There have been some interesting threads started by Jonathan Chance on election predictions (such as here and here).
Let’s assume we take only the top 3 predictions for each poster (to keep the maths simpler.)
For example, we could have predictions like these examples for Iowa:
Poster A
Trump 38%
Cruz 20%
Rubio 15%
Poster B
Cruz 30%
Trump 25%
Bush 20%
The actual result was:
Cruz 28%
Trump 24%
Rubio 23%
So we can make a formula for each poster’s prediction based on (say):
- the difference between predictions and actual for the top three candidates (always take as positive)
- a ‘penalty’ for each top three candidate missed (I’ll use 25%)
- then take the total of the above two and subtract from 100 (just to give a positive result)
Using my formula, the above two predictions would score:
100 - (14+8+8) = 70 for Poster A
100 - (2+1+25) = 72 for Poster B
So B does slightly better, despite missing one of the top 3.
What do you think of my formula?
Have you got a better one?!
If I get enough interest, I’ll have a go at ranking the posters for each State.