TL;DR
This paper introduces a mathematical model for how "soft technologies" (like management fads) spread through a population. It shows that even with weak objective evidence, adoption patterns can appear deterministic due to social influence, and that these patterns are more robust than traditional "rational cascade" models suggest.
What Problem Does This Paper Address?
Many innovations, especially in fields like management and education, gain traction without clear evidence of their effectiveness. These "soft technologies" diffuse in ways that are puzzling, often adopted with fanfare and later discarded without definitive proof of their performance. The authors aim to model this diffusion process, acknowledging that decision-makers are influenced by both their own (imperfect) assessments and the choices of their peers. They propose that when objective evidence is weak, people rely more heavily on social cues, a phenomenon they link to Festinger's Hypothesis.
Methods
The core of the paper is a probabilistic model of diffusion. Two alternatives, A and B, spread through a population. Each period, one agent decides to adopt A or B. This decision is a weighted average of two factors:
- Individual Judgment (p): The probability of choosing the objectively superior option (A) if isolated from social influence. This reflects the quality of evidence.
- Social Influence (α): The weight given to the choices of already converted agents. A higher α means more reliance on social cues.
The model is analyzed both analytically (deriving propositions about means, variances, and convergence) and computationally (through simulations). They explore extreme cases (pure social influence, pure individual judgment) and the complex interplay between them.
Datasets
Not discussed in the paper.
Key Insights and Ideas
- Apparent Determinism from Stochasticity: The model demonstrates that even with random individual choices, the aggregate diffusion process can settle into stable patterns that appear deterministic, especially when objective evidence is weak.
- Soft Technologies are "Sticky": Unlike rational cascade models where conformity can be fragile, this model suggests that once a soft technology gains traction, it's more robust to new information. Small infusions of evidence won't easily derail a well-established trend.
- The Role of "p" and "α": The quality of objective evidence (p) and the reliance on social cues (α) interact in complex ways. Higher p generally leads to better outcomes, but the relationship with α can be nuanced.
- Festinger's Hypothesis in Action: The model explicitly incorporates the idea that as evidence (p) becomes more equivocal (closer to 1/2), reliance on social cues (α) tends to increase.
Key Results
- Convergence: The proportion of adopters of the superior technology (A) converges to a value determined by the quality of evidence (p) and the influence of social cues (α).
- Weak Evidence, Unpredictable Outcomes: When objective evidence is weak (low p), the final proportion of adopters is not predetermined by initial conditions; any outcome is roughly equally likely ex ante.
- Strong Evidence, Determined Outcomes: When evidence is strong (high p), the outcome is more determined by the evidence quality and agent competence.
- Robustness: The model's diffusion processes stabilize once public opinion is established, unlike the fragile cascades in fully rational models.
- Initial Conditions Matter: The initial bias (fo) and the size of the initial seed influence the speed and direction of convergence.
Limitations and Future Work
- Rationality Assumption: While the model moves beyond perfect Bayesian rationality, it still assumes agents make choices based on a weighted average of individual judgment and social influence. The paper acknowledges that agents may not fully understand their own decision processes.
- Endogenizing Festinger's Hypothesis: The paper explores how α could be a function of p, suggesting that improvements in evaluation technology might have counterintuitive effects on diffusion dynamics.
- Diffusion in Structured Communities: The authors suggest extending the model to account for status hierarchies and clustered interactions within communities.
- Leakage of Information: Future work could model how p might change over time as information "leaks" out.
- Waves of Innovations: The paper speculates that the model's findings might extend to understanding waves of innovations, particularly if they are soft technologies.
Summary updated