A probability density describes the relative likelihood of values for a continuous hypothesis or random variable. Unlike discrete probability mass, it does not give the exact probability of one value. Integrating the density over a range gives the probability of that range. section 3.2
Think of the density curve as a landscape: probability comes from the area beneath a portion of the curve, not from the curve’s height alone. A uniform density from 0 to 1 has density 1 everywhere because its total area must equal 1; that does not mean every individual hypothesis has a 100% chance. section 3.2
The distinction from discrete probability becomes clearer in practice:
- Use a probability mass function for discrete hypotheses, where individual probability values can be added. section 3.2
- Use a probability density function for continuous hypotheses, where integration produces the probability of an interval. section 3.2
- Interpret a density’s height as relative concentration rather than as the probability of the exact value beneath it. section 3.2
Density as a regression likelihood
In a linear model, y = mx + ε. If the slope m has a normal prior and ε is normally distributed noise, the observed y has a normal distribution whose PDF supplies its likelihood. For a prior 𝒩(100, 100²), multiplying the slope by x gives 𝒩(100x, 100²x²), and adding noise gives 𝒩(100x, 100²x² + 20000²). section 3.13







