Super 18 Predictions for Round 4
Team Ratings for Round 4
The basic method is described on my Department home page.
Here are the team ratings prior to this week’s games, along with the ratings at the start of the season.
| Current Rating | Rating at Season Start | Difference | |
|---|---|---|---|
| Hurricanes | 16.98 | 13.22 | 3.80 |
| Chiefs | 11.28 | 9.75 | 1.50 |
| Highlanders | 8.08 | 9.17 | -1.10 |
| Crusaders | 8.08 | 8.75 | -0.70 |
| Lions | 6.71 | 7.64 | -0.90 |
| Waratahs | 2.92 | 5.81 | -2.90 |
| Brumbies | 2.89 | 3.83 | -0.90 |
| Stormers | 2.68 | 1.51 | 1.20 |
| Sharks | 2.05 | 0.42 | 1.60 |
| Blues | 0.88 | -1.07 | 2.00 |
| Bulls | -0.76 | 0.29 | -1.00 |
| Jaguares | -2.83 | -4.36 | 1.50 |
| Cheetahs | -7.01 | -7.36 | 0.40 |
| Force | -8.10 | -9.45 | 1.40 |
| Reds | -9.14 | -10.28 | 1.10 |
| Rebels | -12.37 | -8.17 | -4.20 |
| Kings | -19.01 | -19.02 | 0.00 |
| Sunwolves | -20.42 | -17.76 | -2.70 |
Performance So Far
So far there have been 26 matches played, 18 of which were correctly predicted, a success rate of 69.2%.
Here are the predictions for last week’s games.
| Game | Date | Score | Prediction | Correct | |
|---|---|---|---|---|---|
| 1 | Chiefs vs. Hurricanes | Mar 10 | 26 – 18 | -3.60 | FALSE |
| 2 | Brumbies vs. Force | Mar 10 | 25 – 17 | 15.40 | TRUE |
| 3 | Sharks vs. Waratahs | Mar 10 | 37 – 14 | 0.40 | TRUE |
| 4 | Blues vs. Highlanders | Mar 11 | 12 – 16 | -3.70 | TRUE |
| 5 | Reds vs. Crusaders | Mar 11 | 20 – 22 | -14.70 | TRUE |
| 6 | Cheetahs vs. Sunwolves | Mar 11 | 38 – 31 | 18.80 | TRUE |
| 7 | Kings vs. Stormers | Mar 11 | 10 – 41 | -16.50 | TRUE |
| 8 | Jaguares vs. Lions | Mar 11 | 36 – 24 | -7.90 | FALSE |
Predictions for Round 4
Here are the predictions for Round 4. The prediction is my estimated expected points difference with a positive margin being a win to the home team, and a negative margin a win to the away team.
| Game | Date | Winner | Prediction | |
|---|---|---|---|---|
| 1 | Crusaders vs. Blues | Mar 17 | Crusaders | 10.70 |
| 2 | Rebels vs. Chiefs | Mar 17 | Chiefs | -19.70 |
| 3 | Bulls vs. Sunwolves | Mar 17 | Bulls | 23.70 |
| 4 | Hurricanes vs. Highlanders | Mar 18 | Hurricanes | 12.40 |
| 5 | Waratahs vs. Brumbies | Mar 18 | Waratahs | 3.50 |
| 6 | Lions vs. Reds | Mar 18 | Lions | 19.80 |
| 7 | Sharks vs. Kings | Mar 18 | Sharks | 24.60 |
| 8 | Jaguares vs. Cheetahs | Mar 18 | Jaguares | 8.20 |
David Scott obtained a BA and PhD from the Australian National University and then commenced his university teaching career at La Trobe University in 1972. He has taught at La Trobe University, the University of Sheffield, Bond University and Colorado State University, joining the University of Auckland, based at Tamaki Campus, in mid-1995. He has been Head of Department at La Trobe University, Acting Dean and Associate Dean (Academic) at Bond University, and Associate Director of the Centre for Quality Management and Data Analysis at Bond University with responsibility for Short Courses. He was Head of the Department of Statistics in 2000, and is a past President of the New Zealand Statistical Assocation. See all posts by David Scott »
Hi David
I am interested as to the excact nature of your model. I suspect you are basing your predictions on previous results calculating factors such as a team’s past record at a venue, a team’s record against their opponent and the scoring history in the last 10 games they have played. If i am correct its very similar statistical analysis to what a betting agency uses I suspect.
However I am curious as to why you don’t calculate other factors. Like player selection, Named officials, expected weather ect. Are those factors to fickle to accurately calculate? It seems to me that your margin for error must almost solely lie in those factors and I just wonder if you think it’s possible to develop a method to calculate these factors aswell.
8 years ago
Hi David
I am interested as to the excact nature of your model. I suspect you are basing your predictions on previous results calculating factors such as a team’s past record at a venue, a team’s record against their opponent and the scoring history in the last 10 games they have played. If i am correct its very similar statistical analysis to what a betting agency uses I suspect.
However I am curious as to why you don’t calculate other factors. Like player selection, Named officials, expected weather ect. Are those factors to fickle to accurately calculate? It seems to me that your margin for error must almost solely lie in those factors and I just wonder if you think it’s possible to develop a method to calculate these factors aswell.
The blues crusaders game is a prime example for me. Certainly based on the first set of factors I mentioned I would totally agree with that scoreline. But I can’t help but feel the exclusion of so many primary players from the crusaders like up gives them a distinct disadvantage. I expect somewhere in the vicinity of 12 points to the blues.
8 years ago
My model only uses the difference between the scores of previous games and the home ground advantage.
Why not use other things? Collecting and organizing a lot of other information kills that idea straight off. I would be full time at it. The only way to collect a lot of information without spending time is using programs to scrape information from the web and that is difficult with lots of Java Script behind table production, not to mention frequent website changes which mean rewriting programs. Betting companies employ a lot of people to do that sort of stuff.
Even given that you have such information, how do you use it in a model? Even home ground advantage raises questions. When two Sydney teams play at one of the main stadiums in Sydney does the nominal home team have a home ground advantage? If I know a particular referee is officiating, how do I take that into account?
And finally, your idea that you can endlessly reduce risk with more information. There is a limit. In the end the only sensible model is that to some extent, sporting contests are random (providing no Indian or other bookmakers are involved). You can make marginal gains with more information and that is what betting agencies and professional punters do. Talking at conferences to people from betting companies, these days a lot of information, model building and computer power is used.
The lesson of the model I use is that with very limited information, with a very simple model and well chosen parameters, you can get as good a prediction as newspaper pundits. And it has the advantage of being objective: no feelings, team allegiances, etc, just the data and the model.
Blues v Crusaders? Crusaders are always well-organised it seems to me whoever they have in their team. Blues are quixotic and could easily find a way to lose as they did last game, or instead put 6 tries on the opposition. We will see.
8 years ago
There is also lots of confounding between variables mainly because there aren’t really that many observations over a season.
I remember looking at some variables to add to a model like this and the main effect of one variable was completely confounded with an interaction of two other variables. And all these variables were pretty highly regarded as important to the outcome.
I can’t remember those variables off-hand but traveling east or west for a game (which from American sports was thought to be highly predictive) and playing in either South Africa, Australia or NZ (over and above home advantage) were quite problematic to model together.
8 years ago
The east versus west travel is interesting and not something that I had thought of. I am not sure if reading your comment on Friday triggered something in my brain, but I was watching the Rebels Chiefs game on Saturday live and it occurred to me that it was late at night (9:30 pm start NZ time) and if the Chiefs hadn’t adjusted to that time it would be quite difficult for them to play at their best. Going the other way a 7:30 game would be at 5:30 by body clock for an Australian team playing in New Zealand, which would appear to be much less of a problem.
8 years ago
I am going to reply to my own comment here. I think in a way the two possibilities I suggested for Blues v Crusaders both came true, just in different halves.
8 years ago
Hi there David, I really appreciate efforts like these to attempt to objectively predict results and such (looks like your model is performing well so far). I just wanted to ask a very basic question which probably has a simple answer: as a statistician when you encounter a value such as the 3.50 prediction for the Waratahs vs Brumbies game, is there any reason to round up or down in that case (assuming you were going to make a bet with it or something), or would either be arbitrary? Thank you
8 years ago
If you wanted to round to a whole number of points that is where I would go with personal judgement.
8 years ago