Normal Distribution Extensions — AHL 4.9 Normal distribution applications and approximations (HL extension) - IB Mathematics Analysis and Approaches

SECTION 1

Historical Context & Motivation

The normal distribution is often called the most important probability distribution in all of statistics. Its familiar bell-shaped curve appears everywhere — from test scores and human heights to measurement errors in physics experiments. But the story of how mathematicians realized that this single curve could stand in for many other distributions is a fascinating journey that stretches over three centuries.

Early probability theory focused on discrete distributions like the binomial distribution, which counts successes in repeated trials. As scientists tackled problems involving hundreds or thousands of trials, computing binomial probabilities by hand became impractical. Mathematicians began searching for a continuous curve that could approximate these tedious discrete calculations, eventually discovering that the normal distribution was the key.

1733 • De Moivre's Approximation

Abraham de Moivre showed that the binomial distribution for a fair coin could be approximated by a smooth bell curve as the number of trials grew large — the first normal approximation.

1809 • Gauss and Error Theory

Carl Friedrich Gauss used the normal distribution to model measurement errors in astronomy, showing that the bell curve naturally arises when many small, independent errors combine.

1812 • Laplace's Central Limit Theorem

Pierre-Simon Laplace proved a version of the Central Limit Theorem, establishing that sums of many independent random variables tend toward a normal distribution regardless of the original distribution.

1901 • Pearson's Chi-Squared and Approximations

Karl Pearson developed goodness-of-fit tests and further formalized how the normal distribution could approximate the Poisson and other distributions for practical statistical inference.

1920s– • Modern Statistical Framework

Ronald Fisher and others built the modern framework of hypothesis testing and confidence intervals, relying heavily on normal approximations that remain central to IB Mathematics HL today.

The central question this lesson addresses is: When and how can we use the normal distribution to approximate discrete distributions like the binomial and Poisson? Understanding these approximations allows you to solve problems that would otherwise require impossibly long calculations, and it reveals a deep mathematical unity underlying seemingly different probability models.

SECTION 2

Core Principles & Definitions

Before diving into approximations, let's establish the foundational ideas. The normal distribution is a continuous probability distribution defined by two parameters: the mean μ (which locates the center) and the standard deviation σ (which controls the spread). In this HL extension, you will learn to bridge continuous and discrete worlds using specific conditions and corrections.

Normal Approximation to Binomial

When n is large and p is not too close to 0 or 1, the binomial distribution B(n, p) can be approximated by N(np, np(1 − p)). The rule of thumb is np ≥ 5 and n(1 − p) ≥ 5.

Normal Approximation to Poisson

When the Poisson parameter λ is large (typically λ ≥ 15), the Poisson distribution Po(λ) can be approximated by N(λ, λ), since for a Poisson variable the mean and variance both equal λ.

Continuity Correction

Because we are using a continuous distribution to approximate a discrete one, we adjust boundaries by ±0.5. For example, P(X ≤ 7) becomes P(Y ≤ 7.5). This adjustment is called the continuity correction and improves accuracy.

Linear Combinations of Normals

If X ~ N(μ₁, σ₁²) and Y ~ N(μ₂, σ₂²) are independent, then aX + bY ~ N(aμ₁ + bμ₂, a²σ₁² + b²σ₂²). This allows you to model sums, differences, and weighted averages of normal variables.

Standardization (Z-scores)

Any normal variable X ~ N(μ, σ²) can be converted to the standard normal Z ~ N(0, 1) using Z = (X − μ)/σ. This transforms every normal problem into a lookup or GDC computation on a single universal table.

✦

KEY TAKEAWAY

Think of the normal distribution as a universal translator. Just as a translator lets you convert between different languages, the normal curve lets you convert difficult discrete probability calculations into simple continuous ones. The continuity correction is like adjusting for the slight differences in grammar between the two 'languages' — it accounts for the fact that discrete values are whole numbers while the normal curve is smooth.

SECTION 3

Visual Explanation — The Approximation in Action

The diagram below shows a binomial distribution B(20, 0.4) represented as bars alongside its normal approximation N(8, 4.8). Notice how the smooth curve closely follows the tops of the bars. The shaded region demonstrates the continuity correction: to find P(X ≤ 10) using the normal curve, we compute P(Y ≤ 10.5) rather than P(Y ≤ 10), capturing the full width of the bar at X = 10.

The violet bars show the exact binomial probabilities for B(20, 0.4). The cyan curve is the normal approximation N(8, 4.8). The shaded area under the curve up to x = 10.5 (pink dashed line) represents the continuity-corrected approximation of P(X ≤ 10).

As you can see in the diagram, the normal curve fits the binomial bars remarkably well when n = 20 and p = 0.4. The continuity correction of +0.5 ensures we capture the full area corresponding to the bar at X = 10, rather than cutting through its middle. Without this correction, the approximation would systematically underestimate the true probability.

SECTION 4

Mathematical Framework

The mathematical machinery behind these approximations rests on matching the mean and variance of the discrete distribution to a normal distribution, then applying the continuity correction to adjust for the discrete-to-continuous gap. Below are the key formulas you need for the IB HL exam.

NORMAL APPROXIMATION TO BINOMIAL

X ~ B(n, p) ⟹ X ≈ Y ~ N(np, np(1 − p))

Valid when np ≥ 5 and n(1 − p) ≥ 5. Here μ = np and σ² = np(1 − p).

NORMAL APPROXIMATION TO POISSON

X ~ Po(λ) ⟹ X ≈ Y ~ N(λ, λ)

Valid when λ ≥ 15 (some texts use λ ≥ 10). Here μ = λ and σ² = λ, so σ = √λ.

CONTINUITY CORRECTION RULES

P(X ≤ k) ≈ P(Y ≤ k + 0.5), P(X ≥ k) ≈ P(Y ≥ k − 0.5), P(X = k) ≈ P(k − 0.5 ≤ Y ≤ k + 0.5)

When converting from discrete X to continuous Y, expand the boundary by 0.5 in the direction that includes the value. For P(X = k), use a 1-unit-wide interval centered on k.

LINEAR COMBINATION OF INDEPENDENT NORMALS

If X ~ N(μ₁, σ₁²) and Y ~ N(μ₂, σ₂²) are independent, then aX + bY ~ N(aμ₁ + bμ₂, a²σ₁² + b²σ₂²)

This extends to sums of many independent normals. Note: for X − Y, use a = 1 and b = −1, giving variance σ₁² + σ₂² (variances always add, even for differences).

⚠

Common Exam Trap

When finding the variance of X − Y, students often subtract variances. Remember: variances always add for independent variables, whether you are adding or subtracting. Var(X − Y) = Var(X) + Var(Y), not Var(X) − Var(Y).

SECTION 5

Continuity Correction — A Detailed Breakdown

The continuity correction is the most commonly misapplied step in normal approximation problems. Since discrete distributions assign probability to individual integers while the normal curve assigns probability to intervals, we need a systematic way to translate between the two. The table below shows exactly how to apply the correction for every type of inequality you will encounter.

Discrete Probability	Continuity-Corrected Normal	Why?
P(X ≤ k)	P(Y ≤ k + 0.5)	Include the full bar at X = k
P(X < k)	P(Y ≤ k − 0.5)	Exclude the bar at X = k
P(X ≥ k)	P(Y ≥ k − 0.5)	Include the full bar at X = k
P(X > k)	P(Y ≥ k + 0.5)	Exclude the bar at X = k
P(X = k)	P(k − 0.5 ≤ Y ≤ k + 0.5)	Capture just the single bar's width
P(a ≤ X ≤ b)	P(a − 0.5 ≤ Y ≤ b + 0.5)	Include full bars at both endpoints

Continuity correction rules for all inequality types

To approximate P(X = 8) for a discrete variable, we find the area under the normal curve between 7.5 and 8.5 (the pink dashed boundaries). The amber-highlighted bar represents the single discrete value X = 8, and the cyan shading under the curve captures the equivalent continuous probability.

💡

Memory Trick

Ask yourself: "Am I including or excluding this integer?" If including, extend the boundary by 0.5 to wrap around the integer. If excluding, pull the boundary by 0.5 to avoid the integer. The correction always moves in the direction that matches whether you want the bar in or out.

SECTION 6

Worked Example — Normal Approximation to Binomial

A factory produces light bulbs with a 15% defect rate. A quality inspector tests a random sample of 80 bulbs. Using a normal approximation with continuity correction, find the probability that at most 10 bulbs are defective.

Normal Approximation to B(80, 0.15)

1
Step 1 — Identify the Distribution

Each bulb is either defective or not, with p = 0.15 and n = 80 independent trials. So X ~ B(80, 0.15), where X counts defective bulbs.

X ~ B(80, 0.15)
2
Step 2 — Check Validity Conditions

We need np ≥ 5 and n(1 − p) ≥ 5. Here np = 80 × 0.15 = 12 ≥ 5 ✓ and n(1 − p) = 80 × 0.85 = 68 ≥ 5 ✓. The normal approximation is valid.

np = 12 ✓, n(1−p) = 68 ✓
3
Step 3 — Calculate Mean and Standard Deviation

μ = np = 80 × 0.15 = 12. σ² = np(1 − p) = 80 × 0.15 × 0.85 = 10.2, so σ = √10.2 ≈ 3.194. We approximate X with Y ~ N(12, 10.2).

μ = 12, σ ≈ 3.194
4
Step 4 — Apply Continuity Correction

We want P(X ≤ 10). Since we are including X = 10, we use P(Y ≤ 10.5).

P(X ≤ 10) ≈ P(Y ≤ 10.5)
5
Step 5 — Standardize and Compute

Z = (10.5 − 12) / 3.194 = −1.5 / 3.194 ≈ −0.4696. Using the GDC or z-table: P(Z ≤ −0.4696) ≈ 0.3193.

P(X ≤ 10) ≈ 0.319
6
Step 6 — Interpret the Result

There is approximately a 31.9% chance that at most 10 out of 80 bulbs are defective. For comparison, the exact binomial answer (computed on a GDC) gives P(X ≤ 10) ≈ 0.3209, so our approximation is very close.

Approximation error: about 0.5 percentage points

SECTION 7

Strengths, Limitations & When to Use Each Approximation

Not every problem benefits from a normal approximation. Understanding when the approximation works well — and when it doesn't — is crucial for choosing the right approach on the IB exam and in real applications.

Feature	Normal Approx. to Binomial	Normal Approx. to Poisson
Condition	np ≥ 5 and n(1 − p) ≥ 5	λ ≥ 15 (preferably λ ≥ 20)
Mean	μ = np	μ = λ
Variance	σ² = np(1 − p)	σ² = λ
Strength	Excellent for large n with moderate p; avoids enormous factorial calculations	Very useful for large-count rare events; simplifies cumulative Poisson calculations
Limitation	Poor when p is near 0 or 1 (skewed distribution); loses accuracy at tails	Poisson is right-skewed for small λ; approximation is poor below λ ≈ 15
Continuity correction	Always recommended for better accuracy	Always recommended for better accuracy

Comparison of normal approximation conditions

✦

KEY TAKEAWAY

Think of the normal approximation like using a GPS route that follows straight roads instead of winding country lanes. When the terrain is fairly flat and symmetric (large n with moderate p, or large λ), the straight GPS route matches the actual path very closely. But if the terrain is lopsided (p near 0 or 1, or small λ), the GPS route diverges from reality and you should use the exact distribution or a different approximation instead.

SECTION 8

Connection to the Central Limit Theorem

Everything in this lesson connects to one of the deepest results in probability theory: the Central Limit Theorem (CLT). The CLT states that the sum (or average) of a large number of independent, identically distributed random variables is approximately normally distributed, regardless of the original distribution. The binomial approximation is a special case: a binomial variable is the sum of n independent Bernoulli trials, so for large n the CLT guarantees normality.

Concept	This Lesson (AHL 4.9)	Advanced Theory (CLT)
What is approximated?	Binomial or Poisson probabilities	Distribution of sample means X̄ from any population
Conditions	np ≥ 5 and n(1−p) ≥ 5; or λ ≥ 15	n ≥ 30 (rule of thumb) for arbitrary distributions
Parameters	μ and σ² matched from original distribution	X̄ ~ N(μ, σ²/n) where μ and σ are population parameters
Applications	Quick probability calculations for discrete distributions	Confidence intervals, hypothesis testing, inferential statistics

From specific approximations to the general Central Limit Theorem

As you continue into university-level statistics, you will find that the CLT underpins virtually all of inferential statistics — from confidence intervals to hypothesis tests. The normal approximations you are learning now in AHL 4.9 are your first practical encounter with this powerful theorem. Mastering the mechanics here — checking conditions, computing parameters, applying continuity corrections — builds the intuition you will need for more advanced work.

ℹ

Looking Ahead

In the IB exam, you may also encounter problems combining linear combinations of normals with approximation techniques. For example, if X₁ and X₂ are independent binomials that each meet the approximation conditions, you can approximate each normally and then use the linear combination rule to find the distribution of X₁ + X₂ or X₁ − X₂.

SECTION 9

Practice Problems

Question 1: CONCEPTUAL

A random variable X follows a binomial distribution B(50, 0.02). A student proposes using a normal approximation to find P(X ≤ 3). Explain why this approximation would be inappropriate and suggest a better alternative.

Show answer

The normal approximation requires np ≥ 5 and n(1 − p) ≥ 5. Here np = 50 × 0.02 = 1, which is far below 5. The binomial distribution is highly right-skewed when p is this small, so the symmetric normal curve would be a poor fit. A better alternative is the Poisson approximation with λ = np = 1, since n is large and p is small. You could then compute P(X ≤ 3) directly using the Poisson formula, which gives approximately 0.981.

Question 2: BASIC CALCULATION

Emails arrive at a server at a rate of 20 per hour, following a Poisson distribution. Using a normal approximation with continuity correction, find the probability that in a given hour, the server receives fewer than 15 emails.

Show answer

Since λ = 20 ≥ 15, we can approximate Po(20) with N(20, 20), giving σ = √20 ≈ 4.472. We want P(X < 15) = P(X ≤ 14). With continuity correction: P(Y ≤ 14.5). Standardize: Z = (14.5 − 20)/4.472 = −5.5/4.472 ≈ −1.230. From the standard normal table or GDC: P(Z ≤ −1.230) ≈ 0.109. There is approximately an 10.9% chance of receiving fewer than 15 emails in the hour.

Question 3: INTERMEDIATE

A biased coin has P(heads) = 0.35. It is flipped 200 times. Using a suitable normal approximation with continuity correction, find the probability that the number of heads is between 60 and 80 inclusive.

Show answer

X ~ B(200, 0.35). Check conditions: np = 70 ≥ 5 ✓ and n(1−p) = 130 ≥ 5 ✓. Parameters: μ = 70, σ² = 200 × 0.35 × 0.65 = 45.5, σ ≈ 6.745. We need P(60 ≤ X ≤ 80). With continuity correction: P(59.5 ≤ Y ≤ 80.5). Standardize: Z₁ = (59.5 − 70)/6.745 ≈ −1.557 and Z₂ = (80.5 − 70)/6.745 ≈ 1.557. So P(−1.557 ≤ Z ≤ 1.557) = Φ(1.557) − Φ(−1.557) ≈ 0.9403 − 0.0597 ≈ 0.881. There is approximately an 88.1% probability that the number of heads falls between 60 and 80 inclusive.

Question 4: APPLIED

Two machines in a factory operate independently. Machine A produces items with a defect rate of 12% and outputs 100 items per day. Machine B produces items with a defect rate of 8% and outputs 150 items per day. Let

D_A

and

D_B

denote the number of defectives from each machine. Using normal approximations, find the probability that Machine A produces more defectives than Machine B on a given day.

Show answer

D_A

~ B(100, 0.12) ≈ N(12, 10.56) and

D_B

~ B(150, 0.08) ≈ N(12, 11.04). Both satisfy np ≥ 5 conditions. Let W =

D_A

−

D_B

. Since

D_A

and

D_B

are independent normals, W ~ N(12 − 12, 10.56 + 11.04) = N(0, 21.6), so σ_W = √21.6 ≈ 4.648. We want P(

D_A

D_B

) = P(W > 0). With continuity correction (since W involves differences of integers): P(W ≥ 0.5) ≈ P(Z ≥ 0.5/4.648) = P(Z ≥ 0.1076) ≈ 1 − 0.5428 ≈ 0.457. There is approximately a 45.7% probability that Machine A produces more defectives than Machine B.

Question 5: CRITICAL THINKING

A student uses the normal approximation to estimate P(X = k) for a binomial variable X ~ B(n, p) and obtains an answer of 0.045. They then compute the exact binomial probability and get 0.061. Discuss three possible reasons for this discrepancy. Under what circumstances would you expect the discrepancy to decrease, and why?

Show answer

Three possible reasons: (1) The value of n may not be large enough for the approximation conditions to hold well, especially if np or n(1−p) is only marginally above 5. (2) The value k may be in the tail of the distribution, where the normal curve decays as

e^{−z²/2}

but the binomial can have a different tail behavior, causing larger approximation errors. (3) The student may have failed to apply the continuity correction, using P(Y = k) instead of P(k − 0.5 ≤ Y ≤ k + 0.5), which would significantly underestimate the probability since a continuous distribution assigns zero probability to a single point. The discrepancy would decrease as n increases (with p held constant and away from 0 and 1), because the Central Limit Theorem guarantees that the standardized binomial converges to the standard normal as n → ∞. The convergence is fastest when p ≈ 0.5, where the binomial is already symmetric.

SUMMARY

Lesson Summary

Summary

The normal distribution serves as a powerful approximation tool in IB Mathematics HL. For a binomial distribution B(n, p), the approximation N(np, np(1 − p)) is valid when np ≥ 5 and n(1 − p) ≥ 5. For a Poisson distribution Po(λ), use N(λ, λ) when λ ≥ 15. In both cases, the continuity correction (adjusting boundaries by ±0.5) is essential for accurate results because it bridges the gap between discrete bars and a smooth curve.

The linear combination rule lets you add or subtract independent normal variables: aX + bY ~ N(aμ₁ + bμ₂, a²σ₁² + b²σ₂²), remembering that variances always add even for differences. These techniques connect directly to the Central Limit Theorem, the fundamental result explaining why the normal distribution appears so universally in statistics. For exam success, always state the approximation being used, verify the validity conditions, show the continuity correction, standardize to Z, and interpret your final answer in context.