Predicting Media Winners - Britain's Got Talent (#bgt)

Seven weeks ago Susan Boyle exploded into world consciousness with an unliley and amazing performance in the early rounds of Britain's Got Talent. Tonight was the final, where she was up against 9 other acts that had also come through the process. For the record, the results were:

1st: Diversity (dance act)
2nd: Susan Boyle (singer)
3rd: Julian Smith (saxophonist)

I was quite interested to see how various types of prediction of the winner may work. I looked at Prediction Markets, Twitter, and the Delphi Technique.


Prediction Market

The diagram above show how wrong the Hubdub prediction market was. At close, Susan Boyle was at 78%, Diversity was 6%, another singer was at 6% and Julian Smith a lowly 1%. As you can see it precited many others ahead of Julian.


Twitter

I looked at two potential sources - Trending, and my own Twitterstream.

- Trending had Susan Boyle, Diversity and Stavros Flatley ( a p*ss-taking dance act ) in the top 3 places in that order at point of close.

- My own stream was very much in favour of Diversity, loved Stavros Flatley, and treated Susan Boyle as no longer cool.

Neither registered Julian Smith in any significant way.


Delphi Technique

The Delphi technique involves gathering a diverse group of experts to cover a subject. It differs from crowdsourcing in that its a small (ie non statistically relevant) group rather than a mass. I used my family in this test, and the support was:

Diversity 60%
Stavros Flatley 20%
Julian Smith 20%

When asked about Susan Boyle, the view was "she'll probably win, but this is who we like".


Results

Delphi technique was the most accurate (assuming we adjusted for the opinions on Susan Boyle as everyone's 2nd vote), which is a fascinating result.

Prediction market did the worst, which is equally fascinating. I'd have predicted the reverse would have been true.

Neither the Twitter Trending nor my Followed stream saw Julian Smith at all - hypothesis is that the people he would appeal to are by and large not yet on Twitter - ie my family group is probably a better representative sample of the overall market than Twitter is today.

Another reason to beware of Twitter buzz (see my earlier discussion of this over here).

(By the way, the post event Twitterstream is broadly in agreement with the outcome, as opposed to last night's furious reaction to the outcome)