Statistical hypothesis testing

Welcome to Statistical Hypothesis Testing!

Ever wondered how scientists "prove" a new medicine works, or how a factory knows if its machines are under-filling cereal boxes? They use statistical hypothesis testing. Think of it as a formal, mathematical way of being a detective. We start with a "boring" assumption (the status quo) and see if our data is weird enough to suggest something has actually changed.

Don't worry if this seems tricky at first! We are going to break down the "stat-speak" into plain English and show you the step-by-step recipe for passing these questions in Paper 3.

1. The Language of the "Courtroom"

In the UK legal system, someone is "innocent until proven guilty." Hypothesis testing works exactly the same way. We have two competing ideas:

The Null Hypothesis (\(H_0\)): This is the "innocent" or "nothing has changed" position. We assume this is true unless we have very strong evidence otherwise. It always involves an "equals" sign (e.g., \(p = 0.5\) or \(\mu = 10\)).

The Alternative Hypothesis (\(H_1\)): This is the "guilty" or "something is up" position. This is what you, the researcher, suspect might be happening. It uses symbols like \(<\), \(>\), or \(\neq\).

Key Terms You Need to Know:

• Test Statistic: This is the actual value you get from your sample (like the number of people who liked a product).
• Significance Level (\(\alpha\)): The "threshold" for evidence. Usually 5% (0.05). It’s the probability of being wrong if we decide to reject \(H_0\).
• Critical Region: The "rejection zone." If your test statistic falls in here, the result is so unlikely to happen by chance that we reject the null hypothesis.
• Critical Value: The border line between the "acceptance region" and the "critical region."
• p-value: The probability of getting your result (or something more extreme) if \(H_0\) is actually true.

Memory Aid: "The p-value Rule"
If the p is low (less than the significance level), then \(H_0\) must go!

Summary Takeaway:

Hypothesis testing is about deciding between \(H_0\) (no change) and \(H_1\) (change) based on how unlikely our data is under the assumption of \(H_0\).

2. Testing Proportions (Binomial Distribution)

We use this when we are dealing with "success or failure" scenarios—like the probability of a coin landing on heads or the percentage of voters supporting a candidate.

Step-by-Step Process:

1. State your hypotheses: \(H_0: p = \dots\) and \(H_1: p <, >, \text{ or } \neq \dots\)

2. Define the distribution: Assume \(X \sim B(n, p)\) using the values from \(H_0\).

3. Find the p-value: Calculate the probability of getting your result or more extreme.
Example: If you suspect a coin is biased toward heads and you get 8 heads out of 10, find \(P(X \geq 8)\).

4. Compare: If your p-value \( < \) significance level, reject \(H_0\).

5. Conclude in context: Always write a sentence like: "There is significant evidence at the 5% level to suggest that the proportion of... has increased."

1-tail vs 2-tail Tests:

• 1-tail: You suspect a change in one specific direction (e.g., "The drug is better than the old one").
• 2-tail: You just think it's different (e.g., "The machine is no longer accurate"). Crucial Tip: In a 2-tail test, you must split your significance level in half (e.g., 2.5% at the top and 2.5% at the bottom).

Did you know? The significance level is actually the probability of making a "Type I error"—which means rejecting the Null Hypothesis when it was actually true. It's the risk we take of being a "false alarm."

Quick Review:

For Binomial tests, use your calculator's Binomial Cumulative Distribution (BCD) function. Always check if the question asks for "at least" or "more than" to get your inequality right!

3. Testing the Mean (Normal Distribution)

We use this when we are measuring something continuous, like weights, heights, or times. For this section of Paper 3, we assume we know the population variance (\(\sigma^2\)).

The "Big Change": The Sample Mean

When we take a sample of size \(n\), the Sample Mean (\(\bar{X}\)) follows a specific distribution:
\(\bar{X} \sim N(\mu, \frac{\sigma^2}{n})\)

Don't forget: You must divide the variance by the sample size \(n\). This makes sense—bigger samples lead to more consistent, less spread-out averages!

Common Mistake to Avoid:

Students often forget to use the "Standard Error" \(\frac{\sigma}{\sqrt{n}}\) on their calculator instead of just \(\sigma\). If you forget the \(\sqrt{n}\), your whole test will be wrong!

Summary Takeaway:

Testing the mean is almost identical to the Binomial process, but you use the Normal Distribution and your test statistic is the average (\(\bar{x}\)) of your sample.

4. Correlation Hypothesis Testing

Sometimes we want to know if two things are related (like hours spent studying and exam scores). We use the Product Moment Correlation Coefficient (PMCC), denoted by \(r\).

The population correlation is represented by the Greek letter rho (\(\rho\)).

• \(H_0: \rho = 0\) (There is no correlation).
• \(H_1: \rho > 0, \rho < 0, \text{ or } \rho \neq 0\).

You don't need to calculate \(r\) by hand (your calculator does that!), but you do need to compare your calculated \(r\) value to a Critical Value from a table provided in the exam or use a p-value. If your \(r\) is further from zero than the critical value, you've found a real relationship!

Analogy: Imagine trying to hear a whisper in a noisy room. The correlation is the "whisper" (the signal) and the random variation is the "noise." Hypothesis testing helps us decide if the whisper is real or if we're just imagining it in the noise.

Quick Review:

Correlation does NOT mean causation. Even if you reject \(H_0\) and find a strong correlation, it doesn't mean one thing caused the other—they might just be linked by something else!

5. Final Tips for Exam Success

To get full marks on Paper 3, follow these "golden rules":

• Always define your parameter. Don't just write \(p\); write "\(p\) is the probability that a seed germinates."
• Don't be too certain: Never say "This proves the null hypothesis is true." Instead, say "There is insufficient evidence to reject the null hypothesis."
• Context is King: Your final sentence must mention the actual situation (the seeds, the coins, the weights, etc.). Marks are often lost for being too "mathsy" at the very end.

Don't worry if this feels like a lot to juggle. With practice, the steps become second nature. You're just checking if the data is "weird enough" to be interesting!