 had 88 predictions while Week 13 and Week 14 had 56 and 65, respectively. The first half of the season had no fewer than 75 predictions and the second half only eclipsed 75 once.
2. Orange County (to everyone's surprise, I'm sure) had the most predictions - 84.20% of them over the course of the season.
3. Yellowknife and Philadelphia were in 2nd and 3rd with 66.86% and 64.49% respectively.
4. San Jose and Baltimore made up the third tier with 59.75% and 57.81% respectively.
5. Arizona was a distant 6th with 43.33%.
6. New Orleans barely beat out Colorado for not-last with 12.04% to Colorado's 11.39%.
7. 62.57% of predictions were for the home team.
8. The "chalkiest" week was Week 12 where we saw Philadelphia (88%) beat Yellowknife, Orange County (100%) beat Arizona, Baltimore (100%) beat Colorado, and San Jose (98.67%) beat New Orleans. Out of the 300 total predictions (75 people x 4 games), only 10 were wrong.
9. Week 7 was the most chaotic week with San Jose (3.57%) beating Yellowknife, Arizona (29.76%) beating Philadelphia, Baltimore (3.57%) beating Orange County, and New Orleans (53.01%) beating Colorado. Yeah, a week where the only favorite to actually win is New Orleans is pretty crazy.
10. The lowest confidence in Orange County was Week 13 when they played San Jose on the road and only got 41.07% of predictions. That was the only time the team was below 50%.
11. Colorado was the only team to get zero predictions for a home game and it happened twice: Weeks 8 and 9 when they played San Jose and Orange County back to back.
12. San Jose's Week 7 road victory over Yellowknife and Baltimore's Week 7 road victory over Orange County were the biggest upsets last season - just 3 of 84 people predicted each. ralams123, nunccoepi, and ckroyal92 predicted San Jose and DestaDanger, Unicorn, and manicmav36 predicted Baltimore. No one, obviously, predicted both.
13. Depending on how you view it, Yellowknife's Week 4 tie at New Orleans was another major upset. 78 of 79 people (98.73%) picked Yellowknife in that game (yours truly having the only dissent).
"Hey Beaver, what would the standings have looked like if the team with more predictions won every game?" Ah yes, I'm glad you're totally wondering about that because I'm about to tell you.
In the NSFC we would have Philadelphia win the division with a 9-4-1 record and Yellowknife would've snuck into the playoffs ahead of Baltimore with a 9-5 record to their 8-6. Colorado would've brought up the rear at 1-13.
In the ASFC we would've had Orange County leading the way at 12-1-1 with San Jose getting the other playoff spot at 9-5 and Arizona at 6-8 and New Orleans at 1-13 golfing early.
Yes, that's right. The standings positions would've stayed exactly the same. The only difference is that in this hypothetical scenario Arizona would've flipped with Baltimore in the draft.
If instead of having the win/loss cut-off at 50% and instead break it down into thirds like we did for the heuristics earlier (above 2/3 is a win, below 1/3 is a loss, anything in between is a toss up) we find the following:
Philadelphia 9-4 with 1 toss up game (home vs Orange County)
Yellowknife 8-2 with 4 toss up games (away vs Baltimore, away vs Arizona, away vs San Jose, and home vs Orange County)
Baltimore 7-5 with 2 toss up games (home vs Yellowknife and away vs Arizona)
Orange County 11-0 with 3 toss up games (away vs Philadelphia, away vs San Jose, and away vs Yellowknife)
San Jose 7-5 with 2 toss up games (home vs Yellowknife and home vs Orange County)
Arizona 5-7 with 2 toss up games (home vs Yellowknife and home vs Baltimore)
Meanwhile the only time Colorado and New Orleans got more than 1/3 of the predictions was when they played each other.
So the playoff races would have been very much in doubt. Yellowknife went 0-4 in reality in their 4 toss up games so it's very likely that it would've also come down to the Week 14 Yellowknife at Baltimore game just like in reality. Philadelphia isn't necessarily safe either but the most safe of the 3. Orange County went 2-1 in their 3 toss ups in reality but even if they lost all 3 they could've have been caught for home field. Arizona went 2-0 in their toss ups and San Jose split theirs so that could have been a tight race.
Yeah so anyway this has been a very productive use of a Friday at work. If you're scrolling down to try to find the tl;dr it's: when aggregated, predictions are usually pretty accurate.
Not really sure how else I can milk this for word count. Well, if we go back to the thing about the record between the 6 non-bum teams and excise Arizona from that since they were pretty far behind the top 5 that 19-11 record for home teams becomes 12-8 which is a further strike against the reversion to the default "pick the home team."
Ok, the end. Have a wall of numbers and letters: