Foundations of Statistical Inference
Why Statistical Inference Exists
From data to uncertainty
Statistics formalizes learning from noisy data. The central problem is to infer features of an unknown data-generating process from a finite sample, while quantifying uncertainty about what has been learned.
Core idea
A statistical conclusion is never just a point estimate; it is an estimate plus a measure of uncertainty, and often a decision rule under loss.
Probability Models and Sampling
Statistical model
A statistical model is a family of distributions indexed by an unknown parameter . The observed sample is treated as random, usually conditionally independent given , though modern research frequently relaxes independence and identical distribution assumptions.
Likelihood and estimation
Given data , the likelihood is up to a multiplicative constant independent of . Estimation often begins by maximizing , minimizing an empirical risk, or solving estimating equations of the form .
Which quantity is a function of the parameter given fixed observed data?
The likelihood is viewed as a function of the parameter after the data are observed.
Correct answer: Likelihood
State the difference between a parameter and an estimator.
Parameters are non-random in the classical frequentist setup, while estimators vary across repeated samples.
Correct answer: A parameter is a fixed but unknown feature of the population model; an estimator is a random variable computed from the sample that targets the parameter.
Seminal ideas
Seminal pillars
What does Fisher information primarily measure?
Fisher information quantifies how much the likelihood changes with the parameter, which is linked to estimation precision.
Correct answer: Sensitivity of the likelihood to changes in the parameter
Why is uncertainty quantification essential in statistics?
Inference is about what can be learned reliably from incomplete information.
Correct answer: Because finite samples do not identify the truth exactly, so conclusions must reflect sampling variability and model uncertainty.