3 ms·

Well its pretty easy to tell if its because of the advertising. Just turn it off for a while and see if it affects revenue.

zwaps 7y ago
Really not that easy, except if you can everything the same, aka go back in time. Otherwise, causal inference is sadly not easy.
- DoctorOetker 7y ago
  thats why seller should modulate buying ads with a random stream of bits, such that the impulse response (of linearized approximation in tangent space for sales / advertising dollar) can be estimated. of course if the advertisement platform is simultaneously a dominant search engine, they can simply choose to directly modulate your discoverability through the search engine, in which case it is closer to extortion as others point out. as a potential buyer, I'd much rather have (economical) censorship free decentralized platform where every manufacturer / seller can advertise products / parts with relevant specifications / metrics. Think DigiKey, but without a central arbiter. Obviously a lot of products that are actually superior are superior because they have a better understanding of what people need than what conventional metrics show (say not just the cheapest capacitor, but perhaps the cheapest capacitance divided by electric series resistance or whatever, for a certain niche use). This would result in placing the burden of comparing products among the manufacturers, who will also make up original but useless metric formulas with the sole purpouse of making their product rank highest. So instead of having to browse products, people are browsing metrics, and it's up to both manufacturers and end users to communicate the utility of each end metric. Something I don't like about the parts selectors like DigiKey et al, is that I can filter and rank according to proposed properties / columns, but I can't enter a formula in terms of properties, to show me the highest ranking. I always end up having to download and scrape the full list of parts so I can perform this ranking in a spreadsheet... If their search fields allowed formulas, and if people were logged in, the manufacturers / resellers could contact you to find out what made you formulate this specific metric or loss function, which gives feedback to the manufacturers what buyers are interested in. It would also give the sellers opportunity to correct the buyer that in his or her design the buyers' metric is faulty, and point out why...
  - retsibsi 7y ago
    > seller should modulate buying ads with a random stream of bits, such that the impulse response (of linearized approximation in tangent space for sales / advertising dollar) can be estimated Could you unpack this a bit? (I'm not making a point or anything, just confused and curious and not sure that googling would help me understand.)
buboard 7y ago
That's not what the article is about - but try scaling it back to 50% and see if it results in 50% drop. The graphs show that the relationship is probably very nonlinear

zwaps 7y ago

Really not that easy, except if you can everything the same, aka go back in time. Otherwise, causal inference is sadly not easy.

DoctorOetker 7y ago
thats why seller should modulate buying ads with a random stream of bits, such that the impulse response (of linearized approximation in tangent space for sales / advertising dollar) can be estimated. of course if the advertisement platform is simultaneously a dominant search engine, they can simply choose to directly modulate your discoverability through the search engine, in which case it is closer to extortion as others point out. as a potential buyer, I'd much rather have (economical) censorship free decentralized platform where every manufacturer / seller can advertise products / parts with relevant specifications / metrics. Think DigiKey, but without a central arbiter. Obviously a lot of products that are actually superior are superior because they have a better understanding of what people need than what conventional metrics show (say not just the cheapest capacitor, but perhaps the cheapest capacitance divided by electric series resistance or whatever, for a certain niche use). This would result in placing the burden of comparing products among the manufacturers, who will also make up original but useless metric formulas with the sole purpouse of making their product rank highest. So instead of having to browse products, people are browsing metrics, and it's up to both manufacturers and end users to communicate the utility of each end metric. Something I don't like about the parts selectors like DigiKey et al, is that I can filter and rank according to proposed properties / columns, but I can't enter a formula in terms of properties, to show me the highest ranking. I always end up having to download and scrape the full list of parts so I can perform this ranking in a spreadsheet... If their search fields allowed formulas, and if people were logged in, the manufacturers / resellers could contact you to find out what made you formulate this specific metric or loss function, which gives feedback to the manufacturers what buyers are interested in. It would also give the sellers opportunity to correct the buyer that in his or her design the buyers' metric is faulty, and point out why...
- retsibsi 7y ago
  > seller should modulate buying ads with a random stream of bits, such that the impulse response (of linearized approximation in tangent space for sales / advertising dollar) can be estimated Could you unpack this a bit? (I'm not making a point or anything, just confused and curious and not sure that googling would help me understand.)

buboard 7y ago

That's not what the article is about - but try scaling it back to 50% and see if it results in 50% drop. The graphs show that the relationship is probably very nonlinear