Selection effects in online sharing: consequences for peer adoption

Sean J. Taylor, Eytan Bakshy, Sinan Aral

Abstract

In digital social networks, information can spread through active sharing (manual posting) or passive sharing (automatic notifications). This study utilizes a large-scale field experiment on Facebook to examine how these mechanisms influence peer adoption. We find that active sharing is significantly more efficient than passive sharing due to "selection effects"—specifically, users' ability to target peers most likely to be interested (dyad selection). While passive sharing generates higher total adoption volume through sheer scale, active sharing identifies higher-quality products and more relevant audiences.

Introduction

In traditional models of social contagion, exposure to a product is often treated as an automatic byproduct of adoption. If a neighbor buys a new car, you see it. However, in digital environments, exposure is a deliberate design choice. Platforms can implement passive sharing, where a user's actions are automatically broadcast to their feed (e.g., "User X just listened to a song"), or active sharing, where users must manually choose to recommend content to their network.

This paper addresses a critical gap in understanding digital virality: how does the mechanism of sharing change the likelihood of a peer adopting the product? The authors investigate the presence of selection effects, questioning whether active sharing is more effective because users only share high-quality products (product selection) or because they selectively broadcast to friends with specific interests (dyad selection).

Methodology

The researchers conducted a massive field experiment involving 1.2 million primary subjects on Facebook using Facebook Offers—digital coupons that users can claim and share.

Experimental Design

Users who claimed an offer were randomly assigned to one of two conditions:

  1. Active Sharing: Users were presented with a button to manually share the offer with their friends.
  2. Passive Sharing: The claim was automatically shared to the user's News Feed, which was the platform's default behavior for many actions at the time.
Examples of Facebook Offers and sharing interfaces used in the study, showing active and passive sharing formats.

Econometric Modeling

To disentangle the drivers of sharing effectiveness, the authors developed a joint discrete choice model using a Bayesian framework (Hamiltonian Monte Carlo). This allowed them to estimate correlations between a user's decision to share and their peers' subsequent decisions to adopt, effectively separating the inherent "quality" of the offer from the "affinity" between the sharer and the peer.

Data and Scope

The study utilized a proprietary Facebook dataset characterized by:

  • Scale: 1.2 million primary subjects and over 23,000 distinct offers.
  • Tracking: The data monitored both "impressions" (who viewed the share) and "adoptions" (who claimed the offer).
  • Filtering: To isolate the effect of a single social signal, the analysis focused on the 90% of peers who were exposed to an offer via only one friend's action.

Key Insights: The Efficiency vs. Volume Tradeoff

The study reveals that active sharing functions as a "high-quality, low-volume" channel, while passive sharing is "low-quality, high-volume."

The Power of Dyad Selection

The most significant finding is that the increased efficiency of active sharing is primarily driven by dyad selection. Users are remarkably adept at identifying which specific friends are likely to find a particular offer useful. While users do tend to share higher-quality products more frequently (product selection), this effect is secondary to the precision of user-to-user targeting.

The "Cost" of Sharing as a Filter

Active sharing requires effort, which acts as a natural filter. Users generally only incur the "cost" of sharing when they perceive high utility for their peers. Passive sharing removes this friction, leading to a high volume of "wasted" exposures to uninterested peers.

Key Results

  • Conversion Rates: Peers reached via active sharing were 10% more likely to adopt an offer compared to those reached via passive sharing.
  • Total Impact: Despite lower per-person efficiency, passive sharing was 4.3x more effective at generating total adoptions. This is because the sharing rate for passive conditions was 100%, whereas only ~23% of users chose to share actively.
  • Selection Drivers: Although product-level correlation was statistically higher ($\rho = 0.174$) than dyad-selection correlation ($\psi = 0.025$), the vast variance in dyad-level interests across a social network means that dyad selection actually drove the bulk of the observed efficiency benefit.

Discussion and Limitations

While the study provides strong evidence for selection effects, several factors warrant further exploration:

  • Algorithmic Interference: Facebook’s News Feed algorithm may already perform some "passive" selection by prioritizing stories for relevant friends, potentially masking the full extent of the active sharing advantage.
  • Adoption Costs: Claiming a coupon is a low-friction action. Selection effects might be even more pronounced for "high-cost" adoptions, such as purchasing expensive goods or adopting new health behaviors.
  • Network Effects: The study focused on single exposures. In reality, virality often stems from multiple exposures across a social cluster, a dynamic not fully captured here.

Ultimately, these findings suggest that platform designers face a fundamental choice: maximize reach through passive automation or maximize relevance through active user agency.

Summary updated

BibTeX
@inproceedings{taylor2013selection,
  author = {Sean J. Taylor and Eytan Bakshy and Sinan Aral},
  title = {{Selection effects in online sharing: consequences for peer adoption}},
  booktitle = {EC},
  year = {2013},
  doi = {10.1145/2492002.2482604},
}
Figure