It does do what it should in a certain sense - that means that DR has a high winrate because it shows you're in such a lead that either you can buy your own safely or can snatch one from the losing team. AC has a pretty high winrate because at the point you buy it in most games - you've already won. It pretty much works as it should and anything else would be misleading, though extra info on averaeg acquire time and similar would be helpful.
I think the stats become meaningless when you're looking at a large sample anyways because there are too many factors, no one's gonna bother modeling the scenarious if people pay for plus anyways
The refresher argument makes sense, 'correlation =/= causation'
but what you suggested would still be pretty blunt/simplistic because there's so much more to it other than xpm/gpm
the item you're buying is not the only thing that determines xpm/gpm/kda/whatever
You can make statements like on average a 15 min deso on TA contributed to a higher xpm/gpm/whatever than a 25 min deso, bur really you're only seeing the outcome
Maybe filtering out bot games, filtering out items in stomps (based on XP differential), filter out items bought in the last 5 minutes could start to make data look more realistic. Comparable win rates based on first core item built, such as win rate for BKB/Battefury first item mortred (similiar cost).
I'm not saying it's easy, but there is valuable information I think is hidden in the statistics available.
^you're cutting the sample, I said - accounting for more vatiables is better = more meaning
though, who knows really
the product is definitely unlikely to change unless there's a drop in plus sales ResidentSleeper
I see what you're asking for, but there is just too many variables involved when it comes to item statistics, and I think Dota buff plus does a good job at telling us what we need to know as it is. Some things are just too much work.
Sample sizes being processed are rather big nobody in Dotabuff would give a fuck
you only have timing data from fully parsed games - you only have the final items from non-plus matches.
Veuillez vous connecter pour publier un commentaire.
There is currently no way I can see to figure out which items are best for a hero on the statistics available. You can look at the win rate for item but according to that, Refresher 'leads' to more wins on SF than bkb does.
Often the bigger the item, the higher the winrate, because if you were able to complete that item it means you did pretty well. Is midas really that bad on NP? Or does an NP on the winning team normally sell the midas for a stronger 6 slot? We have very little data on the performance of cheaper items due to the fact they could be sold for a better item.
I don't think normalizing the win rate for items based on cost is enough. I think an algorithm which estimates performance (xp/min, gold/min, kda) after purchasing a chosen item verse it's common alternatives could provide some type of answer to: which item is the best option?