Prior Sensitivity Analysis
You'll learn how to test whether your starting assumptions significantly change your final conclusions in Bayesian reasoning.
When you use Bayes' Theorem, you start with a prior belief (your initial probability before seeing data) and update it with evidence to get a posterior (your revised belief). But what if you're not sure about your prior? Prior sensitivity analysis answers: "How much does changing my starting belief change my final answer?"
The Core Idea
Imagine you're diagnosing a rare disease. If you start with a prior of 1% prevalence versus 5% prevalence, will the positive test result lead to drastically different posterior probabilities? Prior sensitivity analysis tests multiple reasonable priors to see if your conclusion is robust (stable across priors) or fragile (highly dependent on your starting assumption).
When Priors Matter Most
Priors have the biggest impact when:
- Evidence is weak: A mildly positive test result won't overcome a very skeptical or very confident prior
- Sample size is small: With little data, your starting belief dominates
- Priors are extreme: Starting at 0.01% versus 10% creates vastly different posteriors
Priors matter less when:
- Evidence is strong: Overwhelming data swamps your initial belief
- Sample size is large: More observations make the data speak louder than assumptions
- Priors are similar: Starting at 30% versus 35% won't create huge differences
Conducting the Analysis
- Choose a range of plausible priors (e.g., 1%, 5%, 10%)
- Apply Bayes' Theorem with the same evidence for each prior
- Compare the resulting posteriors
- If posteriors are similar, your conclusion is robust; if wildly different, report the uncertainty
Key Takeaway: Prior sensitivity analysis reveals whether your Bayesian conclusions depend critically on starting assumptions—essential for honest, transparent reasoning when initial beliefs are uncertain.