Updating Conjugate Priors with Data
You'll learn: How to mechanically update conjugate prior parameters using observed data to obtain your posterior distribution.
The Updating Mechanics
Once you've established a conjugate prior (like Beta-Binomial, Gamma-Poisson, or Normal-Normal), updating with data is remarkably simple: you just add the data counts to your prior parameters.
For Beta-Binomial conjugacy:
- Prior: Beta(α, β)
- Data: k successes in n trials
- Posterior: Beta(α + k, β + (n - k))
For Gamma-Poisson conjugacy:
- Prior: Gamma(α, β)
- Data: total count x from n observations
- Posterior: Gamma(α + x, β + n)
For Normal-Normal conjugacy (known variance σ²):
- Prior: Normal(μ₀, σ₀²)
- Data: n observations with mean x̄
- Posterior mean: weighted average of μ₀ and x̄ based on relative precisions
Why This Works
Think of prior parameters as "pseudo-observations." If your Beta prior is (α=2, β=3), it's like you've already seen 2 successes and 3 failures. When you observe 5 successes and 1 failure, you simply add them: (2+5, 3+1) = (7, 4).
The posterior is always in the same distributional family as your prior—that's what makes it conjugate. No complex integration needed; just parameter arithmetic.
Practical Example
Prior belief: Beta(10, 10) — moderately confident the coin is fair
Observe: 7 heads, 3 tails
Posterior: Beta(10+7, 10+3) = Beta(17, 13)
Your beliefs shifted slightly toward "more heads" but weren't completely overridden because your prior represented substantial existing evidence.
Key Takeaway: With conjugate priors, Bayesian updating is simple parameter addition—data counts add directly to prior pseudo-counts, yielding a posterior in the same distributional family.