Originally Posted by
ireallyshouldknowbetter
Firstly, as a numbers freak, I welcome this site, it's quite fascinating and I look forward to checking in on its projections.
I would have a question regarding how you have verified its accuracy. Basically, I am wondering if you've backtested it. When building algorithms to project future outcomes, I would say the only way to be confident in future performance would be by running blind tests of previous outcomes using the model.
As others have pointed out, some of the constituencies are likely inaccurate. Much as I would love to think Joan Burton only has a 4% chance of election in Dublin West, she's at evens in the betting markets so factoring in Paddy's enormous over-round, the market gives her about a 40% chance of election.
Again, much as I'd love to think Kelly has only a 3% chance of election - and much as I don't believe either him or Burton will win election - he has more than the 3% probability you give him. He's at 1/3 to win a seat.
Howlin holds arguably Labour's safest seat, so a 1% probability is.... bold. The market has him at 1/10 reflecting, after over-round, a true probability in the region of 85%. You have the Shinner as a 100% probability; he's 2/1 in the markets.
All of which suggests possible structural issues with the model in terms of calculation of Labour candidates probability.