Even simple systems that should make sense can - and do - have lurking issues. This gets a tad bit geekisk, but is an excellent example of how we can fool ourselves - tip of the hat to Bjarne for the link.
I maintain it ca be real folly to claim you can accurately predict something like human behavior with samplings of social media. You can get bits and snatches, but I suspect expectations are way overblown. Human behavior isn't exactly an axiomatic system:)