Exam P Multivariate Random Variables

The July 2026 syllabus weights multivariate random variables at 23-30% and explicitly expects candidates to work with joint distributions, conditional and marginal distributions, covariance, correlation, order statistics, linear combinations, and Central Limit Theorem approximations.

That is a broad list, but it is not a random list. These topics all ask the same deeper question: once more than one random quantity is involved, how do you keep track of dependence, aggregation, and approximation without losing the structure of the problem?

Start with joint distributions because everything else in this block comes from them. Marginal and conditional distributions are just different ways of slicing the same joint information. Once that feels natural, covariance and correlation become much easier because they are summaries of how two variables move together rather than isolated formulas to memorize.

After that, move into correlation, linear combinations, and normal-based approximations. The CLT questions are usually easier when you already understand how means and variances behave under sums. Order statistics are important, but they are usually a narrower subtopic than the joint and aggregation material that shows up across many problem types.

Suppose a joint distribution places probability 0.20 on (X=0, Y=0), 0.30 on (0,1), 0.10 on (1,0), and 0.40 on (1,1). From that one table you can recover the marginal distribution of X, the conditional distribution of Y given X=1, and the covariance between X and Y.

The marginals give P(X=1)=0.50 and P(Y=1)=0.70. Conditioning on X=1 gives P(Y=1 | X=1)=0.40 / 0.50 = 0.80. Because E[XY]=0.40, the covariance is 0.40 - 0.50(0.70)=0.05. That is the core Exam P workflow in miniature: read the joint object correctly, then derive the quantity the question actually asks for.

The most common errors are procedural. Candidates sum the wrong row or column when finding marginals, divide by the wrong denominator in a conditional distribution, or forget the cross term when finding the variance of a linear combination.

Another easy trap is treating zero covariance as automatic independence. It can indicate independence in some familiar settings, but it is not the same statement in general.

This block is the bridge from textbook probability to actual modeling. Joint distributions become multivariate feature relationships, covariance structures drive aggregation and portfolio risk, order statistics matter for extremes, and CLT reasoning explains why averages and totals often look approximately normal even when the raw data do not.