Can You Outpredict Paul the Octopus in the 2014 PF World Cup League?

[micromass]

Ahh yea, that is right. If outcomes were picked at random, and I predicted no draws for any match, and for a big enough sample, is the maximum I can get about 66% or 50% of the predictions ?

My apologies for making this into a maths thread

Anyway, let's answer your question.

A game match is denoted by a variable $X$. So $X$ can have three outcomes: A wins, B wins and draw (denoted by A,B,D). Based on the data of the group matches I propose the following probability model. Team A wins with probability 2/5, Team B wins with probability 2/5 and there is a draw with probability 1/5. So $X$ takes on $A$ with probability $2/5$, it takes on $B$ with probability $2/5$ and it takes on $D$with probability $1/5$.

Your prediction are denoted by a variable $Y$ which has the same distributioin as $X$. So it is again random.

The question is to find the probability $\mathbb{P}\{X = Y\}$. Since $X$ and $Y$ will be assumed independent, we can calculate it as
$$\mathbb{P}\{X = Y\} = \sum_{k\in \{A,B,D\}} \mathbb{P}\{X = k\}\mathbb{P}\{Y = k\} = \frac{4}{25}+ \frac{4}{25} + \frac{1}{25} = \frac{9}{25}$$

So if you would pick the teams totally randomly, you would expect to get a percentage of 36%.

Another strategy is never to choose draws, but always to choose randomly a winning and losing team (or equivalently: always choose that $A$ wins). Then we get

$$\mathbb{P}\{X = Y\} = \frac{2}{5}\frac{1}{2} + \frac{2}{5}\frac{1}{2} + \frac{1}{5} 0 = \frac{2}{5}$$

In this case you would get a percentage of 40%. So we should expect Borek to get a percentage of 40% eventually

The worst strategy is probably to consistently predict a draw. We get

$$\mathbb{P}\{X = Y\} = \frac{2}{5} 0 + \frac{2}{5} 0 + \frac{1}{5} 1 = \frac{1}{5}$$

This will get you 20% correct outcomes.

 

[DennisN]

Anyway, let's answer your question.

I see micromass is performing some math-fu. :thumbs:

My votes for 30th June and 1st July:

June 30th:
France - Nigeria: 2 - 1
Germany - Algeria: 3 - 0

July 1st:
Argentina - Switzerland: 2 - 1
Belgium - USA: 1 - 1 (USA advances) - (guess I don't have to fear any "Boos!" from Americans)

 

[micromass]

Belgium - USA: 1 - 1 (USA advances) - (guess I don't have to fear any "Boos!" from Americans)

Boooo!

 

[DennisN]

Boooo!

Sorry! It can go either way!
EDIT: I'm from Sweden. We didn't even qualify!

 

[D H]

June 30th:

France - Nigeria: 2 - 0
Germany - Algeria: 2 - 0

 

[Intrastellar]

Anyway, let's answer your question.

A game match is denoted by a variable $X$. So $X$ can have three outcomes: A wins, B wins and draw (denoted by A,B,D). Based on the data of the group matches I propose the following probability model. Team A wins with probability 2/5, Team B wins with probability 2/5 and there is a draw with probability 1/5. So $X$ takes on $A$ with probability $2/5$, it takes on $B$ with probability $2/5$ and it takes on $D$ with probability $1/5$.

Your prediction are denoted by a variable $Y$ which has the same distributioin as $X$. So it is again random.

The question is to find the probability $\mathbb{P}\{X = Y\}$. Since $X$ and $Y$ will be assumed independent, we can calculate it as
$$\mathbb{P}\{X = Y\} = \sum_{k\in \{A,B,D\}} \mathbb{P}\{X = k\}\mathbb{P}\{Y = k\} = \frac{4}{25}+ \frac{4}{25} + \frac{1}{25} = \frac{9}{25}$$

So if you would pick the teams totally randomly, you would expect to get a percentage of 36%.

Another strategy is never to choose draws, but always to choose randomly a winning and losing team (or equivalently: always choose that $A$ wins). Then we get

$$\mathbb{P}\{X = Y\} = \frac{2}{5}\frac{1}{2} + \frac{2}{5}\frac{1}{2} + \frac{1}{5} 0 = \frac{2}{5}$$

In this case you would get a percentage of 40%. So we should expect Borek to get a percentage of 40% eventually

The worst strategy is probably to consistently predict a draw. We get

$$\mathbb{P}\{X = Y\} = \frac{2}{5} 0 + \frac{2}{5} 0 + \frac{1}{5} 1 = \frac{1}{5}$$

This will get you 20% correct outcomes.

Great
If we use the group stage results of the world cup:
There were $48$ matches in the group stages and I counted $8$ draws during them.
so modifying your model:

So $X$ takes on $A$ with probability $5/12$, it takes on $B$ with probability $5/12$ and it takes on $D$ with probability $1/6$.

$$\mathbb{P}\{X = Y\} = \sum_{k\in \{A,B,D\}} \mathbb{P}\{X = k\}\mathbb{P}\{Y = k\} = \frac{25}{144}+ \frac{25}{144} + \frac{1}{36} = \frac{3}{8}$$

$$\mathbb{P}\{X = Y\} = \frac{5}{12}\frac{1}{2} + \frac{5}{12}\frac{1}{2} + \frac{1}{6} 0 = \frac{5}{12}$$

$$\mathbb{P}\{X = Y\} = \frac{5}{12} 0 + \frac{5}{12} 0 + \frac{1}{6} 1 = \frac{1}{6}$$

Please correct me if I did anything wrong
(one question, isn't predicting the win of a team equivalent to predicting the loss of the other team ? but predicting the draw of one team is predicting the draw of the other ? does this modify the maths somehow ? should we multiply the probability of a draw by 2 before calculating the probability of a win/loss ?)
Picking the teams totally randomly will eventually give you 37.5%, picking only wins and losses will give you 41.7% and picking only draws will give you 16.7%.

So, if I want to pick the distribution that gives me the maximum number of points.

lets denote to our predictions as $a$ for a win, $b$ for a lose, and $d$ for a draw
$$a = b$$
$$a + b + d = 1$$
(please inform me if these are correct)
$$\mathbb{P}\{X = Y\} = aA + bB + dD$$
~some magic happens~
$$\mathbb{P}\{X = Y\} = a(A + B - 2D) + D$$
If $$(A + B - 2D) > 0$$
then picking the largest possible value for $a$ will give you the highest probability, and if $$(A + B - 2D) < 0$$ then picking the smallest possible value for $a$ will give you the highest probability. of course, $0 \le a \le 1/2$.

Using the above equations and the values I obtained for $A$, $B$ and $D$: the highest possible probability is for choosing no draws $\frac{5}{12}$ and the lowest was for choosing all draws $\frac{1}{6}$

(Looking at the equations above, it seems that the draw being multiplied by 2 is already in the equation, so we don't need to modify anything as I asked above ?)

I can see that the vast majority of us had more than 41.7% outcome accuracy, so even the random picks with only wins and losses will not give us as good a result as the ones we have gotten (reassuring ).
But how does the (presumably correct) maths above translate when we want to pick our outcomes ? will choosing no draws but wins/losses that are not completely random increase our outcome accuracy above our current average (or even maximum) ? can we simply modify the maths using, for example, the FIFA ranking advantage to choose every single outcome ? I guess we have to come up with model for calculating the probability of an outcome based on a function (we have to choose) of the FIFA ranking advantage and compare it to the 48 matches that were played.

Now micromass, will you rise to the bigger challenge ?


Edit: Corrected some typos

 

[trollcast]

June 30th:
France - Nigeria: 2 - 1
Germany - Algeria: 3 - 1

 

[micromass]

Thank you montadhar, your analysis seems perfect!

Choosing draw is clearly not a very good tactic. I think it is clear that the best tactic would be to choose either a winner or a loser each time. So let us ignore draws

Now, I have determined for the group stages which matches were predicted by the FIFA ranking and which were not. It turns out that the FIFA ranking predicted the correct outcome 31 times of the 40.

So if you follow the FIFA ranking each time, then you would get a outcome accuracy percentage of 64.5%. So it seems to me that an optimal strategy would be to follow the FIFA ranking every time.

Now, let's say that you predict a match and you say that Team A wins and you are n% sure. For example, you can say that you are 95% sure that Brazil will win against Greece.

You can now use the FIFA ranking to give you additional information. Let's say that the FIFA ranking puts Team A in a higher spot, then using Bayes' theorem, we can now be an amount of

$$\frac{31n}{9 + 22n}$$

sure of a win of Team A. So in our example, if we were 95% sure of our win and FIFA agrees, then we can now be

$$\frac{31\cdot 0.95}{9 + 22\cdot 0.95} = 0.98$$

certain. If you were only 50% sure of your win, then by taking FIFA rankings in account, you can now be

$$\frac{31\cdot 0.5}{9 + 22\cdot 0.5} = 0.77$$

certain of your choice!

On the other hand, if the FIFA ranking disagrees, then your updated certainty is

$$\frac{9n}{31 - 22n}$$

So if you were 95% sure of your choice, then by taking FIFA into account, you can now only be certain for 84%. If you were only 50% sure of your choice, then by taking FIFA into account, you can now only be 22% certain.

This information can be found in the following graph:

Also, my new predictions:
France - Nigeria: 1 - 0
Germany - Algeria: 2 - 0
Argentina - Switzerland: 3 - 0
Belgium - USA: 1 - 0

 

[DennisN]

Love the look of the leaderboard, now it's really exciting! I think we are in for some real fun! (well, it has already been very fun of course!)

 

[micromass]

Love the look of the leaderboard, now it's really exciting! I think we are in for some real fun! (well, it has already been very fun of course!)

Yes, very exciting! But I really regret my choice of letting the prediction of the winner count for so many points. Because basically you just need to predict the winner and you can win, regardless how you do in the individual matches.

That's why this competition will have two winners! One overall winner for which both the winner prediction and the scores count and one winner who predicted the scores most accurately.

 

[Intrastellar]

Thank you montadhar, your analysis seems perfect!

I'm glad that it is!

Choosing draw is clearly not a very good tactic. I think it is clear that the best tactic would be to choose either a winner or a loser each time. So let us ignore draws

Important note to the other members reading this:
This was clearly the case in the groups stage, where in many cases, scoring and conceding many goals was not very dangerous. But this might or might not be the case in the remaining rounds of the game, since conceding and scoring a goal becomes a lot more critical to the game.
Keep in mind that the equations in my previous post tell us that if on the average of all the games, a win/loss is more likely than a draw, then you should pick all the results to be win/loss.
On the other hand, if on the average of all the games, a draw was more likely, then you should pick all the results to be a draw!

Now, I have determined for the group stages which matches were predicted by the FIFA ranking and which were not. It turns out that the FIFA ranking predicted the correct outcome 31 times of the 40.

Ahh, that is interesting Is that for the groups stage only or did you include the matches in the round of 16 ?

So if you follow the FIFA ranking each time, then you would get a outcome accuracy percentage of 64.5%. So it seems to me that an optimal strategy would be to follow the FIFA ranking every time.

Wow! that is very interesting, it means that had we just followed the FIFA ranking, then we would have produced a better outcome accuracy, and this applies even for the highest of us!

Now, let's say that you predict a match and you say that Team A wins and you are n% sure. For example, you can say that you are 95% sure that Brazil will win against Greece.

You can now use the FIFA ranking to give you additional information. Let's say that the FIFA ranking puts Team A in a higher spot, then using Bayes' theorem, we can now be an amount of

$$\frac{31n}{9 + 22n}$$

sure of a win of Team A. So in our example, if we were 95% sure of our win and FIFA agrees, then we can now be

$$\frac{31\cdot 0.95}{9 + 22\cdot 0.95} = 0.98$$

certain. If you were only 50% sure of your win, then by taking FIFA rankings in account, you can now be

$$\frac{31\cdot 0.5}{9 + 22\cdot 0.5} = 0.77$$

certain of your choice!

On the other hand, if the FIFA ranking disagrees, then your updated certainty is

$$\frac{9n}{31 - 22n}$$

So if you were 95% sure of your choice, then by taking FIFA into account, you can now only be certain for 84%. If you were only 50% sure of your choice, then by taking FIFA into account, you can now only be 22% certain.

Ahh, that is interesting. The difference of 40 FIFA ranks between the playing teams would give a much greater from support than the difference of 3 FIFA ranks, I suppose ?
That is what I meant in my previous post that we need a formula linking the difference in FIFA ranks and the probability of the outcome.
Do you know any notes for Bayes' theorem that I can review ? I'm not very familiar with it, unfortunately.

Of course, if we manage to come up with a model linking the FIFA ranking and the probability of a given outcome, and it manages to make some correct predictions. Then we don't have to worry about the draws as well, as the probabilities should be able to predict that as well

 

[micromass]

Ahh, that is interesting Is that for the groups stage only or did you include the matches in the round of 16 ?

That is for the group stage only.

Wow! that is very interesting, it means that had we just followed the FIFA ranking, then we would have produced a better outcome accuracy, and this applies even for the highest of us!

Ahh, that is interesting. The difference of 40 FIFA ranks between the playing teams would give a much greater from support than the difference of 3 FIFA ranks, I suppose ?
That is what I meant in my previous post that we need a formula linking the difference in FIFA ranks and the probability of the outcome.
Do you know any notes for Bayes' theorem that I can review ? I'm not very familiar with it, unfortunately.

Try this: http://www.math.uah.edu/stat/dist/Conditional.html It's a very good online and free course on probability, even better than most of the books!

I will try to see if I can do something with the difference in ranks.

 

[Intrastellar]

Yes, very exciting! But I really regret my choice of letting the prediction of the winner count for so many points. Because basically you just need to predict the winner and you can win, regardless how you do in the individual matches.

That's why this competition will have two winners! One overall winner for which both the winner prediction and the scores count and one winner who predicted the scores most accurately.

It is fine with me if you change the number of points for the scores as well, since the rounds haven't progressed very much. I hope that it is fine with others as well.

And, here are my predictions

June 30th:
France - Nigeria: 3 - 1
Germany - Algeria: 4 - 1

 

[micromass]

I will try to see if I can do something with the difference in ranks.

The problem with this is that I really don't have that much data. There are only 8 matches in the group stages which did not go according to the FIFA rankings. This is far too few to have any statistical significance in my opinion. I'll still do the analysis, but don't expect anything definite.

Another problem is the following: if Team A is ranked a lot higher than Team B, then somehow we expect Team A to have an "easier time". I am not sure how to quantify this. One way to quantify this (and something I will check) is to check the goal difference, but I doubt this is a very good way of quantifying this.

 

[micromass]

It is fine with me if you change the number of points for the scores as well

What do you propose exactly?

 

[Intrastellar]

What do you propose exactly?

I have proposed something before when you first announced the increase in points.

This is a better idea than simply multiplying by 2 :)

How about a point for identifying each team's goals ? This way the number of points will be doubled from the group stage, and predicting the goals will become more significant.

Another idea can be 1 point for the exact number of goals, and 1 point for the correct goal difference.

For a 2-0 match: 2-0 should give you 4 points, 3-1 is 3 points, 2-1 is 2 points, 1-1 with a penalty win is 1 point, and of course, 0-2 does not give you any point.

For a 2-2 "A" qualifies match: 2-2 "A" qualifies should give you 4 points, 1-1 "A" qualifies is 3 points, 2-2 "B" qualifies is 3 points (is that fine ? or should it be modified as a special case ?), 1-1 "B" qualifies is 2 point (same note as the previous one), 1-0 "A" qualifies is 1 point, and 0-1 "B" qualifies does not give you any point.

The 2-2 "B" qualifies and 1-1 "B" qualifies can have 1 point subtracted to each as a special case.
Or wait, I have an even better idea. Since it seems to be more advantageous to predict win/loss than a draw, having predicting a correct draw (as in the previous 2 noted cases) gives you automatically 2 points should encourage everyone to predict draws, and add more balance to the pointing system

Of course, this is just an idea, any other idea can work. As I implied in my quoted post, having 2 points for correct scores is better than having 1.

 

[trollcast]

1st July
Argentina v Switzerland: 3 - 1
Belgium v USA: 2 - 1

 

[Borek]

I have proposed something before when you first announced the increase in points.


Another idea can be 1 point for the exact number of goals, and 1 point for the correct goal difference.

With some creativity, for every set of results and predictions it should be possible to propose a scoring method that will make selected player a winner.

In a way it is like a referee deciding on penalty in the stoppage time.

 

[D H]

July 1st:

Argentina v Switzerland: 2 - 1
Belgium v USA: 1 - 1, USA advances

 

[Intrastellar]

With some creativity, for every set of results and predictions it should be possible to propose a scoring method that will make selected player a winner.

In a way it is like a referee deciding on penalty in the stoppage time.

One extra point for the correct goal difference.

Brazil - Chile (28 June): 1 - 1 (Brazil advances)
cristo: 4 - 1 +1
DH: 2 - 1 +1
micromass: 3 - 1 +1
Matterwave: 4 - 3 +1
trollcast: 3 - 2 +1
montadhar: 1 - 2
Borek: 1 - 0 +1

No one gets an extra point
total increase: zero

Colombia - Uruguay (28 June): 2 - 0
cristo: 1 - 0 +2
DH: 1 - 0 +2
micromass: 1 - 1 (Uruguay advances)
Matterwave: 2 - 0 +3
trollcast: 1 - 1 (Uruguay advances)
montadhar: 2 - 0 +3
Borek: 1 - 0 +2

total increase:
Montadhar +1
Matterwave +1

Netherlands - Mexico (29 June): 2 - 1
cristo: 2 - 0 +2
DH: 1 - 0 +2
micromass: 2 - 1 +3
Matterwave: 1 - 1 (Netherlands advances) +1
trollcast: 2 - 1 +3
DennisN: 2 - 2 (Netherlands advances) +1
Borek: 0 - 1
montadhar: 3 - 1 +2

trollcast, micromass and DH each will get an extra point

total increase:
Montadhar +1
Matterwave +1
DH +1
micromass +1
trollcast +1

Costa Rica - Greece (29 June): 1 - 1 (Costa Rica advances)
cristo: 1 - 0 +1
DH: 2 - 0 +1
micromass: 2 - 0 +1
Matterwave: 1 - 1 (Greece advances) +2
trollcast: 3 - 1 +1
DennisN: 2 - 1 +1
Borek: 1 - 0 +1
montadhar: 2 - 0 +1

matterwave gets an extra point

final total increase:
Matterwave +2
montadhar +1
DH +1
micromass +1
trollcast +1
_______________________________________
These were the matches that were played when I proposed the extra point. I doubt you bothered to check that yourself. I didn't either anyway, until you posted about it.

You can start the 4 point system at the quarter-finals, or in a later stage, or scrap it altogether if you want. All I did was propose something to make the prediction of the winner less significant, which is what micromass requested.

But I have to say that I am not the type of a person to make cheating to a win my intention. If my previous interactions with PF had led you to this conclusion, then I apologise for that, and would appreciate that you point out where exactly that happened, so I can correct it and avoid it in the future.

 

[Logger]

1st July
Argentina v Switzerland: 2 - 0
Belgium v USA: 2 - 1

 

[micromass]

You can start the 4 point system at the quarter-finals, or in a later stage, or scrap it altogether if you want. All I did was propose something to make the prediction of the winner less significant, which is what micromass requested.

OK, but I doubt that actually solves the problem. You see the issue is that if Brazil wins (for example) that I get 16 points. So even if I do absolutely terrible in predicting the scores, I can still win because of the massive boost of scores due to the winner. Your system will give some extra points, but overall everybody will get many extra points. There won't be something to allow a person to gain a 16 point advantage.

But I have to say that I am not the type of a person to make cheating to a win my intention. If my previous interactions with PF had led you to this conclusion, then I apologise for that, and would appreciate that you point out where exactly that happened, so I can correct it and avoid it in the future.

I am certainly not implying that you did this to cheat. I know you are honest.

 

[D H]

At this point I'd rather leave the PF Futbol points system as-is. Per Arrow's impossibility theorem, the result will be unfair to someone no matter how we tweak the system.

 

[Borek]

At this point I'd rather leave the PF Futbol points system as-is. Per Arrow's impossibility theorem, the result will be unfair to someone no matter how we tweak the system.

That's more or less what I meant.

But I have to say that I am not the type of a person to make cheating to a win my intention. If my previous interactions with PF had led you to this conclusion, then I apologise for that, and would appreciate that you point out where exactly that happened, so I can correct it and avoid it in the future.

Never meant that, and as you have correctly guessed I have not bothered to check what the result of a modified system would be. I was just pointing to the fact there is no such thing as a "just system". Sorry if you got it differently.

 

[Intrastellar]

July 1st:

Argentina - Switzerland: 3 - 1
Belgium - USA: 2 - 1

 

[Matterwave]

One extra point for the correct goal difference.



No one gets an extra point
total increase: zero



total increase:
Montadhar +1
Matterwave +1




trollcast, micromass and DH each will get an extra point

total increase:
Montadhar +1
Matterwave +1
DH +1
micromass +1
trollcast +1



matterwave gets an extra point

final total increase:
Matterwave +2
montadhar +1
DH +1
micromass +1
trollcast +1
_______________________________________
These were the matches that were played when I proposed the extra point. I doubt you bothered to check that yourself. I didn't either anyway, until you posted about it.

You can start the 4 point system at the quarter-finals, or in a later stage, or scrap it altogether if you want. All I did was propose something to make the prediction of the winner less significant, which is what micromass requested.

But I have to say that I am not the type of a person to make cheating to a win my intention. If my previous interactions with PF had led you to this conclusion, then I apologise for that, and would appreciate that you point out where exactly that happened, so I can correct it and avoid it in the future.

Looks like I would profit from this...time to get out the campaign posters...

 

[DennisN]

Yes, very exciting! But I really regret my choice of letting the prediction of the winner count for so many points. Because basically you just need to predict the winner and you can win, regardless how you do in the individual matches.

That's why this competition will have two winners! One overall winner for which both the winner prediction and the scores count and one winner who predicted the scores most accurately.

That's a good idea, IMO.

Now, hopefully we will have some further excitement during the rest of the cup . Looking at the graph in post #5:

I took the lead and passed montadhar after game 21 (x=21). Will montadhar manage to retaliate and pass me? And how will cristo perform?

Trollcast, cristo, micromass and DH have been competing for a long time. Trollcast and cristo have been close during a large part of the cup, while micromass has been a constant challenger to them both, repeatedly passing them/falling behind many times. DH has been somewhat lurking close below trollcast, cristo and micromass, and now DH has managed to reach and match trollcast.

And then we have the late bloomers PhysicsGente, Logger, Matterwave and Borek.

PhysicsGente made a distinct impression during the matches 18 - 21, but has been silent since then. What would have happened if PhysicsGente continued playing? Sadly, we will never know.

Logger has been creeping upwards, but has been a bit unlucky. Also, Logger did not vote for quite a number of games.

Matterwave got his/her first point in match 21. After match 31, Matterwave started to display a quite impressive performance, and has now managed to close in on e.g. trollcast and DH.

Borek, who is using an alphabetical strategy, has managed to pass PhysicsGente.

All in all, an exciting game! :thumbs:

 

[Intrastellar]

There is still much unpredictability with regards to who will get the winner points. What will happen to me and DennisN if Germany is out before the final ? Will we still be in the top if Germany ends up as the runner-up ? There is still much unpredictability, which is what makes the game fun
If Matterwave manages to win this game, it will be the biggest upset of the world cup predictions' predictions.

 

[D H]

Argentina isn't looking that impressive today. At least not so far (40 minute mark).

The supposedly top eight teams just haven't looked that impressive this World Cup to me.

1. Spain. Gone.
2. Germany. Tied the fourth place team (Ghana) in group G; struggled against Algeria yesterday.
3. Brazil. Tied Mexico, the surprise second place team in group A; struggled massively against Chile in the round of 16.
4. Portugal. Gone.
5. Argentina. Not that impressive in a weak group; in serious danger of going home today.
6. Switzerland. So unremarkable I forgot to put them in the list! Far too many goals against in the group stage.
7. Uruguay. Biters!
8. Colombia. Maybe they're the ones this year?

Edit: I just noticed my top eight list only had seven entries. I forgot Switzerland (who are playing now!).

 

[DennisN]

Argentina isn't looking that impressive today. At least not so far (40 minute mark).

The supposedly top eight teams just haven't looked that impressive this World Cup to me.

I agree.

 

[Borg]

Probably too late but I will say this:
Belgium - USA: 0-3

 

[Logger]

Brazil - Colombia (4 July): 1 - 1 Colombia win on penalties

France - Germany (4 July): 0 - 1 Germany win

Netherlands - Costa Rica (5 July): 2 - 1 Netherlands win

 

[micromass]

Well, the round of 16 is over now. Congratulations to montadhar for taking the lead once again! Although the difference is quite small!

 

[D H]

July 4th:
Brazil vs Colombia: 2-1
France vs Germany: 1-1, France advances

July 5th:
Netherlands vs Costa Rica: 3-0
Argentina vs Belgium: 2-0

 

[Matterwave]

July 4th:
Brazil vs Colombia: 2-2 Brazil Advances
France vs Germany: 1-2

July 5th:
Netherlands vs Costa Rica: 2-0
Argentina vs Belgium: 1-0