I've seen predictions based on a large number of simulations, team ratings, recent results and patch fit. How useful are those for picking group stage outcomes, and what should I watch out for?
How much should we trust simulations for TI15 group stage predictions?
Answers
The simulation count tells you how often the model was run, not how good its assumptions are. Running 100,000 trials can give a stable estimate of what the model expects, but it won't fix a bad rating or a questionable patch adjustment.
I'd look for how the model weights recent results against older ones, how it handles uncertainty, and whether the group format and tiebreak rules are represented correctly. Comparing models with different assumptions is more revealing than treating one set of percentages as the answer.
Reply to this answer
Patch fit is the tricky bit for me. Recent tournament results are useful, but a team that looked great on an earlier patch might not look the same on the current one. On the other hand, patch adjustments can just be a guess dressed up as a number.
I'd check which teams move the most when that assumption changes. Those are the picks where the model is telling you less than the percentages suggest.
Reply to this answer
Use the favorite as a starting point, not an automatic pick. Look at the alternatives too, especially when several teams have similar chances. A narrow favorite can still lose quite often, and a surprise result doesn't mean the forecast was necessarily bad.
Reply to this answer
I'd be cautious with a home advantage adjustment unless there's a clear reason for its size. Try the same predictions with and without it. If the ranking flips, that's a useful warning that the pick depends on a debatable assumption.
GBCspriteBudget+Napkin:
That makes sense. Comparing the scenarios seems more useful than just looking at the top pick.
Reply to this answer
Add your answer
Answers are reviewed before publication.
scrimFirstPick+Mug:
Exactly. A prediction can look very precise while still being sensitive to one subjective input.