The Likelihood Function
You'll learn how the likelihood quantifies the probability of seeing your actual data if specific parameter values were true.
What Is the Likelihood?
The likelihood function answers a crucial question: "If my model parameters had specific values, how probable would it be to observe the data I actually collected?"
Unlike a probability distribution that varies over possible data outcomes, the likelihood varies over possible parameter values while treating your observed data as fixed. Think of it like reverse-engineering: you've seen the outcome (your data), and now you're asking which parameter settings would have made that outcome most plausible.
A Simple Example
Imagine you flip a coin 10 times and get 7 heads. The likelihood function tells you how probable this result is for different values of the coin's bias parameter θ (probability of heads):
- If θ = 0.5 (fair coin): the likelihood of seeing exactly 7 heads has one value
- If θ = 0.7 (biased coin): the likelihood has a different (higher) value
- If θ = 0.3: the likelihood is lower
The likelihood function sweeps across all possible θ values, showing which ones make your observed data more or less probable.
Why It Matters in Bayesian Analysis
In Bayesian inference, the likelihood function is what connects your prior beliefs to actual evidence. It's how data "speaks" to update your understanding. Without the likelihood quantifying how well different parameter values explain your observations, you'd have no way to learn from data—you'd be stuck with only your prior assumptions.
Key Takeaway: The likelihood function measures how probable your observed data would be under different parameter values, serving as the critical bridge between theoretical models and empirical evidence.