Adoption, influence, and the value of a growing networkUpdated 25 October 2026
Open the notebookNotebook / live research
A public record of what I am reading, what it is making clearer, and where the answer is still not ready.
The latest batch covers 3 papers and 3 chapters from 2 books: influencer models, referral acquisition, seeding, adoption attributes, and the move from standalone tools to networks.
215 evidence cards reviewed · 31 of 43 essential sources have reviewed evidence · updated 25 October 2026.
Test which seeding findings transfer from low-risk information sharing to adoption that needs social reinforcement. Three scanned sources still await OCR.
Source #011 · Chapter 6 reviewed
A foundational account of how new ideas move through communication channels, over time, within social systems.
Working note: the potential adopter’s perceived advantage and fit matter in Rogers’ framework. Chapter 6 also cautions against explaining adoption only through innovations that already succeeded. This is scholarly synthesis, with its cited studies still needing separate evaluation.
Chen proposes that standalone utility can support a product before its network forms. He also warns that tool users may never make the transition: sharing or collaboration needs a close connection to the original task. This is a practitioner hypothesis illustrated by company cases, not a measured causal effect.
Open the original sourceChen proposes that standalone utility can support a product before its network forms. He also warns that tool users may never make the transition: sharing or collaboration needs a close connection to the original task. This is a practitioner hypothesis illustrated by company cases, not a measured causal effect.
Rogers organizes adoption attributes around perceived advantage, fit, complexity, trialability, and visible results. The framework is a way to ask better questions about a product; the five attributes can overlap and may miss something important in a particular setting. It does not establish five universal growth levers.
In the mobile referral campaign, more connected customers were more likely to participate and made more referrals when they did. Among active referrers, connectivity did not significantly predict conversion per referral. These are observational associations; a nonsignificant result does not prove equal persuasiveness.
The authors reported roughly three weeks of modeled signup response to referral activity at one site. The April 2008 manuscript gives an unclear uncertainty rule, so that duration should not be read as a precise cutoff. Signup changes were temporary in their model; neither retained growth nor a causal effect of redesigning referrals was established.
Van den Bulte and Joshi separate people by assumed adoption drivers and direction of influence, rather than when they adopt. Their model shows why timing alone need not reveal influence. Fitting an aggregate growth curve does not demonstrate that the proposed social mechanism caused it.
Voluntary shares were slightly more effective per exposed peer, but automatic sharing produced far more total downstream claims because everyone shared. The setting was Facebook Offers, not general product adoption.
Personal invitations converted better per message, while automated broadcasts produced more total peer adoption because they happened much more often. The result comes from a short-run Facebook app experiment.
For Facebook photos that had already received a few reshares, recent sharing tempo was especially useful for predicting continued growth. That is a forecasting result, not evidence that speeding a cascade up would cause virality.
A large cascade may be a hub-driven burst or many generations of person-to-person spread. Its size alone does not reveal what kind of diffusion happened.
Across seven online domains, 73–95% of observed cascades did not diffuse beyond their starting point. Rare spectacular outcomes are a poor picture of ordinary diffusion.
In this retailer referral program, the association between incoming recommendations and purchases varied by product category and appeared to saturate. The incentives and recorded purchases make this an imperfect stand-in for ordinary word of mouth.
Long ties can help simple contagions travel, yet fail for behaviours that need repeated affirmation. In the model, a bridge’s capacity for reinforcement matters as much as its reach.
Word of mouth, visible social signals, and network effects can all increase adoption through different routes. Treating them as a single viral force makes it difficult to know what is actually working.
Facebook invitation acceptance tracked the number of distinct social contexts around a recipient better than raw contact count. It is a predictive pattern, not a randomized test of diversity.
Friends adopting the same product can reflect influence, but also shared preferences and circumstances. In this study, better matching sharply reduced the apparent peer effect; the remainder was still an upper bound.
For a low-cost registration behaviour, clustered networks spread adoption faster and farther than degree-matched random networks. The simplified experiment was designed to isolate topology, not recreate every real social relationship.
More amusing or anger-inducing story versions increased stated willingness to share in experiments. Those experiments measured intended sharing, rather than observed transmission in a live network.
In a fact-check-labeled Twitter sample, false stories traveled farther and faster than true stories. The study describes a striking pattern, but does not show that falsity itself caused the difference.
YouTube’s recommender narrowed millions of videos to a few personalized candidates. It shows how ranking shapes exposure, but it is one dated platform architecture—not direct evidence of interpersonal virality.