With population N, a frienship whose members have f and g friends might be either picked with probability 1/fN weighing f or probability 1/gN weighing g, so on average all friendhips weigh equally 2/N.

]]>Darla can do better the more work she is willing to do per questionnaire. Set a probability p < 1. Pick a random person. If that person has k friends and kp 1, generalize this to picking either

ceil(kp) or floor(kp) of their friendships. Smaller p reduces the favoring of friendships sharing a common popular person. If p is small enough, Darla will likely

check every person and not save any work.

The question makes implicit assumptions about the data structure representing the friendship graph, since it is computationally cheap to pick a random vertex but expensive to pick a random edge.

]]>