Notes: HKUST notes on hypothesis testing

A major part of statistical inference is hypothesis testing, a decision procedure that gives probabilistic conclusions on underlying population from limited sample.

Working Assumptions

A working assumption is a statement used for pragmatic necessity to construct theoritical arguments. It should be falsifiable and frequently examined.

You start from an assumption that you believe is at least partially true, and the results will tell you that you used a good assumption, or that your assumption needs to be modified, or that your assumption is flat wrong.

Working assumptions cannot be logically false, nor logically true.

In contrast, a core assumption, like a principle, is a proposition that is either logically true or accepted as fundamental for pragmatic use. Core assumptions are thus rarely examined.

Elements of Hypothesis Test

Hypothesis is a statement about a population parameter. We referred to the hypothesis we test on as the Null hypothesis \( H_0 \). If the null gets rejected, we accept the alternative hypothesis \( H_1 \) instead.

Hypothesis test is a rule that specifies for every sample point whether to reject or accept the null. A hypothesis test divides the sample space into a rejection region and an acceptance region.

A typical hypothesis testing procedure:

  1. State the null and alternative hypothesis: \( H_0 = \theta \in \Theta_0 \quad H_1 = \theta \in \Theta_0^C \)
  2. State the hypothesis test
    1. a test statistic \( W(\mathbf{X})\)
    2. a rejection interval \(R\)
  3. Sampling and computing test statistic.
  4. Rejection or acceptance of the null hypothesis.

Types of hypothesis tests:

  1. significance test

Significance Test

passive acceptance

statistical significant difference rejection

Likelihood Ratio Test (LRT)

Def: likelihood ratio statistic

Def: likelihood ratio test

Def: Nuisance parameter

Uniformly Most Powerful (UMP) Test

Def: Type I Error, Type II Error

Def: power function

Def: size

Def: level

Def: unbiased

Def: uniformly most powerful (UMP) test

Thm: (Neyman-Pearson)

Def: monotone likelihood ratio (MLR)

Thm: (Karlin-Rubin)


🏷 Category=Statistics