Modelling with probability

Introduction: Why Model with Probability?

Welcome to one of the most practical parts of your A Level Statistics course! Modelling is the process of taking a messy, real-world situation and simplifying it into a mathematical framework. In this chapter, we focus on how to choose the right "tool" (like the Binomial or Normal distribution) to represent a real-life scenario.

Think of a mathematical model like a map. A map isn't the actual ground—it's a simplified version that helps you navigate. If the map is too simple, you’ll get lost; if it’s too complicated, you can’t read it. Learning to "model with probability" is about finding that perfect balance!

1. What Makes a Good Model?

A statistical model is a simplified description of a real-world process. We use them to calculate probabilities and make predictions without having to test every single possibility in real life.

To create a model, we have to make assumptions. An assumption is something we "pretend" is true to make the math work. For example, when flipping a coin, we assume it is "fair" and that one flip doesn't affect the next.

Quick Review: Discrete vs. Continuous

Before choosing a model, check your data type:

• Discrete Data: Things you count (e.g., number of students, goals scored). We usually use the Binomial Distribution here.
• Continuous Data: Things you measure (e.g., height, time, weight). We usually use the Normal Distribution here.

Key Takeaway: A model is only as good as its assumptions. If your assumptions are wrong, your predictions will be too!

2. Critiquing the Binomial Model

The Binomial Distribution \( B(n, p) \) is a classic model for discrete data. But you can't just use it for everything! To use it, the situation must pass the "BINS" test.

The BINS Test

• B – Binary: Are there only two outcomes? (Success or Failure).
• I – Independent: Does one trial have no effect on the next?
• N – Number: Is there a fixed number of trials (\( n \))?
• S – Success: Is the probability of success (\( p \)) the same every single time?

When the Binomial Model Fails (Real-World Examples)

Don't worry if this seems tricky at first; identifying flaws is a skill you'll get better at with practice! Here are common reasons why a Binomial model might be inappropriate:

Example 1: Predicting if it will rain every day next week.
The Flaw: Independence. If it rains today, there is a higher chance it will rain tomorrow because of a lingering low-pressure system. The trials are not independent.

Example 2: Shooting basketball hoops until you score 5 times.
The Flaw: Fixed number of trials. Here, the number of trials (\( n \)) isn't fixed; you keep going until you hit a target. This fails the "N" in BINS.

Example 3: A student's performance on a 10-question test.
The Flaw: Constant probability. As the student gets tired or hits harder questions at the end, the probability of getting a question right (\( p \)) might change. This fails the "S" in BINS.

Did you know? In biology, the Binomial model is often used to predict the gender of offspring, but even there, scientists debate if the probability of a "male" is truly constant across all families!

3. Critiquing the Normal Model

The Normal Distribution \( X \sim N(\mu, \sigma^2) \) is the "Bell Curve." It is the go-to model for continuous data like heights or exam scores.

When to Use the Normal Model

• The data is continuous.
• The distribution is symmetrical (it looks the same on both sides of the mean).
• Most data points are near the mean, with very few far away (the "tails").

When the Normal Model Fails

Sometimes the "Bell Curve" just doesn't fit the reality of the data:

• Skewness: If your data has a "long tail" on one side (e.g., house prices or incomes, where a few billionaires pull the average up), a Normal model will be misleading.
• Discrete Data: The Normal distribution is for continuous measurements. If you try to model the "number of children in a family" using a Normal curve, you might get a result like "2.4 children," which is impossible in real life!
• Outliers: If there are extreme values that happen more often than the model predicts (like "Black Swan" events in the stock market), the Normal model is too simple.

Memory Aid: Think of the Normal curve as a mountain. If your data looks like a cliff (all on one side) or a flat plain, the Normal model isn't the right map!

4. Refining the Model: Making it Better

If you realize your model is "not appropriate," you have two choices: refine the assumptions or change the model.

Common Refinements

1. Continuity Correction: If you use a continuous model (Normal) to approximate a discrete one (Binomial), you must adjust. For example, if you want "at least 5" (\( X \ge 5 \)), you actually calculate \( P(X > 4.5) \) in the Normal model.
2. Segmenting: If \( p \) isn't constant (like weather), you might model the morning and afternoon separately to keep the probability more stable within each group.

Evaluating the "Likely Effect"

The exam often asks: "What is the effect of this assumption being wrong?"

• If you assume independence but trials are actually linked, your model will usually underestimate the chance of extreme streaks (like 10 wins in a row).
• If you assume constant probability but it's actually decreasing, your model will overestimate the total number of successes.

Summary Table: Choosing Your Model

Quick Review Box:

Feature	Binomial \( B(n, p) \)	Normal \( N(\mu, \sigma^2) \)
Data Type	Discrete (Counting)	Continuous (Measuring)
Shape	Can be skewed	Must be Symmetrical
Key Assumption	Independence & Constant \( p \)	Bell-shaped & No outliers

Common Mistakes to Avoid

• Mistake: Forgetting to check the "BINS" criteria before doing a Binomial calculation. Always mention why it fits the context.
• Mistake: Using a Normal distribution for data that is clearly skewed (like the time people spend on social media).
• Mistake: Assuming that because a sample is large, it must be Normal. Size doesn't fix a bad shape!

Final Encouragement: Modelling can feel a bit "vague" compared to solving equations, but it’s where the real power of math lies. Don't be afraid to critique a model—in the real world, the best statisticians are the ones who know exactly where their models might break!