World Cup 2018 Prediction Contest [R]

I used a website with simulations to get a probability of each country being eliminated in each round and winning the World Cup. Then I took the reciprocal of each probability to make points where predicting less likely outcomes are worth more points. For Morocco, the team most likely to be eliminated in the Group Stage (according to simulations, not me), you get 1.18 for correctly predicting that they were eliminated in the Group Stage. For Tunisia, the least likely to win the World Cup, you get 2,777.78 points for correctly predicting them winning the World Cup. For Brazil, the team who is most likely to reach every round and win the World Cup, you get 12.61 points for correctly predicting them to be eliminated in the Group Stage and 4.12 points for correctly predicting them to win the World Cup. Each person will start the World Cup with a different potential score depending on how likely your picks were. I attached a screen shot of a spreadsheet saying how many points every correct prediction would be worth. I hope people will participate.

@Paul Calixte

Can you sticky this topic?

 

Here's Brazil in table form. At some point I'll type every team that way. Note that R16 means being eliminated in the Round of 16, not reaching the Round of 16 or later, and the same goes for every round.

Team	Group Stage	R16	QF	SF	SECOND	FIRST
Brazil	12.61	5.44	6.94	4.20	8.94	4.12

 

Here's the table. It lists how many points you get for predicting the team would be eliminated in that round.

Team	Group Stage	R16	QF	SF	SECOND	FIRST
Brazil	12.61	5.44	6.94	4.20	8.94	4.12
Germany	11.93	3.15	8.96	4.77	7.69	6.78
Belgium	5.75	3.69	2.58	10.00	27.47	32.20
England	5.41	4.04	2.56	9.57	26.46	29.49
Argentina	5.36	2.53	4.50	12.23	15.74	19.79
France	4.64	4.55	3.57	7.96	14.20	11.29
Uruguay	4.38	2.53	5.73	9.08	19.61	24.11
Portugal	3.91	3.96	6.19	7.34	9.69	11.05
Spain	3.62	4.34	6.71	7.34	9.72	9.48
Russia	3.25	2.40	6.83	12.94	30.75	51.60
Colombia	2.93	2.81	4.80	17.32	45.87	70.22
Poland	2.49	2.66	5.98	26.62	82.30	160.77
Peru	2.04	4.80	5.43	16.42	30.07	42.61
Croatia	1.99	3.01	9.35	31.08	59.99	94.70
Denmark	1.95	4.65	5.78	18.22	36.89	54.29
Switzerland	1.88	3.76	12.44	12.87	35.75	62.34
Mexico	1.75	3.14	20.97	24.08	72.99	124.07
Japan	1.70	3.41	10.40	62.66	207.04	602.41
Iceland	1.65	3.60	12.67	46.02	101.83	176.88
Sweden	1.64	3.34	24.00	29.01	96.99	178.57
Iran	1.62	4.90	11.53	19.73	39.64	70.77
Serbia	1.57	4.35	15.81	20.89	74.29	144.93
Senegal	1.50	4.27	12.80	71.17	263.85	671.14
Egypt	1.47	4.29	17.87	45.27	140.65	378.79
Nigeria	1.42	4.37	20.40	93.55	247.52	529.10
Korea Republic	1.36	4.64	40.57	60.39	226.24	584.80
Costa Rica	1.33	5.91	24.66	34.89	128.21	280.90
Australia	1.28	8.12	14.66	61.80	152.91	341.30
Saudi Arabia	1.27	5.94	30.00	100.60	342.47	1351.35
Panama	1.24	8.50	15.30	113.38	561.80	1886.79
Tunisia	1.20	9.50	19.23	163.40	990.10	2777.78
Morocco	1.18	9.97	33.24	71.58	210.08	500.00

I'm not going to require everybody's predictions to be possible. If you want to predict 20 teams to get eliminated in the Round of 16, you can, but you would be guaranteed to be wrong for 4 teams. It is not necessary to state what group each team is in and/or what place you predict them to finish in.

I don't know what the best format is for people to make predictions. I will accept all of these:

1. Each team and the points so I can tell what round that came from.
2. Each team with GS or Group Stage, Round of 16 or R16, etc.
3. Each team with 1 for the Group Stage, 2 for Round of 16, to 6 for the winner.
4. Each round and listing the teams you predict to be eliminated then.

All predictions must be made at least one hour before the first game of the World Cup.

 

I'll give an example of how the simulations become points with 100 simulations:

Group Stage: 35 (100/35 = 2.86 points)
Round of 16: 30 (100/30 = 3.33 points)
Quarterfinals: 20 (100/20 = 5.00 points)
Semifinals: 10 (100/10 = 10.00 points)
Lose Final: 4 (100/4 = 25.00 points)
Win Final: 1 (100/1 = 100.00 points)

I rounded everything to two decimal places. I'm not saying the simulations are perfect, but I don't know how many sites have something that can be used to give a probability for every team being eliminated in every round. I don't want to do a prediction contest that only includes the Group Stage. If you would like to make a separate contest, you can.

Using these value provides a reason to predict an outcome you don't think will be happen because it will be worth a lot of points if you predict it to happen and you're right. This is true in the real world. If the goal was to predict the outcome of each game, people would predict Barcelona and Real Madrid to win almost every La Liga game. When correct bets on less likely outcomes are rewarded more than correct bets on more likely outcomes, there is a reason to bet on worse teams.

 

For Point 1, the way I intended it was that you only get points for a team if you are exactly right. If I want to give points for each round a person is correct about a team, I might totally redo the points table. Luckily, I have plenty of time.

 

I thought about having people make predictions for the two teams that will advance from each out of six possible combinations. I have simulations that provide probabilities for each team. The problem is that probability of a pair of teams advancing is not the product of the two teams' probabilities of advancing because the teams are not independent events. For example, knowing that Mexico advanced makes it less likely that other teams in their group advanced. Now I'm thinking about having people rank the teams in each group with a point system that does not depend on simulations. I'll give 2 points for each team in the correct position and 1 point for being 1 position off and being correct about if the team advanced.

 

My original plan used probabilities to make ratios. Now I'm thinking about awarding points for correct predictions for each percent chance of that team not doing what they did. Here's Group A:

Uruguay: 77.184% chance at advancing and 22.816% chance at being eliminated
Russia: 69.201% chance at advancing and 30.799% chance at being eliminated
Egypt: 32.094% chance at advancing and 67.906% chance at being eliminated
Saudi Arabia: 21.521% chance at advancing and 78.479% chance at being eliminated

If Uruguay and Russia advance and that is what you predicted, you would get 22.816 + 30.799 + 32.094 + 21.521 = 107.230. In the unlikely event that Egypt and Saudi Arabia advance and that is what you predicted, you would get 77.184 + 69.201 + 67.906 + 78.479 = 292.770. The sum of the points for correctly predicting two opposite pairs of teams to advance will always be 400. If Uruguay and Russia advance and you predicted Uruguay and Egypt to advance, you would get 22.816 + 0 + 0 + 21.521 = 44.337. Group G has the highest combined probability of advancing for its top two teams, so it will have values farthest apart. Correctly predicting Belgium and England would be worth 71.766, and correcting predicting Panama and Tunisia would be worth 328.234.

I could use ELO rating differences to get probabilities for each Knockout Round game and have people predict one game at a time rather than outcomes for every team before the World Cup. Each team will have enough days off between games for fans to watch the game a day or two later and make a prediction here before the next game. I would have to decide how to weight the Group Stage and Knockout Round component.

 

I'm going to do what I proposed in my previous post about your score being the probability of your prediction being wrong. If you are wrong about if a team advances, you get 0 for that team. If you are correct about if a team advances and are one spot off on the rankings (first and second switched or third and fourth switched), your score is the probability of your prediction being wrong. If you are correct about a team's position, you get double what you would get from being one spot off. If Person X has the teams in the correct order and Person Y has first and second switched and third and fourth switched, Person X will get double Person Y for that group.

 

I'm thinking about e-mailing people an Excel file and instructions to enter their predictions in it and e-mail it back to me. That way I don't have to type tables here.

 

Rather than working with probabilities and values that aren't whole numbers, I made a simple format that only includes the Group Stage. Predict the position and points for each team. You do not have to give predictions for each game. You do not have to give scores, but if you predict two or more teams to be tied on points you must specify the tiebreaker order. Goal differential is the first tiebreaker. I will evaluate positions and points off:

Positions: 40%
For each team, you will get 2 points if you were correct about that team's position and 1 point if you were one spot off and correct about whether the team would advance or not. To convert to scores to something out of 40, everybody's score will be multiplied by X (which can be less than 1) so the best person gets a 40 and everybody else gets proportionately less. If you have 60% as many points as the best person, you get 60% of 40, which is 24.

Points Off: 60%
I will do each group separately, and the groups are 7.5% each. I will add up the sum of how many points you were off in each team in the standings. Being X points too high and X points too low are the same. To convert to scores to something out of 7.5 for each group, everybody's score will be multiplied by X (which can be less than 1) so the best person gets a 7.5 and everybody else gets proportionately less. This score is something you want to be low. The best person will get 7.5 and the worst person will get 0, with everybody else's score distributed proportionately in between.

Here are the teams in each group:

A: Russia, Uruguay, Egypt, and Saudi Arabia
B: Portugal, Spain, IR Iran, and Morocco
C: France, Peru, Denmark, and Australia
D: Argentina, Croatia, Iceland, and Nigeria
E: Brazil, Switzerland, Serbia, and Costa Rica
F: Germany, Mexico, Sweden, and Korea Republic
G: Belgium, England, Panama, and Tunisia
H: Colombia, Poland, Senegal, and Japan