🏅 Top traders

my first time being a top trader. It was an honor Isaac!

@PhatFree i made 3 mana :3

For posterity, I had bought this ad for 10 viewers.

@Austin This market suggests that the ad had approximately a 100% clickthrough rate within Manifold (e.g. 10 users saw the add, 10 users bought NO) which seems very unlikely. This also suggests that with 144 positions, it's about 14 times better to create a compelling market than it is to buy an add for a market, and given the unlikely clickthrough rate, it's probably more like 30 times better to create a compelling market than to create an ad.

Further, I think there are a couple more problems with the ad system -- nothing but the first default Manifold pet projects get shown as the first ads when I start looking at them, and you can now only redeem ads one time...so this disincentivizes people from looking at ads, because we have to scroll through the first 4 or 5 pet project ads to get to new ads to try to earn Manna (or see what else is out there or both). Why is ad order not randomized?

Secondly, the ads seem to be net zero...e.g. Manifold earns nothing or very little if they are redeemed? E.g. a buyer spends $M 10 and then a user earns $M 10... why even have ads if Manifold isn't going to earn any compute points back? As an ad buyer it seems like I'm buying a bunch of skips. As an ad consumer I'm being shown a bunch of repeat ads. As an ad buyer I would rather pay for clicks or full watches than skips, rather than pay the same price for both.

I'm sure you are already aware of this but thought it would be nice to get additional comments from a user.

@PatrickDelaney This market doesn't tell us that much. People were only given a suggestion to bet yes or no, but the market incentivizes not following those instructions. If you actually wanted to measure clickthrough rate, you could ask users to vote in a poll on the market to answer where they came from.

@jack Does manifold let you poll or do you need a third party poll?

@jack Can you elaborate? Very respectfully, what you are saying does not make sense. That being said I have definitely enjoyed your previous writing and believe in your analytical capability...it's just, your statement sounds contrary to what would be true, in that - folks had to actually spend to get the results we see above, whereas in a poll, it's just a click. Keep in mind I'm counting number of market participants in that 144 vs 10 figure, not weighted market participation. I removed bot purchases by the way.

@PatrickDelaney I think he is referencing the payout and bet/ante. In a poll there is no external incentive either direction, but in a market the is an internet point incentive which may out weight the non existent external incentive to be truthful.

This market asks "please" bet yes/no depending on how you arrived at the market. But if you read how the market resolves, you will notice that there is absolutely no profit motive to follow the instructions. The market functionally is just "resolves to yes if yes has more holders". So, the optimal bet on the market is independent of how you arrived at the market, and astute readers/traders will have noticed that. So I bet that some people followed the polite suggestion, but many people (myself included) just traded on the market to try to make a profit (for example, following the crowd, since this is a market where whichever side has the most votes wins).

You can easily create a poll like these examples:

Poll yes/no, resolve to percentage of yes: https://manifold.markets/jack/poll-should-it-be-possible-to-dispu https://manifold.markets/jack/standard-poll-rules

Poll a number, resolve to average/median: https://manifold.markets/jack/poll-is-lamda-sentient https://manifold.markets/jack/how-much-do-we-believe-sbf-committe

The author is also clear that they understand this issue - that's what the Schelling point part of the market description is about.

I also don't follow how you calculated the clickthrough rate. Even if we knew exactly how many people came from an ad, we don't know how many times the ad was shown, do we?

@jack Makes sense. We do know how many times the ad was shown by looking at Manifold Sitemap > Posts and then finding this Ad by scrolling down or doing a search. The spend on this ad was $M 100, it was viewed 10 times, and ran once. https://manifold.markets/post/ad-e9b5eca165b7 ... therefore, since there are 10 human YES views (at least those counted, the counting mechanism could be flawed), if we assume the number of times shown is equal to number of YES purchases, that's a 100% clickthrough/purchase rate. On the surface this seems like a fairly weak argument as it also assumes there was 10 views -> clicks -> YES bets, but if you look at the early YES bets, they were on the order of $M10, $M3, $M1, therefore I'm not sure if the Schelling model can be applied as it is a geometric model which assumes the same cost per participant to move across a space (e.g. one move, one unit) whereas in this situation, you had many participants on the YES side betting hundreds of units per Schelling-move, with NO side betting 10's or 1's of units per Schelling-move. So, we don't have to throw it out, we can see that it's probably fairly reliable, you might say ads are 5X or 10X less effective than just sharing markets on the platform. My contention is if you want a customer to actually buy a product, you have to show that product is superior to just their own manual labor, to incentivize them replace their own manual labor, e.g. the ratio should be in the opposite direction, which this above analysis shows that it's not (though it might not be precise).

Hah, I am still learning about just how little I know. It did not occur to me that Schelling point might be a proper noun relevant to the scenario, as opposed to some incidental slang that I was simply unfamiliar with. I am always thankful to have the opportunity to lean new stuff each day.

@PatrickDelaney I think you are talking about https://en.wikipedia.org/wiki/Schelling%27s_model_of_segregation, whereas Isaac and I were referring to https://en.wikipedia.org/wiki/Focal_point_(game_theory). The simple point is that if a bunch of people arrive from ads, they can all loosely coordinate to buy YES and make YES win, whereas if a bunch of people arrive from elsewhere, they can loosely coordinate to buy NO and make NO win. The problem is that this is a very loose correlation indeed and is completely different from actually measuring how many people arrive from where.

Manifold has their own analytics they can presumably use to actually answer the question of clickthrough rates.

Isaac is a very highly followed market creator, and his markets tend to get unusually high views compared to an average market creator. So I agree that probably most people saw this market from non-ad sources, but that doesn't mean that much...

On the ad view count, were there other ads that showed this market, or only one? I know Isaac ran multiple ads for some of his markets, don't know about this one.

Overall, I think there is plenty of useful discussion to have about the experimental ads feature, but I would strongly advise people to use better data - this market is imo nearly useless as a data source and its purpose is for fun.

@jack > On the ad view count, were there other ads that showed this market, or only one? Only one based upon the Posts page. @IsaacKing since you are a super highly followed Manifold user, can you do us a favor and put together another market like this, but instead of the instructions being, "buy," have a poll? I can pay you a Manalink for the cost of running an ad and starting the market if you want.

@ShitakiIntaki you and me both buddy. Honestly I had never even heard of Schelling point, I'm just a math geek so when I look at something like that I try to do the proof in my head before I say anything.

@PatrickDelaney Sure, happy to do that. Note that I'm actually not all that followed; I'm only 17th on the leaderboard, and this market has almost 3x as many traders as I have followers, so I'm not sure how much that matters.

Also by the way, an easier way to find the ad is to go directly to my profile and check the "posts" section under the "comments" tab. (Yes, the naming scheme is terrible.)

(Gonna wait for a few of my other ad tests to end first.)

@jack I attempted to embed a Google Form as a poll in two different ways. There is no way to verify that one account is answering one poll. I believe if you are actively recommending to users that they need to use polls, and that these types of markets are not accurate (which I kind of agree with you on that), then Manifold should have an embed poll feature, or at least allow Google forms as a trusted embed domain perhaps...? https://manifold.markets/PatrickDelaney/testing-embedding-google-form-as-po https://manifold.markets/PatrickDelaney/no-login-embedding-google-form-as-p

@PatrickDelaney I suggest to use a poll like this: https://manifold.markets/jack/poll-is-donating-mana-important-to#IO9faiIylLamfop2ivAg

@jack that is a novel way of conducting a poll. Maybe we should see if Manifold can give market creators the ability to pin a comment.

@ShitakiIntaki Actually there have been quite a few of them done! Agree pinning a comment would be very helpful here.

@PatrickDelaney I'm the one who built the ads feature, so hopefully I can clear up a few things.

• I think we should take a percentage cut of ads. (Perhaps something low like 10%). The intent was always to make ads a mana sink. I just haven't yet because first I wanted to see first see how people would use the feature.

• The ad creator doesn't pay anything when someone skips the ad, they only pay when someone has watched it and clicks the reward button

• Polls would be a great feature! A lot of people want them. I'm sure we'll get to it eventually.

Also about the our pet ads - are you seeing ads that you've already watched again? Some people have run into that issue and I'm looking into it.

@Sinclair i see ads I've already watched again every time (for example the one about the min language always shows up)

@Sinclair I too often see ads I've seen before. Happens pretty much every time I start watching ads.

Team YES may want to take out some more ads. There's a good ~M$5000 to be made by doing so.

@IsaacKing There’s 12 yes, and 100 no, so you’ll need at least 1000+ mana spent in ad views. Not sure what the conversion rate in those ad views will be either.