Posterior Distribution

What Is a Posterior Distribution?

A posterior is your updated belief after seeing the data. It combines a prior (what you believed going in) with the likelihood (what the data says), producing a full probability distribution of plausible parameter values. Unlike a point estimate, the posterior contains complete uncertainty information.

Also Known As

• Data science teams: posterior, posterior belief, updated distribution
• Growth teams: "the updated lift distribution"
• Marketing teams: post-test belief
• Engineering teams: posterior, p(theta | data)

How It Works

Imagine a Bayesian A/B test with 10,000 visitors per variant. You start with a Normal(0, 2%) prior on lift. After observing a 1.5% raw lift with standard error 0.3%, the posterior is approximately Normal(1.3%, 0.28%) — almost identical to the data-only estimate because the sample was large enough to overwhelm the prior. The posterior gives a 99%+ probability the lift is positive, and a 95% credible interval of roughly [0.7%, 1.9%]. Decisions follow directly from the posterior without any additional translation.

Best Practices

• Do summarize posteriors with both point estimates (mean/median) and credible intervals.
• Do compute posterior probabilities of meaningful thresholds (e.g., "P(lift > 1%)").
• Do run posterior predictive checks to validate your model.
• Do not cherry-pick posterior summaries that tell your preferred story.
• Do not confuse a narrow posterior with a meaningful effect; check effect size too.

Common Mistakes

• Reporting only the posterior mean; the full distribution is where the information lives.
• Ignoring sensitivity to prior choice when the posterior is dominated by it.
• Treating posterior probability as a guarantee; models can be misspecified.

Industry Context

• SaaS/B2B: Posterior probabilities communicate faster to executives than p-values.
• Ecommerce/DTC: Decision rules like "ship if P(lift > 0.5%) > 95%" translate directly.
• Lead gen/services: Posteriors let you integrate lifetime-value uncertainty into ship decisions.

The Behavioral Science Connection

Posterior updating mirrors how humans ideally would update beliefs — but rarely do. Kahneman's research on belief perseverance shows we cling to priors even when data contradicts them. Bayesian analysis forces the update to happen mathematically, counteracting the cognitive bias.

Key Takeaway

The posterior is the full answer your Bayesian analysis produces; treat it as a decision tool, not a single number.