Our intuition may tell us that making a sample population more random also makes it more representative, insofar as selection biases are reduced. The mistake lies in assuming that greater randomness increases the potential quality of the sample when it could just as well introduce subtler biases of its own.

Well-designed experiments control for as many variables as possible. Your friends, their friends, and their friends, etc. have far more in common with each other than with random strangers. Even at two degrees removed, you are far less likely to encounter wild divergence than with the erratic spread of a truly random sample. That relatively consistency, not unlike truly controlled variables, makes it all the easier to tease out the data that matter.

Seek out such extended groups for your own controlled experiments, especially for market research. Unless you’re selling a universal commodity like toothpaste, then asking any random person on the street will muddy your data far more than it helps. Besides, an audience of interested friends and friends of friends may prove rather more representative of your target market. Where better to start with research than with people who already care? If they disprove your hypothesis, then what are the chances that the population at large would do otherwise? The broader studies can wait until you have completed your preliminary trials.