Superugby R2 predictions: will they fare any better than R1?
Following on from last week, here is an update from Dr Ian Durbach! Thanks again to Ian for putting the article together and for the data from the guys at Superbru…
Dark days for the Rugby Scientists prediction team this past weekend. In case you missed it, here’s how our predictions fared. Games we got correct are in bold but you won’t need to search too hard – there’s only one.
There are always going to be some upsets that predictions get wrong, that’s what makes sport exciting to watch. Results like the Rebels win were totally unexpected. Only 2% of Superbru users got that one right, die-hard Rebels fans we’d suspect. Models get their power by being right on average, so we’re hoping ours will bounce back. The fact that the away teams won 6 of the 7 games is something to watch, particularly as home ground advantage plays a strong role in our predictions.
This week we’re adding two new elements to our model. Firstly, the good folks at Superbru have provided us not just with the overall prediction given by their users, but how those predictions differ between fans of the teams involved, and neutrals. From some of our previous research we know that fans tend to overestimate their team’s prospects by about 5 points. That means we can adjust for that bias, as well as the home ground bias we talked about last week, before making our own predictions.
Here’s what Superbru fans expect to happen this weekend. Take a look at how different the forecasts of the two sets of fans are. Nothing like some good old-fashioned fan bias!
In addition to the new fan predictions, we’ve now got some game results that we can use to take into account a team’s current form. The idea of “form” suggests that teams that win in one week will, on average, tend to keep on winning. Our previous research has suggested that for every 10 points a team wins by in one week, they tend to, on average, win by 2 points the next week. In other words, scoring 10 points this week is “worth” 2 points the next week. Take the upcoming game between the Chiefs and the Brumbies. Both teams won last weekend, the Chiefs by 5, the Brumbies by 44. In making predictions for this week, the Chiefs 5 points are now “worth” 1 point, the Brumbies 44 points are “worth” 9 points. So with just this information we’d predict the Brumbies to beat the Chiefs by 8 points. Of course this is a pretty crude way of making a prediction, but we’ll add it to our existing model and hope it improves our performance!
To recap, our predictions are now made up of three components: predictions made from data on how teams did in previous seasons, which gives an idea of overall team quality; predictions made from recent performances, which tells us about current form, and Superbru predictions, which captures other kinds of information like team selections, injuries, and so on. Our final prediction is an average of these three predictions, let’s hope it does a little better than last week!
What are your thoughts on our predictions? Are we better off tossing a coin or is there some value in the predictive model proposed in this article?
YES!! I love this!…especially the weighting of the topical/up-to-date knowledge of the superbru users. Brilliant!
Is there any merit in using data from further back than just the last 4 years? (I have it all…but not convinced of it’s relevance).
Keep the predictions coming, I’ll be following with interest.
Thanks Rob, glad you find what we’re doing interesting. Its hard to know exactly how far back to look, but our suspicion is that most of the information will be captured looking back one or two years — those are the players playing now, with a second year smoothing things out a bit in case a team had an off season. Ideally though, one would throw all the data you have into a model, and let it use whatever is useful. If you do this, let us know what you find!