Introduction: Can We Trust the Numbers?

Have you ever seen a claim like "8 out of 10 cats prefer this brand" or "60% of voters support this policy" and wondered if it was actually true? In statistics, we don't just take these numbers at face value. We use Tests for a Binomial Proportion to decide if a sample of data provides enough evidence to support or challenge a claim about a whole population.

In this chapter, we will learn how to take a sample, look at the proportion of "successes," and use probability to decide if our results are just a fluke or if the original claim is likely wrong. Don't worry if this seems tricky at first—once you master the steps, it’s like following a recipe!

1. Setting the Stage: The Hypotheses

Every test starts with two competing statements about the population proportion, which we label as \( \pi \) (the Greek letter pi) or simply \( p \).

  • The Null Hypothesis \( H_0 \): This is the "status quo" or the claim we are testing. We assume this is true unless we find strong evidence against it. It always uses an equals sign.
    Example: \( H_0: \pi = 0.5 \) (The coin is fair).
  • The Alternative Hypothesis \( H_1 \): This is what we suspect might actually be true.
    • 1-tailed test: We suspect the proportion is specifically higher or lower (\( \pi > 0.5 \) or \( \pi < 0.5 \)).
    • 2-tailed test: We just suspect the proportion is different (\( \pi \neq 0.5 \)).

Quick Review: Remember that a parameter (like \( \pi \)) describes the whole population, while a statistic (like the number of successes in your sample) describes your specific data.

2. Method 1: The Exact Binomial Test

When our sample size \( n \) is relatively small, we use the Binomial Distribution directly to calculate the probability of our result occurring by chance.

Steps for an Exact Test:
  1. Define your variable: Let \( X \) be the number of successes, where \( X \sim B(n, \pi) \).
  2. Assume \( H_0 \) is true (use the value of \( \pi \) from \( H_0 \)).
  3. Calculate the p-value: This is the probability of getting a result as extreme as, or more extreme than, the one you observed.
    • If testing \( \pi > k \), find \( P(X \ge \text{observed value}) \).
    • If testing \( \pi < k \), find \( P(X \le \text{observed value}) \).
  4. Compare the p-value to the Significance Level (usually 5% or 0.05).

Memory Aid: If the p-value is low, the \( H_0 \) must go! (If \( p < 0.05 \), we reject the null hypothesis).

3. Method 2: The Normal Approximation

Calculating exact probabilities for huge samples (like \( n = 500 \)) is exhausting. When the sample is large enough, the Binomial distribution starts looking like a Normal Distribution curve.

When can we use this?

According to the Edexcel syllabus, you can use the Normal approximation when:
\( n \ge 20 \) and \( \pi \approx 0.5 \)
OR
\( n\pi > 10 \) and \( n(1 - \pi) > 10 \).

The Formula:

We use the mean \( \mu = n\pi \) and the variance \( \sigma^2 = n\pi(1-\pi) \). To find the test statistic \( z \), we use:
\( z = \frac{X - n\pi}{\sqrt{n\pi(1-\pi)}} \)

Important: The Continuity Correction

Because we are using a continuous curve (Normal) to model discrete steps (Binomial), we must adjust our value of \( X \) by 0.5.
- If you want \( P(X \ge 15) \), you actually calculate \( P(Y > 14.5) \) in the Normal model.
- If you want \( P(X \le 15) \), you calculate \( P(Y < 15.5) \).

4. Critical Regions and P-values

There are two main ways to decide if your result is "significant":

  • P-value approach: Calculate the exact probability of your result. If it's less than your significance level (e.g., 0.05), reject \( H_0 \).
  • Critical Region approach: Find the "cutoff" values (Critical Values) that mark the edge of the 5% tail. If your observed test statistic falls inside this Critical Region, reject \( H_0 \).

Did you know? In a 2-tailed test at a 5% significance level, you split the risk. You look for the extreme 2.5% at the bottom and the extreme 2.5% at the top.

5. Drawing Conclusions (The SEC Way)

In your exam, you will be assessed on the Statistical Enquiry Cycle (SEC). This means you can't just say "Reject \( H_0 \)" and stop. You must interpret the result in context.

How to write a perfect conclusion:
  1. Compare: "Since \( 0.032 < 0.05 \)..."
  2. Decide: "...we reject the null hypothesis."
  3. Contextualise: "There is sufficient evidence at the 5% level to suggest that the proportion of people who prefer coffee has increased."

Common Mistake to Avoid: Never say "This proves the proportion is 0.6." We use phrases like "suggests that" or "provides evidence for." In statistics, a conclusion is never definite because we are only looking at a sample, not the whole population.

Key Takeaways Summary

  • Hypotheses: \( H_0 \) is the claim (\( \pi = \dots \)), \( H_1 \) is the suspicion (\( \pi > \), \( \pi < \), or \( \pi \neq \)).
  • Random Sampling: A test is only valid if the data was collected via a random sample to avoid bias.
  • Exact vs. Approx: Use Binomial for small \( n \); use Normal (with continuity correction) for large \( n \).
  • Significance: If your result is very unlikely to happen by chance (p-value < significance level), it is "statistically significant."
  • The "No Certainty" Rule: Always state your conclusion as "evidence suggests" rather than "it is proven."

Note: For more on errors in testing, see the chapter on "Type I and Type II errors and power." For testing the difference between two different proportions, see "Hypothesis testing for 1 and 2 samples."