Who will stop Italy?

[helene_t]

Oops no, forgot that.

[drinbrasil]

I have fitted my model to the 18 first rounds. These are the probabilities that Italy will beat each of the teams in a 128-board match:

 

Construtive comment. I think this model can be improved. Like we do in models to make odds in soccer, we use some other aspects who in bridge i imagine could be at least the recent h2h (very better information, because in soccer we have many player changes in teams). So you need use weight for actual form (here you used just the RR) and weight for H2H. I imagine h2h is important because we know in bridge our game fit more or less against one or other opponent due to systems or styles for example.

[helene_t]

Yes, my brother who is in fisheries research often uses adaptive filters for similar problems, basically it means exponentially decaying weights.

 

As for H2H I dono how to estimate it. Is it simply taking the RR-result of Italy vs. X into account? It's difficult to see how I could estimate the weights for the H2H without repeat measurements. Of course if used per-board info I would have repeated measurements. I would have to write a script to download the scoreboards then, don't like the idea of copy-pasting each board.

[drinbrasil]

For H2H you should use all last lets say X* year(s) results between teams plus some weight for RR 16boards result this year. Of course perfect would be board a board all time, but general results can be used.

 

*This isnt fixed number, in really you should use many in past datas to make model who fit more for bridge. Of course i am only saying about ideal models, i cant use my time to do this actually, but if you have time to take previous meetings between all teams in past events, you much probabily will get better estimations. Maybe someone has this work partial done around, like data from 2003-2005 and used for 2006 torneys.

[cherdano]

Btw, if you think you understand those odds better than others, you can make a little money on betfair.com :) (Actually not really, the margins between the odds you can get for backing or have to give for laying are way too big.)

[drinbrasil]

little money isnt interesting :)

[skjaeran]

I have fitted my model to the 18 first rounds. These are the probabilities that Italy will beat each of the teams in a 128-board match:

 

[1] "NORWAY 0.848930973373835"

84.9% probability that Italy will beat Norway in a head to head match?

That's ridicolous. They've met in matches like this several times before. Those matches has as a rule been very close. Since 1995 the two teams have met three times at the KO stage of the BB. Norway have won twice, Italy once.

Italy probably are favourites if they meet again, but not by much - maybe 55-60%.

 

And your probabilitiies are likewise way too high for several other teams too.

[joker_gib]

I agree with Harald, this Norvegian team is a very good one !

 

Alain

[Gerben42]

Copy and paste argument for Netherlands :)

The 1993 final replay sounds appealing!

[Fluffy]

Who will stop Italy?

No one, Go Italy!. Kick their arses.

[helene_t]

84.9% probability that Italy will beat Norway in a head to head match?

That's ridicolous.

You're probabbly right, but why does my model say 84.9% then? Which of my assumptions are more likely to be wrong?

[skjaeran]

84.9% probability that Italy will beat Norway in a head to head match?

That's ridicolous.

You're probabbly right, but why does my model say 84.9% then? Which of my assumptions are more likely to be wrong?

I'm no statitican, and I don't really know how you have computed your probabilites, nor what assumptions you've made. But knowing that the score in head to head competition in the KO stage in BB is Norway 2 - Italy 1 and that Norway actually beat Italy (narrowly) in their outing in the RR, common sense alone will tell you that your probability must be way off.

 

And even after Italy beat USA1 63-0 in the RR I can't believe the odds for an Italian victory in an hypothetical semifinal/final encounter with a 16/20 IMP carry over could be more than 60% at most.

 

The Italians are very good at crushing the weaker teams - just look to the European Championships (and there are "weak" teams relatively speaking in the BB too) - and thus always do well in round robins. But the knock out stage is something different. Even in long matches other strong teams have a reasonable chance to beat them, they are far from unbeatable, even if they are a really formiddable team.

[cherdano]

84.9% probability that Italy will beat Norway in a head to head match?

That's ridicolous.

You're probabbly right, but why does my model say 84.9% then? Which of my assumptions are more likely to be wrong?

Are you assuming that the Roundrobin is an accurate reflection of the strength of the teams? Anyway, it could also be that Harald's intuition is wrong for 128-board matches.

[Halo]

Once you define rules of competition, there is little point whining about the consequences.

[Echognome]

84.9% probability that Italy will beat Norway in a head to head match?

That's ridicolous.

You're probabbly right, but why does my model say 84.9% then? Which of my assumptions are more likely to be wrong?

I will have to agree with Harald and Arend here. It wasn't exactly clear *how* you are modeling.

 

Let me give a modest proposal. Suppose that a team's performance in a head-to-head matchup is based on some combination of their overall performance in the RR and their performance in their H2H matchup (also in the RR). Then we have a factor for how well the teams match-up with each other in addition to controlling for them having a bad match (their performance in other matches will make up for it). Start with that combination at say 50% (you can try various values of course) and then see what your numbers say.

 

There are obviously more complex ways to do this, but I'm trying to offer a relatively simple calculation.

 

It would also be interesting to see what your model says about their probability of winning a 16-board match vs a 128-board match to see if the model is internally consistent with the results from the prior 16-board matches.

 

Were using probit to model this?

[pclayton]

Helene - you do realize that you are basically giving 5:1 odds in a Norway / Italy match.

 

Somehow that doesn't feel quite right.

[helene_t]

It wasn't exactly clear *how* you are modeling.

I fit a standard linear regression model to the per-match IMP differences

{IMP difference in match i} = {strength of hometeam in match i} - {strength of vistorteam in match i} + epsilon

 

So this model gets 21 coeffiecients as the strength of USA 2 is per definition zero. Plus a standard deviation of epsilon which turns out to be somewhat more than 24 IMPs. For the purpose of modeling an 8 times longer match I scale the coefficients with a factor 8 and stdev(epsilon) with a factor sqrt(8).

 

Let me give a modest proposal.  Suppose that a team's performance in a head-to-head matchup is based on some combination of their overall performance in the RR and their performance in their H2H matchup (also in the RR).  Then we have a factor for how well the teams match-up with each other in addition to controlling for them having a bad match (their performance in other matches will make up for it).  Start with that combination at say 50% (you can try various values of course) and then see what your numbers say.

 

Tx, I will try that. Also I could try

- shrinking the coeffeicients with Mallow's C.

- there will be correlation between the residual variances of matches played with the same boards, maybe that can be exploited

- some time series analysis could be intersting (some teams may be better in the afternoon than in the evening, some may have started bad and then improved etc.)

 

It would also be interesting to see what your model says about their probability of winning a 16-board match vs a 128-board match to see if the model is internally consistent with the results from the prior 16-board matches.

The residuals look reasonably normal, but I should probably x-validate it.

 

Were using probit to model this?

No, I don't use binary outcomes, I predict the IMP differences. In the Italy-vs-the-others-table I reported the probability that the IMP difference is positive.

[helene_t]

Using RR 18 coefficients (was too lazy to get the last three rounds), the probabilities for the quarter finals, after half-way (round three), are:

 

SA-Ita: 86.1%

USA1-Aus: 99.1%

Nth-Swe: 88.0%

Nor-Chi: 97.0%

[Finch]

I don't think extrapolating from the RR matches to the QFs simply by scaling up is right, because the teams' objectives in a RR match are different, and they have information from each segment.

 

For example, I would expect the volatility of results in later sets to be much higher if the difference is large going into them, as teams start trying to swing. Which means that I think the chance of italy winning is higher than 14%.

[TimG]

Which means that I think the chance of italy winning is higher than 14%.

I bet lots of people would come to the same conclusion without any fancy math!

 

(Yes, I am aware of Italy's big deficit as I am posting this.)

[matmat]

Which means that I think the chance of italy winning is higher than 14%.

I bet lots of people would come to the same conclusion without any fancy math!

 

(Yes, I am aware of Italy's big deficit as I am posting this.)

it's all about the fancy math.

the world would have no sense and life would have no meaning, if there was no fancy math.

[han]

South Africa?

[cherdano]

South Africa?

Yeah why wasn't that an option?

[han]

I have fitted my model to the 18 first rounds. These are the probabilities that Italy will beat each of the teams in a 128-board match:

 

[1] "NORWAY  0.848930973373835"

84.9% probability that Italy will beat Norway in a head to head match?

That's ridicolous. They've met in matches like this several times before. Those matches has as a rule been very close. Since 1995 the two teams have met three times at the KO stage of the BB. Norway have won twice, Italy once.

Italy probably are favourites if they meet again, but not by much - maybe 55-60%.

 

And your probabilitiies are likewise way too high for several other teams too.

Let me just point out that this was not Helene's personal opinion. It is just the outcome of a model that has as input only the IMP scores of the round robin matches.

 

Of course there may be more relevant information, and the model may not be the best model for this purpose, but I expect that the numbers Helene posted are the correct outcomes of her model, and calling them ridiculous doesn't make much sense.

 

It's like saying that 1+1=2 is ridiculous because we should be computing 1+3.

[skjaeran]

I have fitted my model to the 18 first rounds. These are the probabilities that Italy will beat each of the teams in a 128-board match:

 

[1] "NORWAY  0.848930973373835"

84.9% probability that Italy will beat Norway in a head to head match?

That's ridicolous. They've met in matches like this several times before. Those matches has as a rule been very close. Since 1995 the two teams have met three times at the KO stage of the BB. Norway have won twice, Italy once.

Italy probably are favourites if they meet again, but not by much - maybe 55-60%.

 

And your probabilitiies are likewise way too high for several other teams too.

Let me just point out that this was not Helene's personal opinion. It is just the outcome of a model that has as input only the IMP scores of the round robin matches.

 

Of course there may be more relevant information, and the model may not be the best model for this purpose, but I expect that the numbers Helene posted are the correct outcomes of her model, and calling them ridiculous doesn't make much sense.

 

It's like saying that 1+1=2 is ridiculous because we should be computing 1+3.

If so she should have presented her assuptions and how she computed these probabilities and asked the question what's wrong. Since it should be obvious that none of the figures presented could be anywhere near correct.