Influencers are individuals who create content to build followings on social media and influence those followings with ideas relevant to their niche (e.g. makeup tips, pet care, interior design, etc.). “Influencer marketing” refers to advertisers sponsoring those influencers to create content promoting the advertisers’ products. Despite the popularity of this practice, there is limited evidence on its effectiveness: because advertisers cannot access user data from social media, frequently the only data available to the advertiser are the post itself (any data from the image or video), associated public metadata (e.g. the time it was posted), and aggregated performance information (e.g. the content’s total likes/shares).
We propose a model to measure the effectiveness of influencer marketing on sales of the promoted items when only the timing of the promotion is known. The model is based on the causal structural time series regression with a potentially large number of time-varying covariates. It leverages Bayesian model averaging when the dimension of the covariates is high. We conclude with a discussion of identifiability of the causal effect.