Mathematical Statistics Lecture -

Matching Methods for Causal Inference: A Review and a Look Forward Statistical Science (via Project Euclid )

You might be sitting in the lecture hall thinking, "When will I ever derive the Cramér-Rao Lower Bound in a job interview?" The answer: never directly. But the skills you build are invaluable.

To understand the "mathematical statistics lecture," you must understand the student.

Mathematical statistics also explores relationships between multiple variables. Simple Linear Regression

, uniquely determines the behavior of any random variable, whether discrete or continuous. 2. Sampling and Asymptotic Theory mathematical statistics lecture

You cannot do mathematical statistics without measure-theoretic probability (or at least advanced calculus-based probability). Lectures typically spend the first 3-4 weeks on:

If you have $k$ parameters to estimate, set the first $k$ population moments equal to the first $k$ sample moments and solve the system of equations.

Hypothesis testing is a formal statistical framework used to make decisions about population parameters using sample data. The status quo, representing no effect or no difference. Alternative Hypothesis ( H1cap H sub 1 Hacap H sub a ): The claim the researcher wants to prove. Decision Matrix and Error Types Decision \ True State H0cap H sub 0 H0cap H sub 0 Reject H0cap H sub 0 Type I Error ( Correct Decision ( Fail to Reject H0cap H sub 0 Correct Decision ( Type II Error ( Significance Level (

One of the primary goals of mathematical statistics is estimation—using sample data to estimate unknown population parameters (such as the true population mean or variance σ2sigma squared Point Estimation Matching Methods for Causal Inference: A Review and

Λ(x)=L(θ0|x)L(θ1|x)≤kcap lambda open paren x close paren equals the fraction with numerator cap L open paren theta sub 0 vertical line x close paren and denominator cap L open paren theta sub 1 vertical line x close paren end-fraction is less than or equal to k If the likelihood ratio falls below a critical value , we reject H0cap H sub 0 6. Bayesian Inference: An Alternative Paradigm

Moving beyond single-point guesses, we encounter interval estimation. Rather than providing one value, we provide a range—a confidence interval—that is likely to contain the true parameter. It is a common misconception that a 95% confidence interval means there is a 95% probability the parameter is inside. In frequentist statistics, the parameter is fixed; it is the interval itself that is random. Therefore, the 95% refers to the process: if we repeated the experiment many times, 95% of the calculated intervals would contain the true parameter.

A deep lecture does not end with worship of frequencyist methods. The professor will step back and introduce decision theory : a loss function ( L(\theta, a) ), a risk ( R(\theta, \delta) = \mathbbE_\theta[L(\theta, \delta(X))] ). An estimator is admissible if no other estimator has uniformly lower risk. The Bayes estimator —minimizing posterior expected loss—emerges as a natural solution.

), we find the score function (the derivative of the log-likelihood), set it to zero, and solve: Sampling and Asymptotic Theory You cannot do mathematical

An unbiased estimator that hits this lower bound is called a . 4. Interval Estimation (Confidence Intervals)

Does the estimator get closer to the true value as the sample size n → ∞?

A formal lecture on mathematical statistics almost always begins with probability theory. The foundational premise is that data is a realization of a random process.