Conditional Distributions

The "Slicing" Intuition

Imagine you have a joint PDF $f(X, Y)$ representing the height and weight of people. If I tell you, "This person is exactly 180cm tall," we are no longer looking at the whole 3D surface. We are looking at a 2D curve at $X = 180$.

However, because we "cut" the surface, the area under our slice might not equal 1. To make it a valid probability distribution, we must re-normalize it by dividing by the marginal probability.

The Definition

For a fixed value of $Y=y$, the conditional PDF of $X$ is:

$f_{X|Y}(x \mid y) = \frac{f(x, y)}{f_Y(y)}$

This is essentially the multivariate version of the Conditional Probability Definition.

Law of Total Probability (Continuous)

We can find the overall marginal probability of $X$ by summing up all the "slices" of its conditional distribution, weighted by the probability of each slice:

$f_X(x) = \int_{-\infty}^{\infty} f_{X|Y}(x \mid y) f_Y(y) \, dy$

The Law of Iterated Expectations (Adam's Law)

This is a "God-tier" shortcut for finding the mean of a variable. It states:

$E[X] = E[E[X|Y]]$

The "Average of Averages" Intuition

If you want to find the average height of a city, you can:

1. Calculate the average height of each neighborhood ($E[X|Y]$).
2. Then, take the average of those neighborhood averages ($E[\dots]$).

This is extremely useful when the overall mean is hard to calculate, but the means within specific groups are easy to find.

Test Your Knowledge