Nobody's friends constitute a perfectly random sample, but extended social networks are really good approximations of a random sample for questions like this.
Two interesting points: at least 3 people in the 75 had snowmobiles (the family of a college friend who lives in Alaska) -- but also, 0% ownership rate for snowmobiles is not a bad approximation of worldwide snowmobile ownership rate, depending on the application you're using the data for. There are a ton of applications of survey data where assuming a 0% worldwide ownership rate of snowmobiles would be a super useful and perfectly reasonable approximation. Then there are others, like say, running a snowmobile advertising campaign in Minnesota, where that 0% estimate would be terrible.
In fact, I would go a little further and say that for reasonably diverse sets of social connections, it would not be feasible to observe the effect you are saying (either a lot of people satisfy property X or none at all do) unless the property in question is extremely rare in general. Even thinking about connection data sets that are nearly perfectly clustered, like telephone connections in Belgium (which reveal structure based on whether the person speaks primarily French or primarily Dutch), it would be very, very uncommon to have zero opposite-language speakers in a 75-member sample of your network.
I definitely believe that a random batch of 75 connections on e.g. Facebook, for many people, would be perfectly adequate for getting a coarse, approximate sense of something like general interest in a mainstream tech product.
Will it have pitfalls -- of course, but absolutely no one is claiming otherwise. In fact, I am not even claiming the result is accurate. I am asking what data supports the hype over Echo. It's very strange to me that people are more interested in criticizing some openly-acknowledged-to-be-highly-inaccurate-yet-still-suggestive-of-anomaly ad hoc survey simply mentioned in a comment, rather than to actually ask what kind of data might empirically refute the sentiment of that survey.
Two interesting points: at least 3 people in the 75 had snowmobiles (the family of a college friend who lives in Alaska) -- but also, 0% ownership rate for snowmobiles is not a bad approximation of worldwide snowmobile ownership rate, depending on the application you're using the data for. There are a ton of applications of survey data where assuming a 0% worldwide ownership rate of snowmobiles would be a super useful and perfectly reasonable approximation. Then there are others, like say, running a snowmobile advertising campaign in Minnesota, where that 0% estimate would be terrible.
In fact, I would go a little further and say that for reasonably diverse sets of social connections, it would not be feasible to observe the effect you are saying (either a lot of people satisfy property X or none at all do) unless the property in question is extremely rare in general. Even thinking about connection data sets that are nearly perfectly clustered, like telephone connections in Belgium (which reveal structure based on whether the person speaks primarily French or primarily Dutch), it would be very, very uncommon to have zero opposite-language speakers in a 75-member sample of your network.
I definitely believe that a random batch of 75 connections on e.g. Facebook, for many people, would be perfectly adequate for getting a coarse, approximate sense of something like general interest in a mainstream tech product.
Will it have pitfalls -- of course, but absolutely no one is claiming otherwise. In fact, I am not even claiming the result is accurate. I am asking what data supports the hype over Echo. It's very strange to me that people are more interested in criticizing some openly-acknowledged-to-be-highly-inaccurate-yet-still-suggestive-of-anomaly ad hoc survey simply mentioned in a comment, rather than to actually ask what kind of data might empirically refute the sentiment of that survey.