An Ads Experiment

Created by
Gabrielle

I was interested in testing the Manifold ads feature to see what interest it could earn for a market. Moreover, I was interested in discovering more about the two markets I had posted. They were both topics that Nate Silver at FiveThirtyEight considered important questions, and they both have large potential impacts on the United States. However, they are both relatively long term markets (one not resolving until the 2024 election, one not resolving until 2030) and had not received very much interaction.

The two markets are:

Both markets had been quite inactive, with the Senate market not having activity in the previous ten days and the Supreme Court market only having one non-bot trader in the previous twenty five days.

Before placing the ads, I added M$300 of subsidies to each and let the subsidies fully settle, to hopefully help add incentive to bet in the market. The Supreme Court market had 11 traders before adding subsidies, and gained 3 more traders after I added the subsidies but before I placed the ads. The Senate market had 13 traders before adding subsidies and gained no more between that and placing the ads.

I did not post the markets in any context other than the ads, so hopefully the activity was entirely from the ads and the subsidies.

On 2023-04-02, I placed ads for M$500 on each market. For this to be directly profitable for me, not even including the subsidies, both markets needed to get 50 additional traders. For it to be worth it from a knowledge perspective, I wanted to see at least 10 additional traders.

The ads were simply an embed of the market, without any additional text advertising the markets or stating why they were important or potentially profitable. While those would likely be good, I wanted to test how people interacted with very basic ads. For context, you can see the (now used up) ads at https://manifold.markets/post/ad-025e55df1bab and https://manifold.markets/post/ad-a64a5b9a4ee3.

I then waited until the ads had been completely viewed. I precomitted to publishing the results, regardless of the outcomes.

As of 2023-04-11, all of the ad views had been watched. For the Supreme Court market, that was 100 watches and 75 skips. For the Senate market, that was 100 watches and 76 skips.

The Supreme Court market had 26 total traders and the Senate market had 18 total traders, meaning 12 new traders and 5 new traders respectively compared to before I placed the ads. From my perspective, it was barely worthwhile for the Supreme Court market, getting slightly more than 10 additional traders, and it was not worthwhile for the Senate market.

Based on this, I would not use the ad system again in this way. Most importantly, I think that I need to add some call to action, saying either that it was particularly profitable (extra subsidies) or particularly important. It also seems like the ad system just doesn't encourage people to interact. Only five out of a hundred watches on the Senate market actually interacted, which is an abysmally low rate. An M$100 bounty would have been equally as efficient as the ads, and it would feel like a much fairer system. All of the users that watched and didn't interact or skipped wouldn't have had their time wasted, the users that watched and interacted would have gotten a much better reward, and it would have cost me the same.

Sinclair avatar

Thanks for running this test and writing it all up!

I want to see if there would be more betting if the viewer can bet from the ad directly without leaving. (Though this task I've been putting off).

I think we still haven't cracked the "pay for information" use case, and getting a lot of bettors is still too dependent 'winning at social media' - having engaging writing / building an audience / sharing it elsewhere.