Statistics › Section F · The normal distribution › Topic F.1

The standard normal distribution

Heights cluster symmetrically around an average, measurement errors wander around the truth, test scores pile into a bell. The normal distribution is the curve behind all of them - and one standardised version of it serves every problem, whatever the units.

The most famous curve

Many measured quantities share one shape: values crowd near an average and thin out evenly on both sides. Repeated measurements of the same length scatter around the true value; scores on a well-designed test bunch in the middle. The normal distribution is the model for all of these - symmetric, bell-shaped, and fully described by two numbers: its centre and its spread. This topic builds the standard member of the family, the toolkit for reading its areas, and the quick-estimate rule everyone carries.

The normal curve

A normal distribution is a symmetric, bell-shaped curve, centred at its mean \(\mu\), with its spread set by the standard deviation \(\sigma\). The curve's inflection points - where it switches from steepening to flattening - sit exactly one \(\sigma\) either side of the centre, which is why \(\sigma\) is visible in the silhouette: small \(\sigma\) draws a tall, tight bell; large \(\sigma\), a low, wide one. Notation: \(X \sim N(\mu, \sigma)\) means \(X\) is normal with mean \(\mu\) and standard deviation \(\sigma\).

One table for every normal

There are infinitely many normal curves, but they all share one shape: rescale any normal's axis - shift by \(\mu\), stretch by \(\sigma\) - and you land on the same standard curve, the standard normal, \(Z \sim N(0, 1)\), with mean 0 and SD 1. That single fact is why one table of standard-normal areas serves every problem: convert to \(Z\), read the area, done. Without it, every \(\mu\) and \(\sigma\) would need its own table.

Standardising

The conversion is the z-score, already met in the spread topic as "how many SDs from the mean":

z = \frac{x - \mu}{\sigma}

Subtracting \(\mu\) centres the curve at 0; dividing by \(\sigma\) rescales the spread to 1. A height of 178 cm in \(N(170, 8)\) has \(z = \frac{178-170}{8} = 1\): one standard deviation above average. The z-score keeps the value's position while discarding its original units, which is precisely what makes one table enough.

Reading standard-normal areas

For the standard normal, areas come from a table or software, and always as left-areas: \(P(Z < z)\) is the area to the left of \(z\). Four rules turn any question into left-areas. Below 0: \(P(Z < 0) = 0.5\) by symmetry. Above a value: \(P(Z > a) = 1 - P(Z < a)\), the complement. Below a negative value: \(P(Z < -a) = 1 - P(Z < a)\), because the curve's mirror halves swap the areas. Between two values: subtract the lower left-area from the upper.

The 68-95-99.7 rule

Three landmarks let a normal be read without any table: in any normal distribution, about 68% of values lie within one standard deviation of the mean, 95% within two, and 99.7% within three. In \(N(0, 1)\): about 68% of the area sits between -1 and 1, 95% between -2 and 2, and all but 0.3% between -3 and 3. The rule powers quick sanity checks - exact enough for estimates, with the table handling precision work.

Worked example: reading the standard-normal table

a) \(P(Z < 1.96)\): the table gives 0.9750 - about 97.5% of the area lies below 1.96, leaving 0.025 in the upper tail. b) \(P(Z > 1.96) = 1 - 0.9750 = 0.025\), the tail seen from the other side. c) \(P(-1 < Z < 1.68)\): the left-areas are \(P(Z < 1.68) = 0.9535\) and \(P(Z < -1) = 0.1587\), so the between-area is \(0.9535 - 0.1587 = 0.7948\). d) Check: each piece is a left-area or a difference of two, and the interval in (c) spans most of the curve, matching 0.79's size.

Worked example: symmetry in action

a) \(P(Z < -1)\): the table lists positive values, but symmetry answers: \(P(Z < -1) = 1 - P(Z < 1) = 1 - 0.8413 = 0.1587\). b) \(P(Z > -1.5)\): everything above -1.5 is the mirror of everything below 1.5, so \(P(Z > -1.5) = P(Z < 1.5) = 0.9332\). c) Check: (a) should be small - a far-left tail - and (b) large, most of the curve; both came from one table column plus the mirror trick. Symmetry means the standard normal table never needs negative entries.

Common mistakes to avoid

First, using \(\sigma^2\) in the z formula - the denominator is the standard deviation, not the variance. Second, forgetting to standardise before touching the table: it serves \(Z\), not \(X\). Third, reading a right-tail request as a left-area: "above" needs the complement. Fourth, mishandling negative z-values - mirror, then complement, in that order. Fifth, applying the 68-95-99.7 landmarks to non-normal or skewed data. Sixth, confusing \(P(Z < 1.96) = 0.975\) with the 0.025 tail - the two numbers answer different questions.

Key takeaways

Normal curves are symmetric bells defined by \(\mu\) and \(\sigma\), with inflection points one \(\sigma\) from the centre. The standard normal \(Z \sim N(0,1)\) is the family's common shape: every normal converts to it via \(z = \frac{x-\mu}{\sigma}\), so one table serves all. Read areas as left-areas; above-questions take complements; negative values mirror; between-questions subtract. The 68-95-99.7 rule gives the landmarks: 1, 2 and 3 SDs hold about 68%, 95% and 99.7%. Worth memorising: \(P(Z < 1.96) = 0.975\), so 2.5% sits above \(z = 1.96\).

Flashcards

Flashcard

Normal curve

Symmetric bell, centre μ, spread σ

Inflection points one σ from the centre.

Test your understanding

Question 1 of 5

For the standard normal distribution, what is P(Z < 0)?

Question 2 of 5

X ~ N(170, 8). What is the z-score of 178?

Question 3 of 5

In a standard normal distribution, about what percent of the area lies between -2 and 2?

Question 4 of 5

What is P(Z > 1.96)?

Question 5 of 5

Using the table (P(Z < 1) = 0.8413), what is P(Z < -1)?

← Previous topicNext topic →