Statistician proves he can pick the winners

David Scott is a statistician who picks great NRL results using a simple mathematical formula. Photo / Natalie Slade

David Scott, an associate professor, professes to know little about league yet he tips NRL results with the best of them.

While you may worship the relentless Michael Luck's crunching tackles, Dr Scott tackles Lady Luck by number crunching.

"I don't watch a whole lot of games but I follow the tipsters in the paper and I'm right up there," the transplanted Australian says from his tiny office in the statistics department at Auckland University.

Scott is running at a 63.2 per cent success rate this season, and so ranks alongside the best NRL newspaper tipsters and well ahead of others.

Bookmakers employ statistics experts to improve their returns and one such employment alert arrived in Scott's email the day of the Herald interview.

Crazy as it seems though, amateur tipsters are usually stats illiterate and rely on impressions gleaned from watching games and get overly caught up with what happened last week. Fools. Newspaper tipsters are also prone to rash predictions to make up lost ground, which invariably creates more lost ground. Bad seasons turn to disaster on the catch-up principle, the downward spiral.

In contrast, Scott relies on results, margins and the long haul. He's not overly interested in recent results, and apart from factoring in home advantage, uses a form guide built up over time.

His formula needs only limited input, unlike the one created by his initial role model, a Melbourne statistician, Stephen Clarke. The Swinburne University professor has beaten Aussie Rules allcomers for 25 years using a formula that sweeps across a blackboard like Billy Slater on a dazzling run.

Clarke predicts winners at close to 70 per cent by ranking teams on past performances and margins, taking extra account of the most recent results. Clarke, like Scott, subtracts the away team ranking from the home teams, adds in a home ground factor, and ... hey presto ... you have the winner nearly 70 per cent of the time.

Scott was lured into the sports tipping game by a request from TV3's Campbell Live. The segment never went to air, leaving Scott rating the show as a big fat zero.

But he pursued the project in order to promote the relevance of statistics to everyday life, and has done very nicely with Super rugby and the NRL.

And this from an Aussie Rules fan who hardly watches league, although he follows the Super 15 and has "watched Australian Universities play on a Saturday afternoon".

Scott scored at about 70 per cent in Super Rugby. On noticing that NRL tipsters struggled in comparison, he switched his code to the other code.

Scott's formula is much simpler than Clarke's yet it produces a similar success ratio.

Scott says: "The superscript you see in his formula doesn't seem to improve things so I left it out. Clarke appears to take into account who is playing. (The i and j subscripts in his formula). That will cause instabilities and continual adjustments. There is evidence in other variations of exponential smoothing that too much adaptation based on recent observations is not good for prediction. Also, more data is needed for estimating extra parameters as well."

Which, funnily enough, is precisely what we were saying in the Herald office this week.

On with Scott's countdown.

Once an initial rating for each team is established, the system relies on win-loss margins and a home ground factor.

He started the project by collating NRL results from 2007 to 2009 to establish the ratings then applied those to the 2010 results to find a fraction by which to multiply the margins to adjust the ratings each week. His special ratio: 0.075 in case you are interested. The home ground factor is five.

Scott's system suffers when a team like unfancied Newcastle roars out of the blocks in a new season. It also bypasses outlandish rogue factors, such as selection eccentricities.

Clarke was pipped by another Aussie Rules tipster last season, but protested that the Fremantle Dockers tripped him up with a second string lineup in one game. Scott says: "If someone used my formula for betting, it would be best to keep the option of picking a different result now and then. But my impression is that no sports system around the world beats the 70 per cent mark over time."

He has reached one hard and fast conclusion: The Super 15 is more predictable than the NRL, for both the man in the street and the man with the stats. Scott says: "I think two reasons are possible ... the salary cap makes the NRL more even and union requires more team work and organisation and thus more stable teams remain as the good teams.

"League turns much more on individual brilliance or mistakes, like Lance Hohaia in the test last week, or Billy Slater in previous tests."

Ah ha. So Scott does watch league.

Not that he uses anything beyond the results to pick the results. But he doesn't knock the newspaper tipping panelists either. "It [sports tipping] is inherently unpredictable. What you see in the newspapers is close to the limits of predictability."

The Scott footy formula

David Scott is picking the Newcastle Knights to pip the Warriors at Ausgrid Stadium tomorrow.

Before last week's win over the Titans, the Warriors had a Scott Performance Rating of 2.88. They beat the Titans by a margin of 20, which Scott multiplies by 0.075. This gives 1.5 which is added to the 2.88, giving a new rating of 4.38.

The Knights rate at -0.55, so the Warriors have an advantage which rounds to 4.90.

But add the Scott Standard Home Advantage factor of 5, and the Knights are 0.10 ahead. Simple really.