Secondary Effects

in #hf215 years ago

I'm compiling a series of thoughts about the economic proposal a lot of people have been talking about here. I don't expect to convince anyone of anything, but I hope to at least motivate why these choices are being made. For those of you sick of hearing about it... I'm sure... This will be the last I post about it. Probably.

As a summary, the EIP has the following elements in it:

  • Convergent Linear rewards curve. (Form discussed before, and I've talked about it in my previous posts)
  • 50/50 author curator split (from 75 author / 25 curator)
  • 25% subsidized downvotes

Let's go through a few thought experiments and toy examples.

Downvote Rule

A small thought experiment in a world of only farmers.

Let's say everyone was just self voting themselves. You got 100 people, posting 10 posts a day for 7 days, and upvoting them with full power.

Let's now look at if the new rules can break this equilibrium. Initially was going to look at everything, but instead let's just focus on the 25% downvote pool.

The dynamics are immediately interesting, because it's clear that to maximize they need to downvote. But who? They get more also by ganging up on others.

So let's assume that 20 of those farmers are downvoted out, meaning 80 have used all free downvotes on the other 20. Now what?

Those 20 will now see that upvoting their own posts is a waste, and will stop doing so, looking for curation opportunities and also exercising their own downvotes.

From here, it's not clear to me what would happen as votes shuffle back and forth. Even in a world where there is no content. Those kind of dynamics make things much more exciting and having something akin to content or engagement quality can serve to highlight that behavior above the rest. I mean there's nothing that makes anyone choose content quality as a guide but certainly the better networked you are, the better you will do. And that is arguably an important metric for how valuable you can be to an information network.

There's a lot missing here also. Starting distributions can definitely affect things. You don't want to be in a situation where the top user(s), with 25% has enough to lock out everyone else. I think 25% is chosen with that in mind.

Adding Convergent Linear

If we add convergent linear to the mix, there's a slight incentive to consolidate votes. In a cooperative world, that would be where they all vote on a single post and split the spoils fairly. But we know this cannot happen on a global scale. On top of that, the curation curve makes it so that earlier voters get more.

As mentioned before, the intent of convergent linear is to make it unprofitable to create thousands of posts with small votes in an attempt to hide from everyone. You can still do it, but the curve will penalize it. Also, from the way things are looking, it's not going to be highly nonlinear at the bottom. It's going to start near the middle and end linear. Meaning that it doesn't completely kill small posts. But it will still penalize low end farming.

When people study the curve, they need to understand a very crucial point:

  • The payouts are highly dependent on the distribution of posts.

As an illustration, consider a simple case where in linear, two people have 1000 1 rshare posts. Together in linear, 2000 rshares of posts, and if it were distributing 20$, that's 1 cent per post for 10$ each.

Let's use the curve (r^2 + 2r)/(r + 4). Assuming they don't change, each post has a score of 3/5, for a total score of 1200. With 600 each, and 20$ to distribute, that's again 1 cent per post for 10$ each.

But let's say one of them reduces to 500 posts with 2 rshares each. Now one person continues to have a total score of 600, while the other has 500 * 8/6 = 2000/3 or around 666. That means with 20$ to distribute, one gets $9.48 while the other gets $10.52.

At an extreme, let that other person go to 10 posts with 100 rshares each... This becomes 10 * 10200/104 or around 981. Now the distribution looks like one getting $7.59 and the other $12.41.

In any case it's clear, even with a mild curve, it can really shift incentives away from low level post farming.

Why do we care about that, you might ask? It makes it much easier to downvote with the freebie downvotes, which as mentioned above, everyone has incentive to use. They will have to balance hiding in the shadows to exposure to downvotes.

Of course, this is a simplistic example and a lot of things depends on the curve parameters and distribution of other post values.

But it's clear that it becomes that much more important to exercise downvotes to make sure the ones that deserve rewards get the rewards.

The 50% Thing

I don't have much to say about 50/50. But I will remind you that as with above, the payouts are dependent on the distribution of votes on posts. Even if at the moment of flipping the switch it cuts by 1/3, the equilibrium will shift over time as the incentives kick in and people adjust. And there's certainly a lot to think about in relative terms. Curation getting a boost, despite the skepticism about it not doing anything, will incentivize a broader distribution. People will be encouraged not only to post and leave, but also to go find and vote other content based on the potential gain. I don't think everyone necessarily benefits relative to before. But I think the result will be more fair.


In case you couldn't tell, I am pro EIP. Though you might also note that in all cases, downvotes are the key. I would be okay with just downvotes too. But I like the thought behind the other initiatives. Anyway, let's make incentives work better for the platform.

Posted using Partiko Android

Sort:  

Very interesting thoughts! thanks.
As you, I am pro EIP.

Very well analyzed

The examples of the convergent curve should be simple enough for anyone to follow and is very helpful.

I'll just add that with 50% curation and 25% downvotes, for every 'abuser' at any given time we'd need 2 good actors of equal stake to bring their self voted rewards down to average (mean) curation. That is to very roughly say if two others of similar stake are attacking your vote farms, you might as well just go curate honestly.

The 50% curation number was chosen because we believe 2 good actors for every 1 abuser is hopefully attainable in the new equilibrium, and therefore is a sensible number to try on our initial attempt at economic reform.

If we lower this curation value, not only would we require more good actors per abuser, it'll also increase the pay gap between curation and abusing in favor of the latter, likely resulting in fewer good actors per abuser.

Basically the higher curation is, the higher the probability of successfully changing the status quo voting behavior to something that's broadly honest, but we also want to leave enough to reward and incentivize authors too.

One fork at a time. :)

Plus, wake me up when we start promotion and marketing and quit catering to our internal holders.

:) I did read and enjoy your thoughts though.

Fair enough :)

I'd help with any promotional efforts for sure!

lol,

thanks. i'm trying to see how well i do in ROI as a 'leisure' reader and voter.

michealb 295.9 min 1.028 SBD 0.000 SBD 0.376 SP 14.4 %
https://beempy.com/hf21/@eonwarped/secondary-effects-9r35cbuc

currently the autovoters / voters under the 11 minute mark are ones kicking the goals. :)

thoughts?

For things that are paid out, I have this:
https://helpienaut.github.io/steem/curation_analysis.html?account=michealb&votedUser=
which you can use as a guide. It sees the affect of if you changed your vote timing (while preserving the vote value).

But yeah, it definitely depends, something you know is popular, you have to weight the early penalty loss against what you think you'll gain from being early.

thanks

Has anyone checked to see what percentage the bottom 75% of steem earners extract from the reward pool? How many accounts from top down hold 25% of the Steem Power? is it 100? 500? more?

I myself am not for the downvote pool.

The top 12 accounts (minus steemit) that have activity within last two years hold 35% from the query I pulled for steem alliance voting. I did forget to remove misterdelegation and steem though, so that is going to be lower. But... Still gives you a pretty good idea.

AWQlawzffUQMAAAAAElFTkSuQmCC.png

If you look at account counts and stake %. So two more in that top bucket needs to be factored out but relative number wise that's how it looks.

The way I see it, everything is still proportional to stake and we will see quite a different distribution of posts. I do believe that 25% is not enough for the top few to lock down the rest of the system. And keep in mind the top accounts don't necessarily see eye to eye, so that helps too.

Thank you for the information.

This is the greatest explanation of secondary effects I have seen yet! Thanks! Hope more people stumble across this.

Thanks! Though I am realizing that I'm not even sure I mentioned what I wanted to. About ripple effects. It's kind of there. But anyway, glad it helped!

Posted using Partiko Android

To listen to the audio version of this article click on the play image.

Brought to you by @tts. If you find it useful please consider upvoting this reply.

Thank you so much for participating in the Partiko Delegation Plan Round 1! We really appreciate your support! As part of the delegation benefits, we just gave you a 3.00% upvote! Together, let’s change the world!

Hi @eonwarped!

Your post was upvoted by @steem-ua, new Steem dApp, using UserAuthority for algorithmic post curation!
Your UA account score is currently 4.945 which ranks you at #1173 across all Steem accounts.
Your rank has improved 4 places in the last three days (old rank 1177).

In our last Algorithmic Curation Round, consisting of 155 contributions, your post is ranked at #6.

Evaluation of your UA score:
  • Some people are already following you, keep going!
  • The readers appreciate your great work!
  • Great user engagement! You rock!

Feel free to join our @steem-ua Discord server

Coin Marketplace

STEEM 0.36
TRX 0.12
JST 0.039
BTC 70112.96
ETH 3549.99
USDT 1.00
SBD 4.71