Source-linked AI summary
Social Influence in Social Advertising: Evidence from Field Experiments
Eytan Bakshy, Dean Eckles, Rong Yan, Itamar Rosenn
TL;DR
Social advertising raises a causal-identification question because peer-based targeting can reflect correlated characteristics as well as social influence. Using two large Facebook field experiments, the paper finds that social cues increase clicks and connections, with stronger effects for stronger ties, while noting design limitations in Experiment 1.
Problem
Social advertising research had largely been unable to identify how much consumer response was caused by peer social signals rather than correlated peer characteristics.
Method
Two large Facebook field experiments randomized the number or presence of social cues and measured clicks, page connections, and tie-strength differences.
Results
Social cues increase clicks and connections with advertised entities, and cue effects are larger for strong than weak ties.
Takeaways & Limitations
Social advertising systems can benefit from incorporating tie-strength measures when selecting ads and social cues.
Takeaways & Limitations
Experiment 1 lacks a no-cue baseline and confounds more peers with a slightly taller, visually changed ad unit.
Abstract
from arXiv · showhide
Social advertising uses information about consumers' peers, including peer affiliations with a brand, product, organization, etc., to target ads and contextualize their display. This approach can increase ad efficacy for two main reasons: peers' affiliations reflect unobserved consumer characteristics, which are correlated along the social network; and the inclusion of social cues (i.e., peers' association with a brand) alongside ads affect responses via social influence processes. For these reasons, responses may be increased when multiple social signals are presented with ads, and when ads are affiliated with peers who are strong, rather than weak, ties. We conduct two very large field experiments that identify the effect of social cues on consumer responses to ads, measured in terms of ad clicks and the formation of connections with the advertised entity. In the first experiment, we randomize the number of social cues present in word-of-mouth advertising, and measure how responses increase as a function of the number of cues. The second experiment examines the effect of augmenting traditional ad units with a minimal social cue (i.e., displaying a peer's affiliation below an ad in light grey text). On average, this cue causes significant increases in ad performance. Using a measurement of tie strength based on the total amount of communication between subjects and their peers, we show that these influence effects are greatest for strong ties. Our work has implications for ad optimization, user interface design, and central questions in social science research.
1. INTRODUCTION
Social advertising combines peer-based targeting with social cues that may influence consumer responses. The paper uses field experiments to identify social-cue effects, finding responses increase with additional peers, minimal cues, and stronger ties.
- Social cues create a channel for peer influence, extending questions from word-of-mouth research to online advertising and social science.
- Social advertising uses peer information both to target likely adopters and to provide personalized social signals alongside ads.
- Field experiments randomize social-cue exposure to identify effects that prior social-advertising work largely could not isolate from social targeting.
- Showing more affiliated peers can increase positive consumer responses in word-of-mouth-type advertising.
- Minimal single-peer cues substantially increase clicks and connections, with larger influence effects for strong than weak ties.
2. CAUSAL RELATIONSHIPS IN SOCIAL ADVERTISING
Consumer responses to advertising are shaped by individual characteristics, correlated peer characteristics, and potentially peer influence. Because homophily and prior interactions confound observational associations, randomized social-cue assignments are used to identify causal effects.
- Advertisement responses depend on observed and unobserved user characteristics, while the causal model abstracts from ad-creative variables for simplicity.
- Homophily clusters consumer characteristics across networks, making peer behavior predictive of responses even without peer influence.
- Peer behavior can also cause consumer behavior when observed or inferred responses provide social information through advertisements.
- Unobserved characteristics can keep peer responses dependent even after controlling for observed characteristics, complicating causal interpretation.
- Randomly assigning user–ad pairs to different numbers of social cues enables comparisons of response rates across cue exposures.
3. RELATED WORK
Related work situates social advertising within research on information diffusion and experimentally studied social signals. Prior studies distinguish social targeting from cues and manipulate mechanisms such as viral marketing.
- Observational diffusion research has examined links, groups, product recommendations, and user-contributed content across online networks.
- Prior field experiments studied social signals in online advertising and related settings, including social targeting, cues, and viral marketing mechanisms.
4. SETTING AND DATA ANALYSIS PROCEDURES
The experiments use Facebook News Feed advertising and study clicks and page likes as consumer responses. Analyses account for crossed dependence among users and ads, while Experiment 1 randomizes the number of displayed peers.
- The two large Facebook field experiments were conducted during a short period in 2011.
- Facebook users’ friends constitute their peers, and page likes create visible connections and connection stories within the network.
- Socially targeted ad units can display eligible peers who like the advertised page, linking social targeting with social cues.
- Clicks open linked content, while liking the advertised page creates a new user–page connection; both responses are analyzed.
- Because impressions are dependent across users and ads, the analysis uses a multiway bootstrap that resamples both dimensions rather than assuming IID observations.
5. EXPERIMENT 1: INFLUENCE OF MULTIPLE PEERS
Experiment 1 randomly assigned user–ad pairs to show one to three affiliated peers, estimating how additional social cues affect clicks and likes. Within-panel effects increased responses, while between-panel differences remain confounded by user, page, and selection differences.
- Sampling and assignment procedure: User–ad pairs were randomly assigned equal probabilities over feasible cue counts, with 101,633,907 distinct user–ad pairs analyzed.Assignments were limited by each pair’s one to three affiliated peers; all impressions for a pair received the same peer number and order.
- Average cue–response function: 10.3% higher click rates and 10.5% higher like rates resulted from displaying a second peer when two affiliated peers were available.The 95% confidence intervals were [8.7%, 11.9%] for clicks and [8.4%, 12.4%] for likes.
- Average cue–response function: 8.0% higher click rates and 8.9% higher like rates resulted when three affiliated peers were available.These increases were slightly weaker, with confidence intervals of [5.7%, 10.3%] and [6.0%, 12.1%], respectively.
- Average cue–response function: The increase from one to two cues did not significantly differ from the increase from two to three cues, consistent with simple contagion.The reported comparisons had p > 0.1 for clicks and p > 0.3 for likes, though average effects cannot entirely rule out complex contagion.
- Average cue–response function: Between-panel response differences cannot be interpreted straightforwardly because homophily, page popularity, user composition, and endogenous story selection may contribute.Within-panel comparisons identify treatment effects, whereas panels represent populations with different numbers of affiliated peers.
- Average cue–response function: The ad unit lacked a no-cue baseline, and increasing cue count also slightly increased ad height and non-white pixels.These limitations motivated Experiment 2’s test of a more minimal social cue.
6. EXPERIMENT 2: INFLUENCE OF MINIMAL SOCIAL CUES
Experiment 2 randomizes whether Facebook social ads display a personalized cue naming one affiliated peer, then estimates how that cue and peer tie strength relate to clicks and page likes. The cue increases responses, and its relative effect is larger for stronger ties.
- Experimental design: Experiment 2 randomly assigns user–ad pairs to a single-peer social cue or no personalized social cue beneath the advertiser’s creative.The ad unit includes a title, image, caption, and small gray text about people who like the page.
- Average effect of a social cue: 3.8% to 5.4% click-rate increases and 9.6% to 11.6% like-rate increases result from the minimal cue across numbers of affiliated peers.For users with one affiliated peer, referring to that peer increases click rate by 5.2% (CI = [4.0%, 6.5%]) and like rate by 10.3% (CI = [8.4%, 12.3%]).
- Measure of tie strength: Tie strength is measured as the fraction of a user’s Facebook communications directed at or posted by an affiliated peer.The analysis uses communication frequency between users over the preceding 90 days to construct the directed tie-strength measure.
- Caveat: Experiment 2 may underestimate the effect of a minimal personalized cue because the no-cue condition still displays general page-like prevalence information.The Dia = 0 stimulus is not non-informative text; it can itself function as a non-personalized social cue.
- Results: Response rates increase with tie strength both when the minimal cue is present and when it is absent.Predicted response rates are plotted for users at the median total communication count.
- Results: 0.083 click-risk-ratio difference and 0.151 like-risk-ratio difference separate stronger from weaker ties between Wij = 0 and Wij = 0.045.Both differences favor stronger ties, indicating a larger relative increase from the social cue for stronger relationships.
7. CONCLUSION
The paper shows that minimal social cues substantially affect advertising responses, with stronger effects for stronger ties, while highlighting limits in measuring and randomizing tie strength.
- Contributions: The cue–response function differs dramatically from naïve observational estimates, demonstrating the value of experimentation for measuring social influence.The authors construct this function from randomized variation in the number of social signals received.
- Tie strength: Response rates increase with tie strength both when minimal social cues are present and when they are absent.Figure 7 reports model-fitted response rates for users at the median total communication count.
- Tie strength: Risk ratios for clicking and liking increase with tie strength, so social cues have stronger effects for stronger ties.Figure 8 compares predicted responses with and without the cue across measured tie strength.
- Contributions: Social cues can substantially affect consumer responses beyond correlations attributable to homophily.The contribution concerns minimal personalized social signals in advertising.
- Implications: The authors suggest incorporating tie-strength measures into the selection of advertisements and social cues.This recommendation follows their finding that cue effects vary with the strength of the consumer–peer connection.
- Limitations: The study’s tie-strength measure uses communication behavior, limits some analyses to sufficiently active consumers, and does not support causal inference about changing peer tie strength.The authors also note that only two methods of presenting social cues were examined.