Statistics › Section F · The normal distribution › Topic F.2

Using the normal distribution

Nobody measures a child's height in z-scores. Real quantities come with real units, their own mean and SD - and a three-step pipeline connects them to the standard-normal table, in both directions.

Real normals, real units

Standard normal values are z-scores - handy, but no scale reads in z-scores. Real quantities come with real units and their own \(\mu\) and \(\sigma\): IQ centred at 100, adult heights near 170 cm, loaf weights in grams. The previous topic built the table; this one runs the full pipeline on real normals - convert to \(z\), read the table, convert back - in both directions.

The recipe

Every "find the probability" problem on a normal \(X \sim N(\mu, \sigma)\) follows three steps. 1. Standardise each boundary: \(z = \frac{x - \mu}{\sigma}\). 2. Read the standard-normal table for the required left-area, complement or difference. 3. Report the probability. "Below a value" reads one left-area; "above" takes a complement; "between" standardises two boundaries and subtracts. Direction matters at step 1: a boundary below the mean gives a negative \(z\), and the sign must survive the journey.

Worked example: IQ scores

IQ scores are approximately normal with \(\mu = 100\) and \(\sigma = 15\). a) \(P(X > 130)\): \(z = \frac{130-100}{15} = 2\), so \(P(Z > 2) = 1 - 0.9772 = 0.0228\) - about 2.3%; the 68-95-99.7 rule's "2.5%" is the rougher estimate. b) \(P(X < 85)\): \(z = \frac{85-100}{15} = -1\), and \(P(Z < -1) = 0.1587\). c) Check: 130 sits exactly 2 SDs up, 85 exactly 1 SD down, so the tail and near-sixth sizes match the landmarks.

Worked example: between two heights

Suppose adult heights in an illustrative population are approximately normal with \(\mu = 170\) cm and \(\sigma = 8\) cm. What share falls between 162 and 178 cm? 1. Standardise both boundaries: \(z_1 = \frac{162-170}{8} = -1\) and \(z_2 = \frac{178-170}{8} = 1\). 2. Between-area:

P(Z < 1) - P(Z < -1) = 0.8413 - 0.1587 = 0.6826

3. About 68% - the empirical rule's first landmark, now computed from the table. 4. Check: symmetric boundaries at \(\pm 1\sigma\) must give the rule's 68%, and they do.

Working backwards: from area to value

Cutoff problems run the pipeline in reverse. 1. Draw the area: "the top 10%", "the lowest 5%", "the middle 95%". 2. Convert the area statement to a \(z\) value, reading the table backwards or using known landmarks: the top 10% begins at \(z \approx 1.28\) (since \(P(Z < 1.28) \approx 0.90\)); the top 5% at \(z \approx 1.645\). 3. Un-standardise:

x = \mu + z\sigma

The formula reverses \(z = \frac{x-\mu}{\sigma}\): take the position in SDs and put the units back.

Worked example: the top 10% of IQ

A scholarship programme admits students from the top 10% of IQ scores, \(N(100, 15)\). 1. The top 10% begins at the 90th percentile: \(z \approx 1.28\). 2. \(x = 100 + 1.28 \times 15 = 100 + 19.2 = 119.2\). 3. Check: 119.2 sits somewhat more than one \(\sigma\) (15 points) above the mean - sensible, because 10% is a moderate tail, not an extreme one. The cutoff is about 119.

Worked example: a bakery's tallest loaves

Loaf heights are normal with \(\mu = 11\) cm and \(\sigma = 0.8\) cm, and a display shelf fits all but the tallest 5%. 1. The top 5% begins at the 95th percentile: \(z \approx 1.645\). 2. \(x = 11 + 1.645 \times 0.8 = 11 + 1.316 = 12.3\) cm. 3. Check: 12.3 cm is about 1.3 cm above the mean, a little under two \(\sigma\) (1.6 cm). That fits: the 95% landmark leaves only 2.5% above \(2\sigma\), so the larger top 5% must begin a little below it.

Checking that the data are normal

Every calculation above assumes the bell shape - so check it. A histogram should look roughly symmetric and bell-shaped, with the mean near the median. A normal probability plot, which software draws, should show points close to a straight line. Skewed quantities - incomes, waiting times, house prices - fail these checks and should not be pushed through normal calculations directly. The assumption is not a formality: normal formulas applied to skewed data produce confidently wrong answers.

Why the normal is everywhere

What makes one curve deserve all this machinery? The central limit theorem - previewed here, developed in its own chapter - says that averages of many independent observations are approximately normal, whatever the shape of the original data. Heights and measurement errors look normal directly; even when raw data are skewed, their averages turn bell-shaped. That is why normal-based methods run through statistics: when the quantity of interest is an average of many independent values, the normal model usually fits it, even if it fits the raw data badly.

Common mistakes to avoid

First, standardising with the variance instead of the SD: \(\sigma\), not \(\sigma^2\), goes in the denominator. Second, skipping the standardising step and looking up raw x-values in the Z-table. Third, losing the sign: boundaries below the mean give negative \(z\). Fourth, reading "above" as \(P(Z < z)\) without the complement. Fifth, forgetting to un-standardise in backward problems - 1.28 is a \(z\)-score, not an IQ. Sixth, running normal calculations on visibly skewed data without a shape check first.

Key takeaways

The pipeline runs on any normal: standardise with \(z = \frac{x-\mu}{\sigma}\), read the standard-normal table, and report. Below reads a left-area, above takes a complement, between subtracts two. Backward problems reverse the pipeline - area to \(z\) to \(x = \mu + z\sigma\) - with \(z \approx 1.28\) for the top 10% and \(z \approx 1.645\) for the top 5%. IQ above 130 is \(z = 2\): 2.28% from the table, 2.5% by rule of thumb. Check the bell before trusting the arithmetic - histogram, probability plot, mean near median - and remember the central limit theorem explains why averages so often earn the assumption.

Flashcards

Flashcard

The pipeline

Standardise, look up, report

Three steps, any normal.

Test your understanding

Question 1 of 5

IQ scores are approximately N(100, 15). What is P(IQ > 130)?

Question 2 of 5

Adult heights are approximately N(170, 8). What share lies between 162 and 178 cm?

Question 3 of 5

A programme takes the top 10% of IQ scores (N(100, 15)). What is the cutoff, approximately?

Question 4 of 5

In N(11, 0.8), what x corresponds to z = 1.645?

Question 5 of 5

Which observation suggests data are NOT normal?

← Previous topicNext topic →