Conjugate Normal-Normal Model
What You'll Learn
You'll learn how normal priors and normal likelihoods combine algebraically to produce a normal posterior distribution—one of the most elegant results in Bayesian statistics.
The Core Idea
When your data follows a normal distribution (a very common assumption for continuous measurements) and you place a normal prior on the mean parameter, something beautiful happens: the posterior distribution is also normal, with a closed-form solution you can calculate directly.
This is called the conjugate normal-normal model, and it's the Normal distribution's version of what you saw with Beta-Binomial and Gamma-Poisson conjugacy.
How It Works
Suppose you're estimating a population mean μ, and you believe it follows a normal distribution. You have:
- Prior belief: μ ~ Normal(μ₀, σ₀²) — your initial guess with uncertainty
- Likelihood: Data points x₁, x₂, ..., xₙ ~ Normal(μ, σ²) — assuming known data variance σ²
- Result: The posterior μ | data is also Normal, with updated parameters
The math combines your prior precision (1/σ₀²) and the data's precision (n/σ²) to weight how much you trust each source. More precise information (smaller variance) gets more weight in determining the posterior mean.
The Formula (Simplified)
The posterior mean is a weighted average of your prior mean (μ₀) and the sample mean (x̄), where weights depend on relative precisions. The posterior variance shrinks—you're always more certain after observing data.
Key Takeaway: The conjugate normal-normal model lets you analytically update beliefs about a mean by combining normal prior and normal likelihood into a normal posterior, avoiding numerical approximation and providing intuitive precision-weighted estimates.