Conjugate Prior Updating

Description

Carries evidence forward across iterative studies by updating a conjugate prior (e.g. Beta for completion rates) with each round's data, yielding a running posterior.

Conjugate prior updating exploits the algebraic property that, for certain likelihood and prior pairs, the posterior belongs to the same family as the prior. For binary UX outcomes such as task success, a Beta(a,b) prior combined with s successes and f failures yields a Beta(a+s, b+f) posterior; Gamma-Poisson and Normal-Normal pairs handle counts and continuous measures. The posterior from one study becomes the prior for the next, so evidence accumulates transparently across iterative rounds instead of being thrown away. Its rigour lies in explicit prior specification, sensitivity checks against weakly informative priors, and decision rules stated in probability terms (e.g. P(success rate > 0.8) > 0.9) before data are collected. It also forces teams to decide when a redesign is substantial enough to warrant discounting or resetting prior evidence.

Attributes

Loss type
Unmigrated
Stage
Evaluation
Effort
Hours

Steps

  1. For binary UX outcomes (completion rate, error rate): use Beta-Binomial conjugate model.
  2. Prior: Beta(α₀, β₀) where α₀ = prior successes + 1, β₀ = prior failures + 1.
  3. After new data (s successes, f failures): Posterior = Beta(α₀ + s, β₀ + f).
  4. Posterior mean = (α₀ + s) / (α₀ + β₀ + s + f). 95% credible interval: qbeta(0.025, α, β) to qbeta(0.975, α, β) in R.
  5. Each new study updates the prior rather than starting from scratch.

Sources

  • Probability and Statistics — Morris H. DeGroot

Also known as

  • Beta-binomial updating
  • Sequential Bayesian updating
  • Bayesian accumulation across studies

Categories

  • Analytics & Behaviour
  • Statistics & Modelling