Null Hypothesis

Fundamental principle in statistical hypothesis testing and inference

Definition and Purpose

Concept

Null hypothesis (H0): default assumption of no effect, no difference, or status quo in statistical inference.

Purpose

Serves as baseline to test evidence against; foundation for statistical decision-making and inference.

Context

Used in scientific experiments, surveys, clinical trials, quality control, and social sciences.

Formulation of Null Hypothesis

Expression

Typically stated as equality: parameter equals specified value (e.g., μ = μ0, p = p0).

Parameter Types

Mean, proportion, variance, correlation coefficient, regression slope, etc.

Notation

Denoted as H0, contrasted with alternative hypothesis HA or H1.

Role in Hypothesis Testing

Framework

Defines null condition to be tested; statistical test evaluates observed data under assumption H0 true.

Decision Rule

Reject H0 if data contradict null beyond threshold; fail to reject otherwise.

Objective

Control error rates; maintain objectivity by presuming no effect until evidence shows otherwise.

Test Statistics and Null Hypothesis

Definition

Numerical summary computed from sample data to assess compatibility with H0.

Types of Test Statistics

t-statistic, z-statistic, chi-square, F-statistic, depending on data type and hypothesis.

Distribution Under H0

Sampling distribution of test statistic assumed known or approximated under null hypothesis.

P-Value and Decision Making

Definition

Probability of observing test statistic as extreme or more extreme than observed, assuming H0 true.

Interpretation

Small p-value indicates data unlikely under H0, suggesting evidence against null.

Thresholds

Common significance levels: α = 0.05, 0.01, 0.10; compare p-value to α for decision.

Type I and Type II Errors

Type I Error (α)

Rejecting true null hypothesis; false positive; controlled by significance level.

Type II Error (β)

Failing to reject false null hypothesis; false negative; depends on test power.

Trade-off

Reducing one error type increases the other; balanced by sample size, effect size, α level.

Error TypeDescriptionConsequence
Type IReject true H0False alarm
Type IIFail to reject false H0Missed detection

Alternative Hypothesis

Definition

Contrasts null; represents presence of effect, difference, or relationship.

Types

One-sided (directional): e.g., μ > μ0 or μ < μ0. Two-sided (non-directional): μ ≠ μ0.

Role

Defines rejection region; guides interpretation of results beyond null assumption.

Practical Examples

Example 1: Mean Comparison

H0: μ = 100 (population mean equals 100). HA: μ ≠ 100.

Example 2: Proportion Test

H0: p = 0.5 (coin is fair). HA: p ≠ 0.5 (coin biased).

Example 3: Regression Slope

H0: β = 0 (no relationship). HA: β ≠ 0 (significant predictor).

Assumptions and Conditions

Random Sampling

Sample data collected randomly and independently from population.

Distributional Assumptions

Normality (often assumed for parametric tests), homoscedasticity, and independence.

Sample Size

Sufficient size for reliable approximations of sampling distributions.

Limitations and Criticisms

Misinterpretation

Rejecting H0 ≠ proving HA true; p-value ≠ probability H0 true.

Overemphasis on Significance

Neglect of effect size and practical relevance; dichotomous decision oversimplifies.

Dependence on Sample Size

Large samples detect trivial effects; small samples may miss real effects.

Extensions and Related Concepts

Bayesian Hypothesis Testing

Incorporates prior beliefs; compares hypotheses using Bayes factors.

Confidence Intervals

Alternative approach summarizing parameter uncertainty without binary decision.

Multiple Testing

Adjustments (e.g., Bonferroni) to control error rates across many hypotheses.

Key Formulas

Test Statistic for Mean (Z-test)

z = (x̄ - μ₀) / (σ / √n)

Test Statistic for Mean (t-test)

t = (x̄ - μ₀) / (s / √n)

P-Value Calculation

For two-sided test:

p-value = 2 × P(T ≥ |t|) under H₀ distribution

Power of Test

Power = 1 - β = P(reject H₀ | Hₐ true)

References

  • Fisher, R.A., "Statistical Methods for Research Workers," Oliver & Boyd, 1925, pp. 1-100.
  • Neyman, J., & Pearson, E.S., "On the Problem of the Most Efficient Tests of Statistical Hypotheses," Philosophical Transactions of the Royal Society A, vol. 231, 1933, pp. 289-337.
  • Casella, G., & Berger, R.L., "Statistical Inference," Duxbury Press, 2002, pp. 100-150.
  • Wasserman, L., "All of Statistics: A Concise Course in Statistical Inference," Springer, 2004, pp. 50-85.
  • Lehmann, E.L., & Romano, J.P., "Testing Statistical Hypotheses," Springer, 2005, pp. 30-90.