I've seen an analysis arguing that Dota+ win probabilities can be misleading in professional matches. Are they still useful for following a game, or should we ignore them?
How much should we trust Dota+ win probability in pro games?
Answers
I'd treat the number as a rough prompt, not a verdict. If it suddenly favors one team, look at what changed in the game and see whether that explanation makes sense. A percentage on its own doesn't tell you why a team is ahead.
The analysis raises concerns about a preference for Dire and predictions being pulled toward 50-50 in pro games. Those are reasons to question the display, though the source summary doesn't give enough detail to independently check the size of those effects.
Reply to this answer
The distinction between public and professional matches matters here. The same analysis described the public-match predictions as generally very good, while arguing that the pro-game predictions were much worse. That doesn't mean the feature is equally unreliable everywhere.
For a proper test, you'd want to compare predictions across plenty of pro games with actual outcomes, including how often teams win when given similar probabilities. You'd also want to check the claimed side bias and draft performance rather than judge the model from one strange-looking game.
Reply to this answer
For watching a match, I wouldn't throw it out completely. It can give you something to discuss, but I'd rather hear an explanation of the draft and what's happening on the map than have the percentage settle the argument.
Reply to this answer
Add your answer
Answers are reviewed before publication.
P2!ArmorIndex:
And a prediction being wrong in one game isn't enough to show it's bad. Even a team given a 20 percent chance should win sometimes.