Welcome to Hypothesis Testing!
Hi there! Welcome to one of the most practical and powerful chapters in Further Statistics 1. Have you ever wondered how scientists decide if a new drug actually works, or how a factory confirms their machines aren't faulty? They use Hypothesis Testing.
In your standard A Level Maths, you learned how to test the Binomial distribution. In Further Mathematics (9FM0), we take those same ideas and apply them to two new "characters": the Poisson Distribution and the Geometric Distribution. Don't worry if it sounds intimidating—at its heart, it’s just a formal way of asking: "Is this result just a fluke, or has something actually changed?"
1. Testing the Mean of a Poisson Distribution
In a Poisson distribution, we are usually looking at the rate at which events happen over a fixed interval (like time or space). We call this rate \(\lambda\) (lambda) or sometimes \(\mu\).
Setting the Scene: The Hypotheses
Every test starts with two competing statements:
1. The Null Hypothesis (\(H_0\)): This is the "status quo." We assume the rate \(\lambda\) hasn't changed. We always write this as \(H_0: \lambda = \text{number}\).
2. The Alternative Hypothesis (\(H_1\)): This is what we suspect is actually happening. It could be that the rate has increased (\(>\)), decreased (\(<\)), or just changed (\(\neq\)).
Real-World Analogy: The Shooting Star
Imagine you usually see an average of 3 shooting stars per hour (\(\lambda = 3\)). One night, you see 8. You might think, "Wow, the rate of shooting stars has increased!" A hypothesis test helps you decide if seeing 8 was just a lucky night (the null hypothesis) or if there is actually a meteor shower happening (the alternative hypothesis).
Step-by-Step: How to Test
1. Write your hypotheses clearly using the parameter \(\lambda\).
2. Identify the distribution under the null hypothesis: \(X \sim Po(\lambda)\).
3. Calculate the probability of getting your observed value or more extreme. Use your calculator's Poisson CD function.
4. Compare to the significance level (\(\alpha\)). If your probability is smaller than \(\alpha\), it's "too weird to be a coincidence," and we reject \(H_0\).
5. Write a conclusion in plain English, relating it back to the context (e.g., "There is sufficient evidence to suggest the rate of shooting stars has increased.")
Quick Review: Remember that if you change the size of the interval (e.g., from 1 hour to 2 hours), you must adjust your \(\lambda\) accordingly before starting the test!
Key Takeaway: For Poisson tests, we are checking if the observed number of occurrences is "too high" or "too low" compared to the expected average rate.
2. Testing the Parameter \(p\) of a Geometric Distribution
The Geometric Distribution is all about "waiting time"—how many trials does it take until the first success? Here, we are testing the probability of success, \(p\).
Understanding the Logic
If the probability of success \(p\) is very high, you’d expect the first success to happen quickly. If \(p\) is very low, you’d expect to wait a long time. If you have to wait much longer than expected, maybe the true \(p\) is smaller than you thought!
Setting the Hypotheses
For a Geometric test, your hypotheses must be stated in terms of \(p\):
\(H_0: p = \text{value}\)
\(H_1: p < \text{value}\) (or \(>\) or \(\neq\))
The Formula Trick
While you can use your calculator, the Geometric distribution has a very simple formula for "waiting more than \(k\) trials":
\(P(X > k) = (1 - p)^k\)
This is because for the first success to happen after trial \(k\), you must have had failures for the first \(k\) trials. This is often the easiest way to calculate probabilities in these tests!
Common Mistake to Avoid: In a Geometric test, the "more extreme" direction can feel backwards. If you are testing if \(p\) has decreased (\(H_1: p < 0.2\)), the "extreme" result is a large value of \(X\) (taking a long time to get a success).
Key Takeaway: For Geometric tests, we use the number of trials until the first success to decide if the probability of success \(p\) is what we claim it is.
3. Significance Levels and Critical Regions
Significance Level (\(\alpha\)): This is the "threshold for weirdness." Common levels are 5% (0.05) or 1% (0.01). If the probability of our result (the p-value) is less than this, we reject the Null Hypothesis.
One-Tailed vs. Two-Tailed
1. One-Tailed: You only care if the parameter has gone in one specific direction (e.g., "Has the rate increased?"). You compare the whole 5% to your result.
2. Two-Tailed: You care if the parameter has changed at all (e.g., "Is the rate different?"). You must split your significance level in half (e.g., 2.5% at the top end and 2.5% at the bottom end).
Memory Aid: Think of a two-tailed test as a double-edged sword. You have to be careful on both sides, so you split your "danger zone" (the significance level) in two!
Did you know? A Type I error happens when you reject the null hypothesis even though it was actually true. The significance level is actually the probability of making this mistake!
4. Summary Checklist for Success
Don't worry if this seems tricky at first! Just follow this checklist for every question:
1. Define the Parameter: State clearly what \(\lambda\) or \(p\) represents in the context of the question.
2. State \(H_0\) and \(H_1\): Use the correct symbols (\(\lambda, \mu,\) or \(p\)).
3. Check the Distribution: Write down \(X \sim Po(\dots)\) or \(X \sim Geo(\dots)\).
4. Find the p-value: Calculate the probability of the observed value or more extreme.
5. Compare and Decide: Is \(p \text{-value} < \alpha\)? If yes, reject \(H_0\).
6. Contextual Conclusion: Always end with: "There is [sufficient/insufficient] evidence at the [x]% level to suggest that [context]..."
Final Tip: When testing Poisson, if the question gives you a total number of events over several intervals, you can either adjust \(\lambda\) to match the new time period or keep \(\lambda\) the same and use the "Additive Property" of Poisson to create a new distribution. Both work, but adjusting \(\lambda\) is usually simpler!
Key Takeaway: Hypothesis testing is a logical process. If you follow the steps and keep your notation tidy, you'll find these are some of the most reliable marks in the Further Statistics paper.