Modeling Diffusion of Many Innovations via Multilevel Diffusion Curves: With an Application to Payola in Pop Music Radio (M1138V1)

We introduce a new statistical method – multilevel diffusion curves – to model how multiple innovations spread through an industry. Specifically, we analyze when radio stations begin broadcasting 534 pop singles. Ordinarily radio stations imitate one another, an endogenous process producing a characteristic “s-curve.” However, payola can dwarf this process and produce a characteristic negative exponential curve, controlling for the song artist's number of successful songs in the past year. Therefore the shape of a song’s cumulative adoption function indicates whether its rise involved corruption. We validate this heuristic against a panel of songs with a documented history of payola and a comparable set of songs with no such allegations. Compared to earlier methods, multilevel diffusion curves allow testing of more types of hypotheses, model a greater range of data, and are statistically more efficient and precise.