Preventing the Need for "Overvalued Post" Flags

This is a repost, but the issue addressed is ongoing so I'd like to share it again.

Loss Aversion

A Major Source of Negativity on Steemit

How does it feel when after making a new post, you see the payout rise to $50 or $500? I know that feeling, and it's pretty great. But after hitting that high, the payout may drop. Perhaps it only drops a little, from $500 to $450, but weirdly that drop hurts more than the initial gain felt good. This post will explain loss aversion and propose a solution to minimize its impact on Steemit.

Rationally Irrational

Wikipedia describes Loss Aversion as follows:

In economics and decision theory, loss aversion refers to people's tendency to prefer avoiding losses to acquiring equivalent gains: it's better to not lose $5 than to find $5. Some studies have suggested that losses are twice as powerful, psychologically, as gains.[1] Loss aversion was first demonstrated by Amos Tversky and Daniel Kahneman.[2]

Twice as powerful! We really dislike losses. This phenomenon does not just apply to humans, it has also been seen among other primates.

Despite being irrational in many contexts, it is likely that loss aversion is evolutionarily beneficial. In our evolutionary past, there would have been an assymetry between losses and gains. It may also lead us to prefer our peers who come through on their promises, rather than overpromise and underreward. Those peers may have been more honest and reliable.

How Loss Aversion Affects Steemit

The basic idea of Steem post payouts is like a proposal going through a committee approval process. The author makes a proposal, submits it to their stakeholder peers, and depending on the final outcome, they get a payout in the end. However unlike most similar systems, Steem is totally transparent, and it needs to be. When the proposal gains support from an influential stakeholder (whale), the author sees that in the form of a rising 'potential payout'. Similarly when another stakeholder opposes the proposal, they see and feel that immediately as loss. The cost of transparency in the Steem system is that every trending author is subject to an emotional rollercoaster as their proposal visibly gains support (upvotes) and opposition (downvotes). There are also other elements which add to the volatility of this rollercoaster such as the votes occurring on other proposals and the inherent volatility of the Steem price.

Example of the Emotional Impact of Loss Aversion

Popular and prolific Steemian @karenmckersie was on the verge of quitting after receiving a downvote from one of our major stakeholders. The following is a rough graph of what the payout estimate may have looked like for the post in question. It only shows accumulated rshares squared over time, it's not the exact payout estimate.

The first thing that should stand out with this graph is how incredibly volatile it is, despite only accounting for the volatility in terms of votes for this post only, it doesn't account for the other two sources of volatility. And while the reward itself was a rollercoaster, it still doesn't reflect the emotional experience for the author @karenmckersie, when we account for the fact the experience of loss has twice the emotional impact of the experience of gain. Here is my approximation of the "emotional chart" by manually adjusting the impact of downvotes.

It's a wash! Due to loss aversion, as far as her experience went, she would have been better off not making the post.

Don't Blame The Stakeholder

After this incident, many people were upset and angry with the stakeholder who applied a large downvote. However as a stakeholder, they have the right to have their say in how the rewards are distributed. This is a fundamental premise of Steem as a stakeholder governance project. Many are also inclined to blame the facility to downvote at all, but an upvote only system is not viable. Stakeholders must be allowed to reign in over-allocation of rewards.

Yet still, downvotes are clearly a harsh punitive measure, with a disproportionate impact as shown above. In all areas of life, it is better to prevent harm than to try to fix it retroactively. Let's imagine a world where the downvote didn't happen, but we still had the same outcome from the reward pool.

What is this fantasy land? I would suggest that the above outcome is both better in practice, and viable to achieve on Steemit. It is better for the author, who has a much better emotional experience in the above scenario. It is better for the stakeholder who never had to come in with a downvote in the first place. And finally it is better for the other stakeholder in this story, who has been entirely left out of the discussion so far. The largest vote was the one which brought the post up to almost $54 in the first place. Do we even know that was their intention?

Upvotes are Not Good For Determining Consensus on Payout

Here I will repeat a suggestion from a previous post:

The purpose of the voting process is to determine how much should be paid out for a post. We want to reward people who contribute to the network appropriately, within the limits that the scarcity of Steem allow.

How Much Is This Post Worth?

Options: Good. Bad.

One of the main reasons we argue over payouts is because the system isn't actually working to achieve the purpose described above. Even ignoring the connotations of the flag symbol ("your post is abusive"), instead of feeding the system with information on what we feel a post should be paid, we can only say "I think this post is worth more" and "I think this post is worth less". There are big problems with this. Worth more than what? The voter may mean more than the amount at the time of the vote. If the payout approaches a certain high or low amount, the vote may no longer accurately reflect what the voter wants.

In practice it seems like an up vote and down vote mostly say, and are interpreted as "This post is good" and "This post is bad" or as above "This post is abusive". Yet those don't even answer the question which the vote is asking, and very frequently they are also not what the voter had in mind. Sometimes the amounts that are paid out may not reflect the intentions of any of the voters.

Solution: Payout Recommendations

A large stakeholder upvoted @karenmckersie's post and the payout estimate went close to $54. I cannot say that is the payout that the stakeholder considered appropriate, only they can. The only information given to the Steem Network was an upvote along with a weight, but that doesn't say very much. If stakeholders include a payout recommendation, we can prevent the need to reign payouts in, because in many cases they may not become excessive in the first place. The system has more information available to it to make better payout decisions earlier.

We already use Median in Steem Because it is a Robust Statistic

The price of Steem Backed Dollars is determined by the median of the price feed over the last week. This reduces the volatility of the peg, while preventing outliers from having adverse impact. These are the same objectives we should have with payout estimates! Volatility is the source of the pain of loss aversion. Excess is what we are trying to police with downvotes. While there will still always be scope for deliberate abuse, preventing the need to police unintentional excess may reduce most of the problem.

As suggested in my previous post, every vote should include a payout recommendation. For the sake of continuation, and simple user experience for new users, defaults can be applied for normal users such as $1,000,000 for a standard minnow upvote. Larger stakeholders should have access to an interface which doesn't just ask them if they like a post, but how much it should be paid.

The final payout can be determined as it is today, but with the additional caveat that the payout will not exceed the median payout recommendation. For the sake of calculating the median, we count every r-share applied, not 1-vote per user. This is straightforward enough to calculate and I don't see it having a serious impact on performance.