The last post looked at how tremendously mediocre (and that is being generous) last season’s predictions were. But in terms of assessing data for predictive purposes, one season does constitute a sufficient sample. Yes, I should have backtested the accuracy of at least the Pythagorean method to determine whether it is worthwhile, or what weighting it should be given, but – and here is the honesty – I forgot! I will try to do that during the season.
So for 2014-15 I have effectively repeated the process from last season, but with some amendments. Last season’s summary came at predicting a team within 2 places of its final position on average, but the variation was large at times, and in a league of only 12 teams it is hardly something to shout about. Nonetheless, let’s see what the components say.
Unsurprisingly this reflects last season very closely, although the model does allow for Bath to pip Harlequins to 4th place. Given that this final play-off position was so close last season it is no real surprise. Mid table is still close, but rather worryingly Gloucester actually appear to have over-achieved last season, and now have a predicted points total of only 41 compared to the dismal 44 they had before. From a personal perspective, I am hoping that this will be countered by the dramatic changes in personnel at Kingsholm!
The bottom three provide a difficulty, as it is not possible to compare the data for the promoted London Welsh team. However, looking at last season’s data things don’t look good for Newcastle who ‘should’ have finished bottom on 19 points. The gap between them, Worcester (22 predicted) and London Irish (40 predicted) shows a clear gap, and no reason to assume that the relegation battle won’t be between Newcastle and London Welsh this year.
Bookmakers and Media
As before, these represent a combination of ‘market’ and ‘expert’ opinion. Creating a bookmaker summary leads to a top four of Saracens, Leicester, Northampton and Bath, which is the same teams as the media, but the press appear to favour a Northampton repeat. All predict the same bottom three of London Irish, London Welsh and Newcastle, with the latter two some way off the pace, but the media ‘favour’ Newcastle to finish bottom whilst the bookies have London Welsh sharing that dubious honour.
Of course, the gut call can’t differ dramatically from the other analysis above. However, in addition to Northampton, Saracens and Leicester I favour Harlequins (actually I placed them third, above Leicester). Two underperforming seasons are likely to see a fightback from that team. I actually placed Northampton at the top of the table; their performance in the final last year (which I was fortunate enough to see first hand) was excellent, and they just seem to have that little bit more in the tank.
Mid-table will be mid-table. My admitted personal bias sees a great improvement in Gloucester, heading up to fifth, whilst I feel that Bath’s results last season were more of an exception than a rule. The bottom three remain the same, with London Irish someway clear in 10th, but I am giving Newcastle the benefit of the doubt in terms of experience in the Premiership, and the difficulty London Welsh might experience in integrating all the player changes. But this will not be the case if Newcastle can’t score more tries – their total of 23 last season was dreadful, and scoring only 281 points was the worst in a decade.
So how was the data used? In this case, for the Pythagorean approach, the press and the gut call a simple points value was applied to the predicted final position; that is 1st place ‘scored’ 12 points, 2nd place 11 points etc. down to 1 place for 12th place. The exception to this rule was the ‘bookie’ predictions, whose scores were weighted to account for their increased accuracy last season (increased by 33%). Each team’s total score was then totaled, with higher scores better (range of 51.96 down to 4.33).
This would appear to indicate a safe bet for the top three, but this time Bath just edge out ‘Quins for the last play off position. And by ‘just’ I mean ‘just’, although my prediction of a good ‘Quins season may have distorted it. Although Gloucester should do better, it won’t be enough (yet) to break into post season action. The bottom two appear clearly adrift, but really too close to call.
Time will tell, but I will be surprised if any team is as far out as Gloucester was last season!
Next on the list is match betting (or match predictions if you prefer). I did try an Elo system for the football last season, but was overwhelmed by the maintenance. Perhaps with 12 teams (rather than 20) it might be easier to manage – at least until I work out how to automate the whole thing!