Practical Computation of Credible Intervals
You'll learn how to actually extract credible intervals from your Bayesian analysis using either analytical formulas or posterior samples.
The Core Idea
Once you've performed Bayesian inference and obtained a posterior distribution, you need a practical way to extract credible intervals. You have two main routes: analytical computation (when conjugate priors give you a known distribution) or sampling-based computation (when you've used MCMC or other sampling methods).
Analytical Approach
When you have a conjugate prior and can write down the exact posterior distribution, computing credible intervals is straightforward:
For a Beta posterior (from Beta-Binomial): Use the quantile function of the Beta distribution to find the 2.5th and 97.5th percentiles for a 95% credible interval.
For a Normal posterior: Use the mean ± (z-score × standard deviation) where z-score comes from the normal distribution.
Think of this like looking up values in a statistical table—you know the distribution's formula, so you just need to find the right cutoff points.
Sampling-Based Approach
When you have posterior samples from MCMC (thousands of draws from the posterior), credible intervals are even simpler:
- Sort your samples in ascending order
- For a 95% equal-tailed interval: take the 2.5th and 97.5th percentiles of your samples
- For a 95% HDI: find the narrowest interval containing 95% of the samples
This is like having a bag of 10,000 balls with numbers on them—just sort them and count to find your boundaries.
Key Takeaway: In practice, compute credible intervals either by using quantile functions of known posterior distributions (analytical) or by taking percentiles of posterior samples from MCMC—both approaches give you the probability ranges you need to make decisions.