Complex Contagions in Charitable Donations

Jie Gao, Golnaz Ghasemi, Jason Jones, Grant Schoenebeck2019 · DOI

TL;DR

This paper demonstrates complex contagion in human decision-making through a series of experiments on charitable donations. By manipulating peer donation behavior, the researchers show that social influence can not only direct donations to unpopular charities but also exhibit non-submodular influence curves, a hallmark of complex contagion. This provides strong experimental evidence for a phenomenon previously difficult to isolate from observational data.

What Problem Does This Paper Address?

The spread of behaviors and information through social networks, known as cascades, can be either "simple" or "complex." Simple cascades are driven by single exposures, while complex cascades require reinforcement from multiple contacts. Understanding which type of cascade is at play is crucial for predicting and influencing behavior. However, disentangling complex contagion from factors like homophily and environmental effects in observational data is challenging. This paper aims to provide direct experimental evidence for complex contagion in human decision-making, specifically in the context of charitable donations, and to show how it can be used to influence behavior.

Methods

The researchers conducted three online experiments using Amazon Mechanical Turk. Participants were rewarded for answering questions and then given the option to donate a portion of their earnings to charity. The core of the experimental design involved manipulating the donation behavior of simulated "neighbors" (bots controlled by the experimenters).

Key methodological elements include:

  • Simulated Social Networks: Participants were told they were part of a group and observed the donation choices of their simulated groupmates.
  • Similarity Manipulation: In Experiments A and B, participants were first asked preference questions to establish a baseline, and then their simulated neighbors were presented as either "High Similarity" or "Low Similarity" based on their answers. This was intended to test if perceived similarity amplifies social influence.
  • Neighbor Donation Manipulation: The number of neighbors who donated to a specific charity (Schistosomiasis Control Initiative - SCI) or to any charity was systematically varied. This allowed for the construction of "influence curves" showing how the probability of a participant donating changes with the number of donating neighbors.
  • Complex Contagion Measurement: Complex contagion was identified by a non-submodular influence curve, meaning that the marginal benefit of an additional donating neighbor increases as more neighbors donate. This was formally tested using the Chen-Szroeter test.

Datasets

The study utilized data from 4474 participants recruited from Amazon Mechanical Turk across three experiments (Experiment A, B, and C). No external datasets were used for training or evaluation. The "datasets" are the experimental results generated from participant interactions.

Key Insights and Ideas

  • Social Influence is Powerful: Participants are significantly more likely to donate and to donate to a specific charity when they observe their peers doing so. This confirms basic social influence.
  • Directing Donations to Unpopular Causes: The experiments successfully directed donations towards the Schistosomiasis Control Initiative (SCI), an otherwise unpopular charity, by manipulating peer behavior. This highlights the practical application of understanding social influence.
  • Evidence for Complex Contagion: The influence curves for donations to SCI, particularly in Experiments B and C, showed non-submodularity. This means that the impact of each additional donating neighbor was greater than the previous one, indicating that multiple exposures are needed for adoption, a key characteristic of complex contagion.
  • Similarity Doesn't Always Matter: Surprisingly, the manipulation of neighbor similarity (high vs. low) did not consistently affect the strength of social influence. This suggests that for donation behavior, the sheer number of peers acting is a stronger driver than perceived similarity.
  • Option Space Matters: The paper suggests that complex contagion is more likely to emerge when there are multiple options, making any single option less inherently appealing. This "space of options" amplifies the salience of social signals.

Key Results

  • Hypothesis I (Social Influence): Strongly supported. The more neighbors who donated, the more likely a participant was to donate. For example, in Experiment A, observing 3 neighbors donate to SCI increased the donation rate by over eleven times compared to observing none.
  • Hypothesis Ia (Similarity): Not supported. The similarity manipulation did not show a consistent or significant effect on influence in any of the experiments.
  • Hypothesis II (Complex Contagion): Supported, particularly for donations to the specific charity SCI.
  • Experiment A showed some support for non-submodularity in SCI donations, though the formal test yielded a modest p-value (0.186).
  • Experiment B provided stronger evidence for non-submodularity in SCI donations (p-value of 0.017).
  • Experiment C also supported non-submodularity in SCI donations (p-value of 0.21, though the authors note this is more likely than not).
  • Donating at All vs. Specific Charity: Non-submodularity was more evident when participants chose a specific charity (SCI) rather than deciding whether to donate at all.

Limitations and Future Work

  • Similarity Manipulation: The authors acknowledge the lack of a significant effect from the similarity manipulation and suggest future work could explore ways to increase its salience or investigate other factors that might mediate similarity's influence.
  • Distinguishing Factors: The study could not fully distinguish between the effect of the number of options and the baseline popularity of the target option. Future research could manipulate these factors independently.
  • Generalizability: While using a large sample size from Amazon Mechanical Turk, the generalizability to real-world donation scenarios with different social contexts and motivations needs further exploration.
  • Future Directions: The authors encourage further theoretical and experimental work on complex contagion, emphasizing the importance of characterizing influence curves as submodular or non-submodular. They also suggest exploring the role of baseline popularity and option space size in future studies. Practical implications for charities, such as using clear names and "social proof," are also highlighted.

Summary updated

Abstract

Cascades over social networks can spread information, beliefs, diseases, technologies, and behaviors. Simple cascades spread from mere contact and produce submodular influence curves. Complex cascades assume agents with thresholding behavior and may produce non-submodular influence curves. In this study, we run three experiments that request charitable donations from human participants and experimentally manipulate whether and where their peers donate. We find evidence that we can (1) direct donations to an otherwise unpopular charity and (2) elicit complex contagion as evidenced by a non-submodular influence curve. The findings represent the most straightforward evidence to date of treatment-induced complex contagion - explicitly and formally defined - in human decision-making.
BibTeX
@misc{gao2019complex,
  author = {Jie Gao and Golnaz Ghasemi and Jason Jones and Grant Schoenebeck},
  title = {{Complex Contagions in Charitable Donations}},
  year = {2019},
  doi = {10.31235/osf.io/tczvw},
}
Figure