Statistics · Eqora guide

Normal distribution in math: which tail counts?

Choose the correct normal distribution area. Standardize z-scores, calculate tails and intervals, and find percentiles with checked examples.

Woman in a red jacket placing a marker on a cord in snowy mountains

A measurement lies above the mean, so its z-score is positive. Does that make the answer a large probability or a small one? Both are possible: the area below that measurement is large, while the area above it is small. Normal distribution questions become much easier when you decide which event you need before looking up a number.

This guide develops a complete paper-and-calculator method using an invented packaging example. You will standardize boundaries, read cumulative probabilities, handle intervals and work backward from a percentile. The photographs are conceptual comparisons of centers, boundaries and regions; they are not scaled probability diagrams or evidence about real manufacturing.

Start with a model and an event

Suppose the mass X of a randomly selected package is modeled by a normal distribution with mean μ = 500 g and standard deviation σ = 8 g. These are assumed model parameters, not measurements gathered for this article. Write them with their units before doing any arithmetic. The normal model has positive spread, symmetry around its mean and total probability one.

A common notation is X ∼ N(500, 64), where the second parameter is the variance in squared grams. Some courses instead put the standard deviation in the second slot. Read the convention explicitly: in our calculation σ is eight, not sixty-four. If your exercise gives a variance, take its positive square root first. The guide to standard deviation explains the distinction between a spread and its square.

Now translate the question into an event. “More than 512 g” means X > 512. “At most 492 g” means X ≤ 492. “Between 488 and 512 g” identifies a bounded interval. Sketch the mean and the boundaries, then shade the requested region. You do not need a perfect bell drawing; you need a reliable record of which side you intend to count.

Turn a boundary into a signed distance

For a boundary x, calculate z = (x − μ)/σ. Subtract the center first, then divide the entire difference by the standard deviation. RMIT’s explanation of normal probabilities describes this conversion to a common scale and the interpretation of probabilities as areas. We will apply that idea to our own numbers.

At x = 512, z = (512 − 500)/8 = 12/8 = 1.5. The boundary is one and a half standard deviations above the mean. At x = 492, z = −8/8 = −1. A negative result describes a position below the center; it does not describe a negative probability. At the center itself, z is zero.

The units cancel: grams divided by grams gives a dimensionless distance. Converting everything into kilograms gives exactly the same z-score. Mixing a value in kilograms with a mean in grams does not. Preserve parentheses in a calculator entry, because 512 − 500/8 is a different expression. Since σ is positive, standardizing preserves the direction of an inequality. X > 512 becomes Z > 1.5 without reversing the sign.

Gloved hands pointing to a washer at the center of a cord on a metal table

Check what your normal table actually reports

Let Φ(z) mean P(Z ≤ z) for a standard normal random variable Z. This is the cumulative area from the far left up to z. It increases from zero toward one, with Φ(0) = 0.5. In a cumulative table, Φ(1.5) is approximately 0.9332. Read that as “about 93.32% below the boundary,” not “a probability of 1.5.”

Not every printed table uses that convention. Some give the area between zero and a positive z; others give a right tail. Inspect the heading and a known entry before using any value. At z = 0, a cumulative table shows 0.5, while a center-to-boundary table shows zero. A right-tail table also shows 0.5 there, but its entries decrease as positive z increases. Check both clues.

For example, the area from zero to 1.5 is approximately 0.4332. Adding the left half produces 0.9332. A right-tail table would directly show about 0.0668. These are three descriptions of the same partition. The conversion must match the table rather than an imagined universal layout. Keep the definition of the returned quantity beside your calculation, especially when switching between a worksheet table and a calculator.

Count the upper tail by taking the complement

Our first question asks for P(X > 512). Standardization gives P(Z > 1.5). The cumulative value includes everything below 1.5, so subtract it from the total: P(X > 512) = 1 − Φ(1.5) ≈ 1 − 0.9332 = 0.0668. That is about 6.68%, conditional on the assumed model.

The size check matters as much as the arithmetic. The boundary is above the mean, so its upper tail must be smaller than one half. An answer of 93.32% would contradict the shaded sketch. A probability of −0.0668 would contradict the range from zero to one. Neither mistake is fixed by attaching a percent sign.

For the lower boundary 492 g, use symmetry: Φ(−1) = 1 − Φ(1) ≈ 1 − 0.8413 = 0.1587. Thus about 15.87% lies below 492 g. In the continuous normal model, the probability at one exact point is zero, so < and ≤ give the same result here. A rounded display reading “492 g” represents a measurement interval, however. Do not use the zero-point result for a whole rounding bin.

Woman holding a red strip beyond the right cord support on a snowy ledge

Subtract cumulative areas for an interval

For P(488 ≤ X ≤ 512), standardize both ends separately. The lower end becomes −1.5 and the upper end becomes 1.5. The probability is Φ(1.5) − Φ(−1.5). Using symmetry, Φ(−1.5) ≈ 0.0668, so the interval probability is approximately 0.9332 − 0.0668 = 0.8664, or 86.64%.

Notice why subtraction works: the larger cumulative area contains the entire smaller one plus the desired middle region. Subtracting removes the unwanted left portion. Adding the two cumulative areas would count overlapping regions and, in this symmetric example, give one. That answer would incorrectly leave no probability outside the interval.

An asymmetric question follows the same method. Between 492 and 516 g, the boundaries become −1 and 2. Therefore P(492 ≤ X ≤ 516) ≈ 0.9772 − 0.1587 = 0.8185 using four-decimal table entries. With unrounded cumulative values the result rounds to 0.8186. The tiny difference is rounding, not a different method. Keep calculator precision until the final answer when available, and state the approximation honestly. For regions outside an interval, subtract the interval probability from one or add its two disjoint tails.

Another useful distinction is density versus cumulative probability. The height of a normal density curve at a boundary is not the chance of that exact value. Probabilities come from areas across intervals. If you change grams to kilograms, the horizontal scale contracts and the density height changes to keep the total area equal to one. The probability of the same physical mass range stays unchanged. This gives a practical check on your calculator function: a density command and a cumulative command serve different purposes. Read the function description rather than selecting the first normal-distribution option that appears.

Hands adjusting a coral strip between two clips on a silver tray

Reverse the calculation for a percentile

Suppose we want a threshold c with 90% of packages at or below it. The event is P(X ≤ c) = 0.90. Find the standard normal value with cumulative area 0.90: z ≈ 1.2816. Undo the standardization with c = μ + σz. This gives c ≈ 500 + 8 × 1.2816 = 510.2528 g, or about 510.25 g.

If the question says “the heaviest 10%,” the same threshold applies because the area above it is 0.10. Using a cumulative input of 0.10 would instead locate the lower tenth. Translate words into an area before pressing an inverse-normal function. Also check whether the function expects μ and σ directly or expects the already standardized distribution.

A percentile has the original measurement unit; its cumulative probability has none. To verify, put the threshold back into (c − 500)/8 and recover approximately 1.2816. Then check that the threshold is above 500 g, as a 90th percentile of this symmetric model must be. A rounded table might use z = 1.28 and give 510.24 g. Report the available precision rather than claiming more accurate digits than your input supports.

Standardizing does not create a normal distribution

The expression (x − μ)/σ can describe standardized distances for many distributions with finite mean and positive finite variance. That operation changes location and scale. It does not remove skewness, merge two peaks or prove that a normal model fits the data. A mean of zero and standard deviation one are not enough to justify Φ.

As a simple counterexample, let a variable take only −1 and 1 with equal probability. Its mean is zero and its variance is one already, but it remains a two-point distribution. Its probability of being at most zero is one half, while its probability of being at most 0.5 is also one half. A normal table gives a different second answer. Matching two summary measures does not match an entire distribution.

Also distinguish an individual package from a sample mean. If sixteen independent packages share our exact normal model, their mean has standard deviation 8/√16 = 2 g. The probability that this mean exceeds 502 g uses z = 1 and is about 0.1587. Using eight in the denominator would answer an individual-package question. Independence and a common model matter here; do not substitute a sample size into every problem automatically.

Solve a changed example and explain the region

Now consider an invented normally distributed rod length Y with mean 80 mm and standard deviation 3 mm. Work out P(Y > 83) and P(74 ≤ Y ≤ 83) before reading on. Write the events, standardize their boundaries and decide which cumulative areas to retain. The arithmetic is deliberately different from the package example, while the method stays the same.

The first boundary has z = 1, giving an upper tail of about 0.1587. The interval runs from z = −2 to z = 1, so its probability is Φ(1) − Φ(−2) ≈ 0.8413 − 0.0228 = 0.8185 from rounded table entries, or 0.8186 at higher precision. Your explanation should identify the removed left region, not merely repeat a calculator output. The independent answer-checking guide offers additional ways to test a result’s meaning and scale.

Finally, describe which assumptions make your answer valid and what precision you used. Eqora publishes this learning guide and offers an optional app for iOS and Android; these calculations require neither an app nor an upload. Any assisted explanation still needs checks against the original wording, units, distribution and event. Follow your course’s rules, and use a second fresh problem to confirm that you can choose the correct region yourself.

Man writing in a notebook inside a mountain shelter

Your next problem, with a clearer path

Ready to understand the math?

Scan a problem, ask follow-up questions, and turn one difficult step into a method you can use again.

Good to know

Questions about this guide