"Tips" is the number of correct tips. Draws are counted as correct.
"Bits" from Monash University Probabilistic Footy Tipping rewards tipsters for saying a win was more likely and punishes them for saying it was unlikely. Higher is better.
"MAE" is Mean Absolute Error, which is the average difference between predicted and actual margins. Lower is better.
"Correct" is the percentage of correct tips, i.e. "Tips" divided by the total number of tips provided (which is normally also the number of games played, assuming the model tips all games).
[+] Bits, MAE, Tips... which is best?
Tips is what most people care about: How many winners did you pick? So it's the
primary variable for the Squiggle leaderboard. But there's a fair bit of luck involved in tipping 50/50 games,
so it's hard to know whether a high Tips score is due to
skill or good fortune.
Bits are earned by confident correct tips, lost by confident incorrect tips,
and not much happens either way for fence-sitting 50/50 tips. The added dimension of "confidence" means Bits are less influenced by luck, so
high scores are likely to indicate models that are good performers over the long-term.
MAE (or Mean Absolute Error) measures how far, on average, margin tips are from the actual margin.
(Lower MAE scores are better.)
It's hard to fluke a good MAE, because your tips have to be consistently close to the real margins, and therefore it's probably the best
of the three metrics at measuring a model's underlying forecasting skill. However, it doesn't measure whether anyone is getting the winners right,
which means it's not directly tracking the most important factor.