The Illusion of Change: Correcting for Bias when Inferring Changes in Sparse, Societal-Scale Data