Let's hypothesize that hyperbolic discounting is a good model for how market research people will value focus-group information: they will pay twice as much if it takes half as much time to perform a focus group and digest the results.
If it currently takes two weeks to do a focus group, and it takes 30 seconds on Facebook, then not only does it become reasonable to ask questions that you need only five minutes to digest the results of --- but you'll be willing to pay 4000 times as much, say, per question. Or, more likely, ask questions that are 4000 times less valuable, and pay the same amount.
That's not plausible --- there's probably some other limiting factor. But it could easily be bigger than Gartner.
Doing a poll on this thing could have the same relationship to traditional polling and focus groups that Google has to library research. I do maybe 100 Google searches a day, maybe 100× as many as I ever did card-catalog searches in a library. Can you imagine marketing guys and politicians doing 100 opinion polls a day?
In my own work, we do a dozen or more experiments every day, and our typical value is p=0.0000001% -- one in a billion or so. Sometimes we get a p of one in a trillion. And even then we agonize over whether we should believe the results of the experiment.
Don't you wish you could do a dozen or more experiments on human subjects each day, with a p of one in a billion?
Comments
Let's hypothesize that hyperbolic discounting is a good model for how market research people will value focus-group information: they will pay twice as much if it takes half as much time to perform a focus group and digest the results.
If it currently takes two weeks to do a focus group, and it takes 30 seconds on Facebook, then not only does it become reasonable to ask questions that you need only five minutes to digest the results of --- but you'll be willing to pay 4000 times as much, say, per question. Or, more likely, ask questions that are 4000 times less valuable, and pay the same amount.
That's not plausible --- there's probably some other limiting factor. But it could easily be bigger than Gartner.
Doing a poll on this thing could have the same relationship to traditional polling and focus groups that Google has to library research. I do maybe 100 Google searches a day, maybe 100× as many as I ever did card-catalog searches in a library. Can you imagine marketing guys and politicians doing 100 opinion polls a day?
Relevant is this snippet from http://norvig.com/experiment-design.html:
In my own work, we do a dozen or more experiments every day, and our typical value is p=0.0000001% -- one in a billion or so. Sometimes we get a p of one in a trillion. And even then we agonize over whether we should believe the results of the experiment.
Don't you wish you could do a dozen or more experiments on human subjects each day, with a p of one in a billion?