Historical Context & Motivation
Imagine you are a quality-control manager at a factory that produces light bulbs. The company claims each bulb lasts an average of 1 000 hours, but customers are complaining. You test a sample of 50 bulbs and find their average life is only 970 hours. Is the company's claim wrong, or did you just happen to grab a weaker batch? This is exactly the kind of question that hypothesis testing and confidence intervals were designed to answer. These tools let us move from guesswork to mathematically justified decisions.
The development of these ideas spans several centuries, growing from early probability theory into the sophisticated framework used in science, medicine, economics, and engineering today. Understanding the history helps us appreciate why these methods work and what assumptions sit beneath them.
The central question these pioneers were tackling remains the same one you will answer in this lesson: how can we use a limited sample to make defensible claims about a much larger population? Whether you are evaluating a new drug, testing a physics theory, or checking factory output, the logic is identical.
Core Principles & Definitions
Before you can run a hypothesis test or build a confidence interval, you need a small toolkit of vocabulary and ideas. Every concept below will appear repeatedly throughout the IB HL Statistics and Probability course, so getting comfortable with them now will pay off later.
Null Hypothesis (H₀)
Alternative Hypothesis (H₁)
Significance Level (α)
p-value
Confidence Interval (CI)
A few more ideas tie these together. A Type I error occurs when you reject H₀ even though it is actually true — a 'false alarm.' A Type II error occurs when you fail to reject H₀ even though H₁ is actually true — a 'missed finding.' The significance level α is exactly the probability of a Type I error that you are willing to tolerate.
Visual Explanation — The Logic of a Hypothesis Test
The diagram below shows how a hypothesis test works when we are testing a population mean using a z-test. The bell curve represents the sampling distribution of x̄ under the null hypothesis. The shaded tails are the rejection regions — if our test statistic lands there, we reject H₀.
Notice three key features. First, the curve is centred on z = 0 because we assume H₀ is true when we calculate probabilities. Second, the total shaded area equals α = 0.05, split equally between two tails for a two-tailed test. Third, the observed test statistic acts like a pin dropped onto the number line — its position determines our decision. If the pin lands in a red zone, the sample evidence is strong enough to reject H₀.
Mathematical Framework
In the IB HL course, you will primarily encounter hypothesis tests for a population mean µ when the population variance σ² is known (z-test) or unknown (t-test), and you will construct confidence intervals for µ. Below are the key formulas you need to master.
One-Tailed vs Two-Tailed Tests & the Relationship with CIs
A common source of confusion is when to use a one-tailed test versus a two-tailed test. The choice depends entirely on the alternative hypothesis. If H₁ states that µ is simply different from µ₀ (µ ≠ µ₀), you run a two-tailed test. If H₁ states a specific direction — µ > µ₀ or µ < µ₀ — you run a one-tailed test, placing the entire rejection region in one tail.
There is an elegant connection between hypothesis tests and confidence intervals. A two-tailed z-test at significance level α will reject H₀: µ = µ₀ if and only if µ₀ falls outside the corresponding (1 − α) × 100 % confidence interval. In other words, constructing a 95 % CI and checking whether µ₀ is inside it gives the same answer as running a two-tailed test at α = 0.05. This duality is worth remembering because IB problems sometimes ask you to use a CI to reach a conclusion about a hypothesis.
Worked Example — z-Test and 95 % Confidence Interval
A coffee-shop chain claims that its large cups contain an average of µ = 350 mL of coffee. A consumer group suspects the cups are under-filled. They take a random sample of n = 40 cups and find x̄ = 343 mL. The population standard deviation is known to be σ = 18 mL. Test the claim at the 5 % significance level and construct a 95 % confidence interval for the true mean.
Strengths, Limitations & Common Misconceptions
| Aspect | Strengths | Limitations / Pitfalls |
|---|---|---|
| Objectivity | Provides a structured, repeatable decision procedure rather than subjective judgement. | The choice of α is still somewhat arbitrary; 5 % is a convention, not a law of nature. |
| Quantified uncertainty | Confidence intervals give a range, acknowledging that point estimates alone can be misleading. | A 95 % CI does NOT mean there is a 95 % probability the true parameter is inside this specific interval. It means 95 % of such intervals, across repeated sampling, would contain the parameter. |
| Sample size sensitivity | Large samples produce narrow CIs and high power, making small effects detectable. | With a very large n, even trivially small deviations from µ₀ become 'statistically significant,' even if they have no practical importance. |
| Assumptions | The z-test relies on a known σ and normally distributed data (or large n by CLT), which are reasonable in many real-world settings. | If data are heavily skewed and n is small, the normal approximation breaks down, leading to unreliable p-values and CIs. |
Connection to Advanced Topics
The z-test and t-test for means are your entry point into a much larger family of hypothesis tests. As you progress in the IB HL course and beyond, you will encounter chi-squared tests for independence, tests for proportions, paired t-tests, and analysis of variance (ANOVA). The logical structure — state hypotheses, compute a test statistic, find a p-value, make a decision — remains exactly the same.
| Feature | Intro / Standard (This Lesson) | Advanced Extensions |
|---|---|---|
| Parameter tested | Population mean µ | Proportions p, variances σ², regression slopes β |
| Distribution used | Standard normal (z) or t-distribution | Chi-squared χ², F-distribution, non-parametric tests |
| Number of groups | One sample vs one hypothesised value | Two-sample tests, paired tests, multi-group ANOVA |
| Confidence intervals | For µ using z* or t* | For differences of means, proportions, odds ratios, prediction intervals |
One particularly important extension is the concept of statistical power — the probability that a test correctly rejects H₀ when H₁ is true. Power depends on α, n, and the true effect size. In research design, scientists choose n large enough to achieve a desired power (commonly 0.80), ensuring their study has a realistic chance of detecting the effect they are looking for.
Practice Problems
Lesson Summary
In this lesson you learned to perform hypothesis tests for a population mean by stating a null hypothesis (H₀) and an alternative hypothesis (H₁), computing a z- or t-test statistic, comparing it to a critical value or finding the p-value, and writing a conclusion in context. You also learned that a confidence interval provides a range of plausible values for the true mean, with wider intervals reflecting greater uncertainty.
Key ideas to retain: the significance level α sets the maximum acceptable probability of a Type I error; a two-tailed test at α corresponds to a (1 − α) × 100 % confidence interval; and statistical significance does not automatically imply practical significance. Always report your results in the context of the original problem, and remember that failing to reject H₀ is not the same as proving H₀ true.