Epidemic Models (SIR/SEIR) for Adoption and Virality

Description

Models users moving between compartments (susceptible, exposed, infected/active, recovered/churned) to estimate spread rates, reproduction numbers and saturation.

Compartmental models describe populations moving between states via rates: in SIR, susceptible people become infected through contact (rate beta) and recover (rate gamma). The basic reproduction number R0 = beta/gamma determines whether spread grows. SEIR adds an exposed (latent) stage; SIRS allows reinfection. Fitted to adoption or referral data, these models describe viral growth, word-of-mouth, feature adoption within organisations and even spread of misinformation or abusive behaviour on platforms. They distinguish contact rate from 'stickiness' (recovery) and quantify interventions, much more precisely than a 'viral coefficient' computed from a single cohort.

Attributes

Loss type
Unmigrated
Effort
Weeks

Steps

  1. Define compartments relevant to the product (e.g. unaware, aware, active, lapsed) and transitions.
  2. Collect time series of counts in each compartment, or referral chains if available.
  3. Fit model parameters by maximum likelihood or Bayesian methods, with uncertainty.
  4. Estimate effective reproduction number over time and check fit to observed curves.
  5. Simulate interventions: increasing contact (sharing), reducing churn (recovery), seeding.
  6. Validate predictions on new cohorts or markets.

Sources

  • Kermack & McKendrick, A contribution to the mathematical theory of epidemics (1927), Proceedings of the Royal Society A
  • Anderson & May, Infectious Diseases of Humans (1991)
  • Keeling & Rohani, Modeling Infectious Diseases in Humans and Animals (2008)

Also known as

  • SIR model
  • SEIR model
  • Compartmental models
  • Kermack-McKendrick model

Categories

  • Epidemiology & Population Health