Master Hypothesis Testing in Econometrics: A Week 5 Tutorial Guide with Real-World Examples

Introduction: Why Hypothesis Testing Matters in 2026

In June 2026, as AI models power stock market predictions and esports analytics, understanding hypothesis testing is more relevant than ever. Whether you're evaluating a new trading algorithm or assessing a player's performance improvement, you're essentially testing whether observed effects are statistically significant or just random noise. This tutorial covers the core concepts from ECON2206 Week 5, including null and alternative hypotheses, t-statistics, p-values, confidence intervals, and F-tests. By the end, you'll be able to apply these tools to real-world data with confidence.

Null and Alternative Hypotheses

The null hypothesis (H0) is a statement about a population parameter, typically that it equals a specific value. For example, in a study of the effect of a new AI training method on test scores, the null might be that the average score increase is zero. The alternative hypothesis (H1) can be one-tailed (e.g., the increase is greater than zero) or two-tailed (the increase is not zero). In esports, you might test whether a new gaming mouse improves reaction time: H0: μ = 0, H1: μ > 0 (one-tailed) or H1: μ ≠ 0 (two-tailed).

Type I and Type II Errors

A Type I error occurs when you reject a true null hypothesis (false positive). The level of significance (α) is the probability of making a Type I error, commonly set at 0.05. A Type II error occurs when you fail to reject a false null hypothesis (false negative). In AI fraud detection, a Type I error might flag a legitimate transaction as fraudulent, while a Type II error might miss actual fraud. Balancing these errors is crucial.

The t-Statistic and Critical Values

The decision rule: “Reject the null if the t-statistic exceeds the critical value.” The t-statistic is calculated as (estimated coefficient - hypothesized value) / standard error. The critical value comes from the t-distribution with n-k-1 degrees of freedom and depends on α and whether the test is one- or two-tailed. For example, with 30 degrees of freedom and α=0.05 (two-tailed), the critical value is about 2.042. If your t-statistic is 2.5, you reject H0.

P-Values Explained

The p-value is the smallest significance level at which you would reject the null, given the observed test statistic. A p-value of 0.03 means you would reject H0 at α=0.05 but not at α=0.01. In practice, if p < α, reject H0. For instance, in a study of a new app's user retention, a p-value of 0.04 indicates a statistically significant improvement at the 5% level.

Confidence Intervals

A 90% confidence interval for a parameter is an interval that, if you repeated sampling many times, would contain the true parameter 90% of the time. It is calculated as estimate ± (critical value × standard error). The width increases with higher confidence levels (e.g., 99% vs. 90%) because you need a wider range to be more certain. Importantly, if a hypothesized value lies inside the 90% confidence interval, you cannot reject the null at the 10% significance level (two-tailed). This equivalence is fundamental: the event “the 90% CI covers the hypothesized value” is the same as “fail to reject H0 at α=0.10.”

Testing a Single Linear Combination of Parameters

Sometimes you need to test a hypothesis like β1 + β2 = 1. You can rewrite it as a single parameter: θ = β1 + β2 - 1, then test H0: θ = 0. The t-statistic is (θ̂ - 0) / se(θ̂), where se(θ̂) is computed from the covariance matrix. This is common in finance when testing whether the sum of coefficients equals a specific value, such as in a portfolio model.

Restricted and Unrestricted Models

The unrestricted model includes all variables. The restricted model imposes the null hypothesis constraints (e.g., setting some coefficients to zero). The F-statistic compares the sum of squared residuals (SSR) from both models: F = [(SSR_r - SSR_ur) / q] / [SSR_ur / (n - k - 1)], where q is the number of restrictions. A large F leads to rejecting the null. For example, testing whether three AI model features jointly have no effect on prediction accuracy.

General Linear Restrictions

General linear restrictions are of the form Rβ = r, where R is a matrix of constants. The F-test can handle multiple restrictions simultaneously. In practice, econometric software computes this automatically, but understanding the formula helps interpret output.

Problem Set Examples

Ch4 Q2: Computing p-values

In part (iii), you need the p-value for a t-test. Using the standard normal approximation (since df is large), if t=2.5, the p-value for a two-tailed test is 2×P(Z>2.5)=2×0.0062=0.0124. With exact t-distribution, it would be slightly larger, but the approximation is fine for large samples.

Ch4 Q5: Confidence Intervals and Hypothesis Tests

This question typically involves constructing a 95% confidence interval for a coefficient and using it to test a hypothesis. Remember the equivalence: if the hypothesized value lies in the interval, fail to reject H0 at the 5% level.

Ch4 Q6: F-test for Multiple Restrictions

You might be asked to compute an F-statistic given SSR from restricted and unrestricted models. For example, SSR_r=150, SSR_ur=130, q=2, n=50, k=5. Then F = [(150-130)/2] / [130/(50-5-1)] = (20/2) / (130/44) = 10 / 2.9545 ≈ 3.38. Compare to F-critical with (2,44) df at 5% (about 3.20). If F>3.20, reject H0.

Trend Connection: AI and Esports

Hypothesis testing is everywhere. In 2026, AI models are used to predict esports match outcomes. A data scientist might test whether a new feature (e.g., player's recent form) significantly improves prediction accuracy. The t-test on the feature's coefficient tells if it's useful. Similarly, in finance, testing whether a trading strategy's alpha is significantly different from zero uses the same principles. Even in school life, analyzing whether a new study method improves exam scores relies on these tools.

Conclusion

Mastering hypothesis testing is essential for any econometrics student. By understanding null hypotheses, p-values, confidence intervals, and F-tests, you can critically evaluate empirical evidence. Practice with the problem set questions, and remember the connections to real-world applications in AI, gaming, and finance. Good luck with your ECON2206 studies!