You are viewing a single comment's thread from:
RE: Steemit Power Relationships between Groups
Good analysis, well done.
Couple of comments about the data. The vests in Accounts table doesn't account for delegations. There are some high SP users who delegate away their power while other minnows might be leasing a lot of SP that would push them up classes. Also, some users vote a lot more at 100% and have a degraded Vote Power which will skew your results too. Then there are the bots too like randowhale and minnowbooster that show up everywhere and would ideally be excluded from the data.
Overall though, really good analytics.
Yes, you are right about delegations. I was looking only at how the groups vote each other based on accounts table. So, even if power is delegated for a user it expires, the user is just a boosted up entity, it's relationship with others might not really change. So delegations do not really matter here.
Bots - I have already said that I don't endorse a manual way of excluding them as it may result in showing bad data.
In the week from Sept 17 to 24 the major bots have the following votes :-
minnowsupport - 5638
minnowbooster - 4120
randowhale - 2735
booster - 1560
bellyrub - 1321
That's a total of 15374 votes and they would almost entirely manifest in your statistic of whales voting for minnows or plankton at low percentages.
I get it and I know it. I am trying to present the data as a whole, not doing sampling. So if you have a way to identify the bots programmatically then please do share as it will be useful for a lot of us.
They are different types of bots with different voting behaviours so it is tricky to identify them programatically. I personally have been been building a register of them with some vital stats so I would use a manual blacklist to exclude them from the data query myself, but if I come up with a better idea I'll let you know.
Keep up the good work.
Thanks for the heads-up. One of the major pitfalls in any data analysis is the manual exclusion part. Once we start doing that cleanup we tend to get more biased on the data to be excluded and the final result may get screwed up. So if we cannot exclude programmatically, it is better to leave it as is. Let me know if you get any breakthrough in finding the bots. Eagerly Waiting!