4 ms·

I think the case they are describing is where the marginal cost is highly nonlinear and the price delta between two reasonable values is so negligible that the

by esrauch 14y ago

I think the case they are describing is where the marginal cost is highly nonlinear and the price delta between two reasonable values is so negligible that the marginal cost isn't meaningful. A 2MB file does not cost twice as much to email send to someone as a 1MB file; you aren't going switch to a different internet connection or email provider because of your file is twice as big. The first 1 byte is very expensive and every subsequent byte has no observable marginal cost until 20 orders of magnitude later.

001sky 14y ago
Appreciate your comment. I think I was unclear earlier in my post. Was not trying to talk about the supply of > undifferentiated bits per-se. Those are trivial to scale, in minor orders. The micro cost is in the acquisition of <unique> bit sets. In your example, its not the cost to send or replicate the e-mail with a 2 mb attachment. Its the cost to acquire, verify, etc the contetents of the 2mb file have any value. [1] You can't create more <valuable> data per-se by making X copies of the same data (in the sense of it having value for marketing/analytics). that is just monetizing existing data. ie, The marginal cost to relicate a set, provided it was given to you for free...just assumes away a non-trivial part of the equation....getting the data. At scales of 100m to a 1Billion...is not trivial or costless. Lastly, if you only have one set of data (say 5m users of data), that is a finite supply. You could have 2 sets (10m users). Thats not the same as having 2 copies of 1 set (of 5 million). A customer might pay per user for a lead, but wont pay twice for two copies of the same info. now, if somebody shows up with 100s million, it might impact supply/demand (depending on comparabilit/uniquenss). But those differences cannot be assumed away at zero cost, imho. Hope this make more sense, was not trying to argue just for the sake of it. ________________ [1] That scales with data entry, etc (if nothing else) at the origin (ie, this is a FB user cost == per user). Even if its non-cash its ~$0.85c per 15 minutes of time for a western eurpoean ABC1, back of the envelope. And that scales linearly.
- esrauch 14y ago
  Thanks for the reply; I honestly am not sure about this situation in particular, I was mostly trying to explain how the statement about marginal costs can be true in any scenario. I didn't actually intend to say that the cost here is mainly bandwidth or X for any X. My point is more than it can be true that the marginal cost of anything can be so many orders of magnitude less than the non-marginal cost for certain ranges that it isn't worth considering. Data sets based on facebook profiles almost certainly come from people approving shady apps or someone set up a crawler in a way that is able to get a lot of information before being detected as a crawler. In either of those scenarios, the person who set it up effectively paid a flat upfront cost and ends up with X number of users, and there are no linear costs (no manual verification or paid data entry at any point). They cannot spend half as much time and get X/2 users or even maybe X/100 amount of information. In practice they could spend more time getting more users to give access to their random app, but the marginal cost function is just insanely nonlinear; its effectively 0 at some places and probably tends towards infinity an order of magnitude higher than that.

by esrauch 14y ago

001sky 14y ago

Appreciate your comment. I think I was unclear earlier in my post. Was not trying to talk about the supply of > undifferentiated bits per-se. Those are trivial to scale, in minor orders. The micro cost is in the acquisition of <unique> bit sets. In your example, its not the cost to send or replicate the e-mail with a 2 mb attachment. Its the cost to acquire, verify, etc the contetents of the 2mb file have any value. [1] You can't create more <valuable> data per-se by making X copies of the same data (in the sense of it having value for marketing/analytics). that is just monetizing existing data. ie, The marginal cost to relicate a set, provided it was given to you for free...just assumes away a non-trivial part of the equation....getting the data. At scales of 100m to a 1Billion...is not trivial or costless. Lastly, if you only have one set of data (say 5m users of data), that is a finite supply. You could have 2 sets (10m users). Thats not the same as having 2 copies of 1 set (of 5 million). A customer might pay per user for a lead, but wont pay twice for two copies of the same info. now, if somebody shows up with 100s million, it might impact supply/demand (depending on comparabilit/uniquenss). But those differences cannot be assumed away at zero cost, imho. Hope this make more sense, was not trying to argue just for the sake of it. ________________ [1] That scales with data entry, etc (if nothing else) at the origin (ie, this is a FB user cost == per user). Even if its non-cash its ~$0.85c per 15 minutes of time for a western eurpoean ABC1, back of the envelope. And that scales linearly.

esrauch 14y ago
Thanks for the reply; I honestly am not sure about this situation in particular, I was mostly trying to explain how the statement about marginal costs can be true in any scenario. I didn't actually intend to say that the cost here is mainly bandwidth or X for any X. My point is more than it can be true that the marginal cost of anything can be so many orders of magnitude less than the non-marginal cost for certain ranges that it isn't worth considering. Data sets based on facebook profiles almost certainly come from people approving shady apps or someone set up a crawler in a way that is able to get a lot of information before being detected as a crawler. In either of those scenarios, the person who set it up effectively paid a flat upfront cost and ends up with X number of users, and there are no linear costs (no manual verification or paid data entry at any point). They cannot spend half as much time and get X/2 users or even maybe X/100 amount of information. In practice they could spend more time getting more users to give access to their random app, but the marginal cost function is just insanely nonlinear; its effectively 0 at some places and probably tends towards infinity an order of magnitude higher than that.