6 Nations Results Machine

Hi 606v2

Long-time reader, first time contributor.

Some friends and I have set up Rugby4Cast, a rugby prediction model which looks at past results and aims to predict future international rugby results.

https://sports4cast.com/

Our model uses the data to work out the average and standard deviation of points scored for each team, both home and away. It then skews these scores depending on each team’s respective ranking (for example, Fiji typically score very highly at home, but wouldn’t be expected to score their usual 35 points if playing New Zealand).

The match is then simulated 5,000 times to obtain the most likely outcome.

https://sports4cast.com/the-data-and-the-model/

We have run this model back over the historical data, and (with a little tweaking to take into account certain variables) would have predicted the correct result in around 80% of the nearly 6,000 matches played, with an average score difference of 10.19.

We did manage to get the first weekend’s results correct but were some way off on the score margins - we are working on this!

Predicted results



We have previewed this weekend’s games and are predicting the following results


FYI these are the finishing positions of the respective sides that we expect:



Please note that there is no national bias in the model and all our predictions are based solely on the results of previous games.

Enjoy and please let us know if you have any questions.

I can't see how a statistical program can give a 20 point win for England vs Wales. (EDIT: I can but think you might be ignoring crucial information).

I'm presuming it's world ranking only and no factor on previous results between the teams? If so that's a mistake IMO.

It's an anti-Welsh machine

Specific head to heads aren't taken into account - the model simply looks at the number of points a team scores at home or away from home and adjusts these based on rankings.

England have rarely failed to score more than 30 points at home over the past couple of years (failing to do so against Argentina in 2016 and 2017 and Ireland and Wales in 2016 - 4 times in almost two years), whereas Wales haven't been scoring particularly well against higher ranked opposition away from home.

Given that the model does need finessing, we will consider whether two sides' head-to-head may be more likely to predict a scoreline that two sides' other results though, so thank you for your observation.

Also I need to point out as a matter of urgency that the man in charge of developing this is Scottish and he has no particular love for the English!

rugby4cast wrote:Specific head to heads aren't taken into account - the model simply looks at the number of points a team scores at home or away from home and adjusts these based on rankings.

England have rarely failed to score more than 30 points at home over the past couple of years (failing to do so against Argentina in 2016 and 2017 and Ireland and Wales in 2016 - 4 times in almost two years), whereas Wales haven't been scoring particularly well against higher ranked opposition away from home.

Given that the model does need finessing, we will consider whether two sides' head-to-head may be more likely to predict a scoreline that two sides' other results though, so thank you for your observation.

Wales do horribly against SH opposition. Rarely do they lose big in the 6 Nations. Stylistic matchups are important IMO. And history for the physiological aspect. When did England last beat Wales by 20?

Just going by rankings England would beat France by more than Fiji. I don't think there's a single person who thinks that is the case.

Scottrf wrote:

rugby4cast wrote:Specific head to heads aren't taken into account - the model simply looks at the number of points a team scores at home or away from home and adjusts these based on rankings.

England have rarely failed to score more than 30 points at home over the past couple of years (failing to do so against Argentina in 2016 and 2017 and Ireland and Wales in 2016 - 4 times in almost two years), whereas Wales haven't been scoring particularly well against higher ranked opposition away from home.

Given that the model does need finessing, we will consider whether two sides' head-to-head may be more likely to predict a scoreline that two sides' other results though, so thank you for your observation.

Wales do horribly against SH opposition. Rarely do they lose big in the 6 Nations. Stylistic matchups are important IMO. And history for the physiological aspect. When did England last beat Wales by 20?

Just going by rankings England would beat France by more than Fiji. I don't think there's a single person who thinks that is the case.

2015, wasn't it? World Cup warm-up game?

EDIT: My mistake, it was the pre-summer tour warm-up game in 2016 I'm thinking of - and now I've looked it up I'm surprised it was only 27 - 13.

It was 5 tries to one.

You need an algorithm in the machine for 'passion' and rivalry! Like with football derbies which often rip up the form book and produce odd looking results.

So we have looked at this a little further and the results were interesting.

Typically, Wales tend to consistently outperform our predictions when playing against England.

Based on our model, England would have beaten Wales in 2008 and 2012 in the Six Nations as well as in the 2015 World Cup game. Given that the percentage of games we have correctly predicted is still over 80%, we think that this points to Wales' ability to lift their game against England rather than a fundamental flaw in the model and it will be interesting to see if this weekend's game bears this out.

rugby4cast wrote:So we have looked at this a little further and the results were interesting.

Typically, Wales tend to consistently outperform our predictions when playing against England.

Based on our model, England would have beaten Wales in 2008 and 2012 in the Six Nations as well as in the 2015 World Cup game. Given that the percentage of games we have correctly predicted is still over 80%, we think that this points to Wales' ability to lift their game against England rather than a  fundamental flaw in the model and it will be interesting to see if this weekend's game bears this out.

What about Scotland vs Australia?

England vs SA?

I predict that Scotland and SA respectively outperform your predictions. Scotland recently at least.

The Oracle wrote:You need an algorithm in the machine for 'passion' and rivalry!  Like with football derbies which often rip up the form book and produce odd looking results.

But are they odd though? Or do we just think they're odd because we are ourselves are swayed by intangibles. I like removing these intangibles. Where it falls down, for me, is that the rankings are very flawed as not everyone plays each other often enough for you to then use them in direct head to head match ups.

If a (normal) coin is tossed 99 times in a row and it has come up heads every time - the probability of it coming up tails on the 100th throw is still 50%!

The Great Aukster wrote:If a (normal) coin is tossed 99 times in a row and it has come up heads every time - the probability of it coming up tails on the 100th throw is still 50%!

No it's not, as it's clearly a loaded coin.

Scottrf wrote:

The Great Aukster wrote:If a (normal) coin is tossed 99 times in a row and it has come up heads every time - the probability of it coming up tails on the 100th throw is still 50%!

No it's not, as it's clearly a loaded coin.

Nothing is loaded in Vegas. Gamblin' is always honest... so stop giving coin tosses a bad name.

I was taking some comfort from the orginal post, until rugby4cast pointed out previous predicted England wins over Wales had turned out to be losses.

I find exercises like this interesting. Some supporters might argue you need to look at the referee as well as the venue. That would be in line with Scottrf's suggestion about style matchups. Some teams just seem to get on the wrong side of certain officials but perhaps not sure how you'd go about measuring that.

Rugby Fan wrote:I was taking some comfort from the orginal post, until rugby4cast pointed out previous predicted England wins over Wales had turned out to be losses.

I find exercises like this interesting. Some supporters might argue you need to look at the referee as well as the venue. That would be in line with Scottrf's suggestion about style matchups. Some teams just seem to get on the wrong side of certain officials but perhaps not sure how you'd go about measuring that.

Wayne Barnes is worth 3 tries to Ireland's opposition.

Scottrf wrote:

Rugby Fan wrote:I was taking some comfort from the original post, until rugby4cast pointed out previous predicted England wins over Wales had turned out to be losses.

I find exercises like this interesting. Some supporters might argue you need to look at the referee as well as the venue. That would be in line with Scottrf's suggestion about style matchups. Some teams just seem to get on the wrong side of certain officials but perhaps not sure how you'd go about measuring that.

Wayne Barnes is worth 3 tries to Ireland's opposition.

I'm dreading the Scotland game... two yellow cards to begin with to Ireland for high tackles in the scrum.... then a red much later in the game (13 minutes in) for cheek directed at ref by Sexton.

rugby4cast wrote:Specific head to heads aren't taken into account - the model simply looks at the number of points a team scores at home or away from home and adjusts these based on rankings.

England have rarely failed to score more than 30 points at home over the past couple of years (failing to do so against Argentina in 2016 and 2017 and Ireland and Wales in 2016 - 4 times in almost two years), whereas Wales haven't been scoring particularly well against higher ranked opposition away from home.

Given that the model does need finessing, we will consider whether two sides' head-to-head may be more likely to predict a scoreline that two sides' other results though, so thank you for your observation.

I think head to heads should be taken into account. Certain teams always seem to struggle against others. Wales v Australia being one of the more extreme examples. Ireland v France another (possibly changing).

Scottrf wrote:

Rugby Fan wrote:I was taking some comfort from the orginal post, until rugby4cast pointed out previous predicted England wins over Wales had turned out to be losses.

I find exercises like this interesting. Some supporters might argue you need to look at the referee as well as the venue. That would be in line with Scottrf's suggestion about style matchups. Some teams just seem to get on the wrong side of certain officials but perhaps not sure how you'd go about measuring that.

Wayne Barnes is worth 3 tries to Ireland's opposition.

He is worth about a 10 point head start Id say.

Hi all

Thanks for the comments. Got the results right (6/6 so far) but the scorelines were pretty far away.

As has been pointed out a few times, certain teams tend to lift their games for certain opposition and we are looking to do a piece on these. I am also inclined to agree that some teams just do better against others - don't know whether this is to do with contrasts in styles or whether it just reflects, as I'm sure a lot of people would say about Wales against England, that some teams seem to want it more in certain games.

Please take a look

https://sports4cast.com/2018/02/12/2018-six-nation-week-2-ireland-v-italy-debrief/

https://sports4cast.com/2018/02/12/2018-six-nations-week-2-england-v-wales-debrief/

https://sports4cast.com/2018/02/12/2018-six-nations-week-2-scotland-v-france-debrief/

FYI, despite their inability to score for 50 minutes at home and Ireland's smashing of Italy, we've still got England as favourites for the tournament.

Also in other news, we ran our algorithm against the Americas Rugby Championship and did an awful lot better...

Do you use this for betting, rugby4cast? And if so, are you making decent money?!

rugby4cast wrote:Also in other news, we ran our algorithm against the Americas Rugby Championship and did an awful lot better...

Hi R4C. You said that you’re a longtime follower of the forum but this is your first post. It would be useful for our Fine Analytics team to know what was the reason for not posting until now? Did you think you wouldn’t fit in? Was it something someone said?

He asked World Rugby could he join but they never got back to him until recently. They were too busy trying to work their way through the oceans of clarification requests over the years from one Warren Gatland.

rugby4cast

Interesting stuff. Two things:
Agree with others that you need to consider H2H performances as well. There are clearly issues of style match ups and also basic psychology - Scotland losing in Wales has become more than just a habit, and similarly Wales inability to beat Australia.

Scores are increasingly being affected by the weather. In wet conditions, defences are generally on top and games will have much lower scores than the equivalent in dry conditions. Probably doesn't affect the winner of the match very often, but will impact the points spread.

The first point I guess could be dealt with by looking at win % between two teams over the previous say 5 meetings and using that as a multiplying factor. Still wouldn't have got you that close to the Wales v Scotland scoreline, as that was just a good Wales performance and an abject Scottish one, which would be a statistical outlier.

Might just be enough data to look at how bad weather affects points scoring overall, but I doubt any team has played enough matches in the rain to allow a reliable analysis to be carried out on how it affects individual performance.

Oh, and is the predicted table based on projections at the start of the tournament, or on current position after the first two results?

OK chaps, we've been beavering away and have updated the model to take into account head to heads! It affects the predictions depending on how many head to heads have been played, and has improved our overall accuracy (but, disappointingly, not massively).

Regarding the overall chance, they are updated at the end of each round. Here's the latest...





Sent from Topic'it App