Abstract:
|
As an increasing amount of daily activity---ranging from what we purchase to who we talk---shifts to online platforms, it is only natural to ask how those platforms impact our behavior. Take, for instance, online recommendation systems: How much activity do recommendations actually cause over and above what would have happened in their absence? Ideally we would estimate their impact via randomized experiments, but in practice such manipulations may be costly or impossible. In this paper we develop an alternative method for estimating causal effects from purely observational data, through an instrumental variable approach to analyzing products that experience large and sudden shocks in traffic. We apply our method to anonymized browsing logs for 2.1 million users on Amazon.com over 9 months and analyze over 4,000 unique products that experience such shocks. We find that while recommendations account for a large fraction of traffic among these products, at least 75% of this activity would likely occur in their absence. Given reasonable assumptions, this method can be applied to other platforms and settings, such as online advertisements.
|