Mathematical Statistics: Lecture

Do not walk into a proof on "Complete Sufficient Statistics" cold. Spend 20 minutes the night before skimming the textbook. Focus only on:

A point estimate like $\hat\theta = 5$ is rarely enough. Is it exactly 5? Probably not. We need a range. This leads to Confidence Intervals.

A $95%$ confidence interval does not mean there is a 95% chance the parameter is in the interval (the parameter is fixed; the interval is random).

The Correct Interpretation: If we repeated the experiment 100 times, calculating a new interval each time, roughly 95 of those intervals would contain the true parameter. mathematical statistics lecture

Mathematically, we construct bounds using probability statements: $$P(L \leq \theta \leq U) = 1 - \alpha$$

This accounts for the sampling error. It transforms a single number into a rigorous statement about uncertainty.

The behavior of an RV is described by:

You cannot transcribe the board. You must filter.

This is the climax of the course.

This is where mathematical statistics distinguishes itself from applied stats. Do not walk into a proof on "Complete

Sufficiency: A statistic $T(X)$ is sufficient for $\theta$ if it contains all the information in the sample regarding $\theta$. Once you know $T$, the individual data points provide no extra information about $\theta$.

The Rao-Blackwell Theorem: This is a profound result. It states that if you have a crude estimator and a sufficient statistic, you can "improve" the crude estimator by conditioning on the sufficient statistic. It guarantees that we never need to throw away data efficiency if we use sufficient statistics.