Section 6.1's machinery put to work. With a calculator or computer returning areas under the normal curve, three questions can be answered about any normally distributed quantity: the probability of a tail, the probability of an interval, and the value at a given percentile. The technology reports only the area to the left, so a right tail is found by subtracting from one and an interval by subtracting one left area from another, while a percentile reverses the process by taking an area and returning the boundary. The genuine difficulty in this section is neither the arithmetic nor the technology but the translation of English phrases into areas to the left: at most, at least, the bottom quartile, the top ten percent and the middle twenty percent all specify an area, and two of them specify the complement of the number they mention.
Subject: Statistics · 68 slides · symbolic lesson
Open the interactive version of this deck
Title
Statistics · Chapter 6 — The Normal Distribution
Every question is an area to the left, or its reverse
Objectives
Section 6.1 taught you to say where a value sits on a normal curve, in standard deviations. This section turns that position into a probability, and runs the question backwards to turn a probability into a value. The arithmetic is short; the thinking is in reading the question.
Everything in this section rests on one fact: calculators, computers and tables all report the same single number, the area to the left of a line. A right tail, a band, a percentile and a quartile are each one or two moves away from that number, and learning which moves to make is the whole skill.
The book shows each answer with TI-83 or TI-84 keystrokes. This deck shows the arithmetic underneath as well, a z-score and an area, so you can follow every example with any tool, or with a printed table.
OpenStax Introductory Statistics 2e, §6.2 Using the Normal Distribution §6.2, pp. 340-347 — the section this deck covers
Warm-up
Example 6.8 is about final exam scores that are normal with mean 63 and standard deviation 5. Before the section gives you any new tool, use what Section 6.1 gave you to place one score and estimate how many students beat it.
Discussion prompt
For X ~ N(63, 5), find the z-score of 65. Then use the Empirical Rule to say roughly what fraction of students score above 65: more or less than a half, and more or less than 16 percent?
Hold on to that estimate: somewhere between 16 and 50 percent, nearer the top. The first worked example computes it exactly, and it is 34.46 percent. An estimate like this is not wasted effort; it is the check that catches a wrong subtraction later.
Also recall from Chapter 5 that for a continuous variable a probability is an area under the density curve, and the total area is one. Every calculation today is an area, and every check is whether the areas add up.
OpenStax Introductory Statistics 2e, §6.2 Using the Normal Distribution §6.2, p. 341 — Example 6.8's distribution
Section
Part 1
Concept
The book opens the section with a single shaded picture. Draw a vertical line at some value x under a normal curve and shade everything to its left. That shaded area is the probability that X is less than x.
Figure (svg): A bell curve for final exam scores with mean 63 and standard deviation 5, ticks from 48 to 78. Everything under the curve to the left of a yellow line at 65 is shaded blue and labelled P of X less than 65 equals 0.6554, area to the left.
\[ P(X < x) = \text{area to the left of } x \]
This is the one quantity that normal tables, computers and calculators give you. A table lists it for standard normal z-scores; a calculator's normalcdf computes it for any mean and standard deviation. Every other question in this section, a right tail, a band, a percentile, is built from left areas.
For the exam scores, the left area at 65 is 0.6554: about 65.5 percent of students scored below 65. Read the picture: the line is just right of the peak, so a little more than half the area is to its left, which is exactly what 0.6554 says.
area to the left — P(X < x), the area under the density curve to the left of the vertical line through x. It is what normal tables and technology report, and every normal probability in this section is found from it.
OpenStax Introductory Statistics 2e, §6.2 Using the Normal Distribution §6.2, p. 340 — Figure 6.4 and the area to the left
Concept
The vertical line splits the total area, which is one, into a left piece and a right piece. Whatever is not on the left is on the right, so the right area is one minus the left area.
\[ P(X < x) + P(X > x) = 1 \]
\[ P(X > x) = 1 - P(X < x) \]
Figure (svg): The standard normal curve. A thin pink sliver left of z equals minus 2 is labelled left 0.0228. Everything to the right of minus 2 is shaded blue and labelled right equals 1 minus 0.0228 equals 0.9772. A note reads total area equals 1.
Example 6.7 is exactly this, with no distribution needed at all. If the area to the left is 0.0228, the area to the right is one minus 0.0228.
\[ 1 - 0.0228 = 0.9772 \]
On the standard normal curve, a left area of 0.0228 belongs to z equal to minus 2: it is the sliver beyond two standard deviations that the Empirical Rule calls about 2.5 percent. Try It 6.7 is the same move with a left area of 0.012: the right area is 0.988.
Always sketch before subtracting. With a picture you can see at once that a thin left sliver leaves a huge right area, and a subtraction that produced a small right area would be obviously wrong.
OpenStax Introductory Statistics 2e, §6.2 Using the Normal Distribution §6.2, pp. 340-341 — P(X > x) = 1 − P(X < x); Example 6.7 and Try It 6.7
Notation
The book's calculator command for any normal probability takes four numbers. Read each one, because the first two are the boundaries of the shaded region and the last two describe the curve.
Annotate
On: \( \text{normalcdf}(\text{lower},\ \text{upper},\ \mu,\ \sigma) \)
Under the hood the calculator computes the left area at the upper value and subtracts the left area at the lower value. With the lower value at minus ten to the ninety-ninth, that second left area is zero, so you get a pure left area; with the upper value at ten to the ninety-ninth, the first is one, so you get one minus a left area.
\[ \text{normalcdf}(a,\ b,\ \mu,\ \sigma) = P(X < b) - P(X < a) \]
So even the calculator is doing the two subtractions of this part. Knowing that lets you do every problem with a printed table instead.
OpenStax Introductory Statistics 2e, §6.2 Using the Normal Distribution §6.2, pp. 341-342 — the normalcdf syntax
Concept
Practice Exercise 45 asks whether the probability that X is less than 1 equals the probability that X is at most 1. For a continuous distribution it does, and the reason is worth seeing once. The two differ only by the probability that X equals exactly 1.
\[ P(X \le 1) = P(X < 1) + P(X = 1) \]
The probability of one exact value is the area of a region with no width, the vertical line itself. A line has no area, so that probability is zero.
\[ P(X = 1) = 0 \]
\[ P(X \le 1) = P(X < 1) \]
That is why the book treats less than x and less than or equal to x as the same, and why at most and less than lead to the same left area. It is also why you never need to worry about whether a boundary value is included.
This is a real difference from the binomial and Poisson distributions of Chapter 4, where X counts whole things and P(X = 3) can be large. There at most 3 and less than 3 are different questions. On a continuous curve they are the same question.
OpenStax Introductory Statistics 2e, §6.2 Using the Normal Distribution §6.2, p. 340; practice p. 353 — continuous distributions; Exercise 45
Concept
The book's Historical Note shows how the area was found before calculators, and it is still how you work from a printed table. A standard normal table lists left areas for z-scores only, so first turn the value into a z-score.
\[ z = \frac{65 - 63}{5} = 0.4 \]
Figure (svg): The exam-score curve with two rows of labels under the axis: scores 48, 53, 58, 63, 65, 68, 73, 78 and, in orange beneath them, z values minus 3, minus 2, minus 1, 0, 0.4, 1, 2, 3. The area left of 65 is shaded, with the note 0.6554 from the table.
The picture has two rulers under one curve. The score 65 and the z-score 0.4 sit at the same place, so the area to their left is the same area. The table, read at z equal to 0.40, gives 0.6554.
\[ P(X < 65) = P(Z < 0.4) = 0.6554 \]
That equality is the whole reason tables work: standardising moves the question from X's curve to the one standard curve without changing any area. Section 6.1 showed every normal curve is the standard curve with a different ruler under it; here that fact pays off.
The book adds that tables are now largely historical, and technology is more precise. Use whichever tool you have, but understand that both are reading the same left area.
OpenStax Introductory Statistics 2e, §6.2 Using the Normal Distribution §6.2, p. 342 — Historical Note
Worked example
Final exam scores in a statistics class were normal with mean 63 and standard deviation 5. Find the probability that a randomly selected student scored more than 65.
Define X and its distribution
Why: the score of one randomly chosen student.
\[ X \sim N(63, 5) \]
Write the question as a probability
Why: more than 65 is a right tail.
\[ P(X > 65) \]
Turn the right tail into one minus a left area
Why: the only area technology returns.
\[ P(X > 65) = 1 - P(X < 65) \]
Before any number, notice what kind of question this is. The line is given, at 65, and an area is wanted; the word more puts that area on the right of the line. Technology only gives the left, so you already know a subtraction is coming.
Standardise the boundary
Why: for a table; a calculator does this internally.
\[ z = \frac{65 - 63}{5} = 0.4 \]
Read the left area
Why: table at z of 0.40, or normalcdf with lower bound minus ten to the ninety-ninth.
\[ P(X < 65) = P(Z < 0.4) = 0.6554 \]
Subtract from one
Why: the right tail.
\[ P(X > 65) = 1 - 0.6554 = 0.3446 \]
Figure (svg): The exam-score curve for mean 63 and standard deviation 5. The region to the right of a yellow line at 65 is shaded orange and labelled P of X greater than 65 equals 0.3446; the unshaded region on the left is labelled 0.6554.
On the calculator the whole thing is one line, and it gives the same number. The upper bound is ten to the ninety-ninth, standing in for no right edge at all.
\[ \text{normalcdf}(65,\ 10^{99},\ 63,\ 5) = 0.3446 \]
Interpret it in context: the probability that a randomly selected student scored more than 65 is 0.3446, so about 34 percent of the class did.
Check against the warm-up estimate
Why: 65 is 0.4 standard deviations up, so the tail had to be between 0.16 and 0.50, nearer 0.50; it is. And the two pieces must add to one.
\[ 0.16 < 0.3446 < 0.50, \qquad 0.6554 + 0.3446 = 1 \]
OpenStax Introductory Statistics 2e, §6.2 Using the Normal Distribution §6.2, pp. 341-342 — Example 6.8(a) and the Historical Note
Error analysis
Four students answered Example 6.8(a), P(X > 65) for X ~ N(63, 5). The correct answer is 0.3446. Find what each wrong attempt did, and what would have caught it.
Annotate
On: \( 0.6554 \quad 1 - 0.4 \quad P(X{<}65) - P(X{<}63) \quad 1 - 0.6554 \)
The first and third errors each produce a genuine area under the curve, which is what makes them convincing. The only protection is the picture: shade the region the question names, then ask which of your numbers is that shaded area.
One quick habit prevents all three errors. Before computing, write down whether the answer must be above or below one half. A region to the right of a value that is above the mean must be below one half, and that alone rules out 0.6554 and 0.6.
The second error mixes up the two rulers from the table slide. A z-score of 0.4 lives on the horizontal axis; the area 0.6554 lives under the curve. Subtracting a position from one means nothing.
OpenStax Introductory Statistics 2e, §6.2 Using the Normal Distribution §6.2, pp. 341-342 — Example 6.8(a)
Worked example
Same class, X ~ N(63, 5). Find the probability that a randomly selected student scored less than 85.
Write the question as a probability
Why: less than 85 is a left area, so no subtraction is needed.
\[ P(X < 85) \]
Standardise to see where 85 sits
Why: far to the right of the mean.
\[ z = \frac{85 - 63}{5} = \frac{22}{5} = 4.4 \]
Read the left area
Why: node gives 0.999995.
\[ P(X < 85) = P(Z < 4.4) = 0.999995 \approx 1 \]
Figure (svg): The exam-score curve with x ticks from 48 to 88 and z values minus 3 to 5 beneath them. The whole area left of a yellow line at 85 is shaded, which is visually the entire curve; the line is labelled 85, z equals 4.4. A note reads P of X less than 85 is about 0.999995.
The book computes normalcdf with a lower bound of 0 rather than minus ten to the ninety-ninth. That is fine here: a score below 0 is 12.6 standard deviations below the mean, so the area it leaves out is far too small to show up.
\[ \text{normalcdf}(0,\ 85,\ 63,\ 5) \approx 1 \]
Interpret it: the probability that one student scores less than 85 is approximately one, or 100 percent. It is not exactly one, and saying rounds to one is more honest than saying certain: roughly five students in a million would score 85 or more under this model.
A z-score beyond 3 in size is already outside the Empirical Rule's 99.7 percent band, so a z-score of 4.4 tells you before any technology that the answer will be one to three or more decimal places.
Check with the complement
Why: the right tail beyond 4.4 standard deviations must be tiny, and the two must add to one.
\[ P(X > 85) \approx 0.000005, \qquad 0.999995 + 0.000005 = 1 \]
OpenStax Introductory Statistics 2e, §6.2 Using the Normal Distribution §6.2, p. 342 — Example 6.8(b)
Concept
Example 6.8(b) gave an answer of one to five decimal places, and it is worth seeing why a normal curve's tails vanish so quickly. The bars below are the area beyond z, the right tail, for z from one half to four standard deviations.
Figure (svg): Eight orange bars showing the standard normal right-tail area beyond z equals 0.5, 1, 1.5, 2, 2.5, 3, 3.5 and 4. The bars fall from 0.3085 to 0.1587, 0.0668, 0.0228, 0.0062, 0.0013, 0.00023 and 0.00003, the last few too short to see.
\[ P(Z > 1) = 0.1587, \quad P(Z > 2) = 0.0228, \quad P(Z > 3) = 0.0013 \]
Going from one to two standard deviations divides the tail by about 7. Going from two to three divides it by about 17, and from three to four by about 42. The density contains the square of z in an exponent, so every step out costs far more area than the step before.
\[ \frac{0.1587}{0.0228} \approx 7, \qquad \frac{0.0228}{0.00135} \approx 17, \qquad \frac{0.00135}{0.000032} \approx 42 \]
That is why the Empirical Rule stops at three: beyond it there is almost nothing left, 0.13 percent on each side. And it is why a normal model makes very confident claims about extremes, which real data do not always honour. Keep that in mind in the paternity case of Part 5, where the answer lives in a tail.
For your own work the practical lesson is a sense of scale. A z-score around 1 leaves a tail of about a sixth, around 2 about a fortieth, around 3 about one in seven hundred. Use those to judge whether any tail probability you compute is plausible.
OpenStax Introductory Statistics 2e, §6.2 Using the Normal Distribution §6.2, p. 342 — Example 6.8(b); §6.1's Empirical Rule
Worked example
Golf scores for a school team were normal with mean 68 and standard deviation 3. Find the probability that a randomly selected golfer scored less than 65. In golf a low score is a good one, but the arithmetic does not care.
Set up
Why: less than 65 is a left area.
\[ X \sim N(68, 3), \qquad P(X < 65) \]
Sketch first. 65 is below the mean of 68, so the region less than 65 is a left tail smaller than one half. Whatever you compute must come out under 0.5; if it does not, you have found the other side of the line by mistake.
Standardise
Why: 3 below the mean, one standard deviation.
\[ z = \frac{65 - 68}{3} = \frac{-3}{3} = -1 \]
Read the left area
Why: table at z of minus 1.00, or normalcdf with lower bound minus ten to the ninety-ninth.
\[ P(X < 65) = P(Z < -1) = 0.1587 \]
Figure (svg): A bell curve of golf scores with mean 68 and standard deviation 3, ticks 59 to 77 and z values minus 3 to 3 beneath. The left tail below 65, exactly one standard deviation below the mean, is shaded and labelled 0.1587.
So the probability that a randomly chosen golfer on this team scored under 65 is 0.1587, about 16 percent. Those are the team's best rounds.
The value 65 lands exactly on a z tick, and whenever that happens the Empirical Rule already knows the answer roughly. Below one standard deviation under the mean is half of what is left outside the 68 percent band.
Check with the Empirical Rule
Why: half of the 32 percent outside one standard deviation.
\[ \frac{1 - 0.68}{2} = 0.16 \approx 0.1587 \]
OpenStax Introductory Statistics 2e, §6.2 Using the Normal Distribution §6.2, p. 343 — Try It 6.8
Concept
Now a band with two edges, a and b. Nothing new is needed. The left area at b covers the band and everything to its left; the left area at a covers only the part to its left. Subtract, and the band is what remains.
\[ P(a < X < b) = P(X < b) - P(X < a) \]
Figure (svg): Three small copies of the same bell curve for X distributed N of 2 and 0.5. The first has everything left of 2.75 shaded blue, labelled 0.9332. A minus sign. The second has everything left of 1.8 shaded pink, labelled 0.3446. An equals sign. The third has only the band from 1.8 to 2.75 shaded yellow, labelled 0.5886.
Example 6.9 is the picture. Household computer use for entertainment is X ~ N(2, 0.5) hours per day. The left area at 2.75 is 0.9332 and the left area at 1.8 is 0.3446. Their difference, 0.5886, is the probability of a day between 1.8 and 2.75 hours.
\[ 0.9332 - 0.3446 = 0.5886 \]
The order is always upper minus lower. Reverse it and you get a negative number, which cannot be a probability, so at least that error announces itself.
This is the same subtraction that normalcdf performs when you give it a lower and an upper value. With it and the right-tail rule you can answer every probability question in this section: left area as it is, one minus a left area, or one left area minus another.
OpenStax Introductory Statistics 2e, §6.2 Using the Normal Distribution §6.2, pp. 343-344 — Example 6.9(a)
Worked example
The average household computer is used two hours a day for entertainment, with a standard deviation of half an hour, and times are normal. Find the probability that a household's computer is used between 1.8 and 2.75 hours a day.
Define X
Why: hours of entertainment use per day.
\[ X \sim N(2, 0.5) \]
Write the band as a difference of left areas
Why: upper minus lower.
\[ P(1.8 < X < 2.75) = P(X < 2.75) - P(X < 1.8) \]
Sketch the band before standardising. It starts a little left of the mean and ends well to the right, so it straddles the peak, where the curve is tallest. Expect it to hold a good share of the curve, somewhere around a half.
Standardise the upper edge
Why: 0.75 hours above, one and a half standard deviations.
\[ z = \frac{2.75 - 2}{0.5} = 1.5 \]
Standardise the lower edge
Why: 0.2 hours below.
\[ z = \frac{1.8 - 2}{0.5} = -0.4 \]
Read both left areas
Why: table or technology.
\[ P(Z < 1.5) = 0.9332, \qquad P(Z < -0.4) = 0.3446 \]
Subtract
Why: upper minus lower.
\[ 0.9332 - 0.3446 = 0.5886 \]
Figure (svg): A bell curve of hours per day with mean 2 and standard deviation 0.5. The band from a pink line at 1.8 to an orange line at 2.75 is shaded yellow and labelled 0.5886. Under the axis, the z row reads minus 0.4 under 1.8 and 1.5 under 2.75.
\[ \text{normalcdf}(1.8,\ 2.75,\ 2,\ 0.5) = 0.5886 \]
Interpret it: the probability that a household computer is used for entertainment between 1.8 and 2.75 hours per day is 0.5886. The lower left area, 0.3446, is the same number as Example 6.8's right tail, because minus 0.4 and plus 0.4 are mirror images on a symmetric curve.
Check by splitting at the mean
Why: the band is the piece from minus 0.4 to 0 plus the piece from 0 to 1.5, each measured from the half-way area 0.5.
\[ (0.5 - 0.3446) + (0.9332 - 0.5) = 0.1554 + 0.4332 = 0.5886 \]
OpenStax Introductory Statistics 2e, §6.2 Using the Normal Distribution §6.2, pp. 343-344 — Example 6.9(a)
Worked example
Back to the golf team, X ~ N(68, 3). Find the probability that a golfer scored between 66 and 70.
Write the band as a difference
Why: upper minus lower.
\[ P(66 < X < 70) = P(X < 70) - P(X < 66) \]
Standardise both edges
Why: each is 2 points, or two thirds of a standard deviation, from 68.
\[ z = \frac{70 - 68}{3} \approx 0.667, \qquad z = \frac{66 - 68}{3} \approx -0.667 \]
Notice that the two z-scores are mirror images. That means the left area at 66 equals the right area beyond 70, a fact you can use as a check, or to skip one of the two look-ups entirely.
Read both left areas
Why: node, from the unrounded z-scores.
\[ P(X < 70) = 0.7475, \qquad P(X < 66) = 0.2525 \]
Subtract
Why: upper minus lower.
\[ 0.7475 - 0.2525 = 0.4950 \]
Figure (svg): The golf-score curve with mean 68 and standard deviation 3. The band from 66 to 70, marked by pink lines, is shaded yellow and labelled 0.4950. Below the axis two orange brackets show that each half of the band is 2 points, or 0.67 standard deviations.
So there is a 0.4950 probability, just under a half, that a randomly chosen golfer scored between 66 and 70. A band of two thirds of a standard deviation on each side captures about half the curve; remember that, because it returns in Part 4 as the quartiles.
If you use a table with z rounded to 0.67, you get 0.7486 and 0.2514 and an answer of 0.4972. The difference is only rounding: two thirds is not exactly 0.67.
Check by symmetry
Why: the band is centred on the mean, so it is twice the area from 68 to 70.
\[ 2(0.7475 - 0.5) = 2(0.2475) = 0.4950 \]
OpenStax Introductory Statistics 2e, §6.2 Using the Normal Distribution §6.2, p. 344 — Try It 6.9
Sorting
Every probability in this part came from one of three moves on left areas. Before computing anything, sort these questions from the section and its exercises by the move they need.
Sort into buckets
Sort each question by how it is built from areas to the left.
The wording decides the bucket, and the numbers never do. At least 4 years is a right tail even though no word like more appears; between is a band whatever the distribution.
Practice Exercise 58 is the last item: for X ~ N(4.1, 1.3), the left areas at 6 and 2.8 are 0.9280 and 0.1587, so the band is 0.7694. If you put it with the right tails, you probably read only the lower edge.
OpenStax Introductory Statistics 2e, §6.2 Using the Normal Distribution §6.2 practice, pp. 353-354 — Exercises 50-58
Trap
Example 6.9(a) again, P(1.8 < X < 2.75) with σ equal to 0.5. A first attempt measures the band's width and standardises that:
\[ 2.75 - 1.8 = 0.95 \]
\[ \frac{0.95}{0.5} = 1.9 \]
\[ P(Z < 1.9) = 0.9713 \]
Wrong. 0.9713 is almost the whole curve.
A band's probability depends on where it sits, not just on how wide it is. A band 0.95 wide in the middle holds far more than one out in a tail. Standardise each edge.
\[ z_{a} = -0.4, \quad z_{b} = 1.5 \]
\[ 0.9332 - 0.3446 \]
\[ = 0.5886 \]
The width alone gives the left area at 1.9, which is a different region entirely. The picture catches it: a band that does not even reach the right tail cannot hold 97 percent of the curve.
OpenStax Introductory Statistics 2e, §6.2 Using the Normal Distribution §6.2, pp. 343-344 — Example 6.9(a)
Ranking
Stay with the final exam scores, X ~ N(63, 5). Rank these four probabilities from smallest to largest using only a sketch and the Empirical Rule. You need the order, not the values.
Put in order
Order from the smallest probability to the largest.
Why: Seventy is 1.4 standard deviations up, so its right tail is under 16 percent: about 0.081. The band from 62 to 64 is only 0.2 standard deviations each side of the peak: about 0.159. Below 60 is 0.6 standard deviations down: about 0.274. Above the mean is exactly one half.
\[ 0.0808 < 0.1585 < 0.2743 < 0.5 \]
The middle pair is the hard one. A band two points wide right at the peak looks like it should be big, and it does collect the tallest part of the curve, but it is narrow: 0.4 standard deviations in all. The left tail below 60 is wider than it looks, because it runs all the way out.
Above the mean is the anchor. Any probability of a region that lies wholly on one side of the mean is under 0.5, and that alone places the last item.
Section
Part 2
Concept
Example 6.8(c) asks for the 90th percentile of the exam scores: the score k that has 90 percent of scores below it and 10 percent above. Compare that with Part 1. There you were handed the line and asked for the area. Now you are handed the area and asked for the line.
Figure (svg): The exam-score curve. The area left of a dashed yellow line is shaded blue and labelled given 0.90; the thin area to its right is orange, labelled 0.10. The dashed line itself is labelled k equals question mark.
\[ P(X < k) = 0.90, \qquad k = ? \]
The book calls k a critical value, a name you will meet again in Chapter 8. It is a value on the horizontal axis, in the units of X, so the answer to a percentile question is a score, a number of hours or a number of centimetres, never a probability.
Because technology reports areas to the left, a percentile question is easiest when the area it gives is already a left area. The 90th percentile is: 90 percent below k. Part 3 handles the questions whose area is on the right.
kth percentile — The value k with k percent of the distribution below it. For a normal X it solves P(X < k) = k/100, and it is found by running the left-area calculation backwards.
OpenStax Introductory Statistics 2e, §6.2 Using the Normal Distribution §6.2, pp. 342-343 — Example 6.8(c): the 90th percentile and critical values
Concept
Running Part 1 backwards takes two moves. First find the z-score whose left area is the given area: that is a standard normal table read from the inside out, or the inverse function on a calculator. For an area of 0.90 it is 1.2816.
\[ P(Z < z) = 0.90 \quad \Rightarrow \quad z = 1.2816 \]
Second, convert that z-score back to a score with the formula from Section 6.1. Start at the mean and walk z standard deviations.
\[ k = \mu + z\sigma \]
\[ k = 63 + (1.2816)(5) = 63 + 6.408 = 69.408 \]
The calculator's invNorm does both moves at once. Notice the pleasing fact hiding in the first move: the z-score for the 90th percentile is 1.2816 for every normal distribution. The 90th percentile always sits 1.28 standard deviations above the mean; only the size of a standard deviation changes.
| percentile | 10th | 25th | 50th | 70th | 75th | 90th | 98th |
|---|---|---|---|---|---|---|---|
| z | −1.2816 | −0.6745 | 0 | 0.5244 | 0.6745 | 1.2816 | 2.0537 |
Every z-score in that table appears in this deck. Percentiles below the 50th have negative z-scores, and the table is symmetric: the 10th and 90th percentiles are mirror images.
OpenStax Introductory Statistics 2e, §6.2 Using the Normal Distribution §6.2, p. 343 — invNorm and the 90th percentile
Notation
The book's inverse command is shorter than normalcdf, and its first argument is the one that trips people up.
Annotate
On: \( k = \text{invNorm}(\text{area to the left},\ \mu,\ \sigma) \)
Compare the two commands side by side. normalcdf takes values and returns an area; invNorm takes an area and returns a value. They undo each other, which gives you a free check on every percentile you compute.
\[ \text{normalcdf}(-10^{99},\ \text{invNorm}(0.9,\ 63,\ 5),\ 63,\ 5) = 0.9 \]
Whenever you are unsure which command a question needs, ask which of the two, a value or an area, the question gives you.
OpenStax Introductory Statistics 2e, §6.2 Using the Normal Distribution §6.2, p. 343 — invNorm syntax; Formula Review, p. 351
Worked example
Exam scores are X ~ N(63, 5). Find the 90th percentile: the score k with 90 percent of scores below it and 10 percent above.
Write the equation for k
Why: 90 percent to the left of k.
\[ P(X < k) = 0.90 \]
The question hands you an area, 90 percent, and asks for a score, so this runs in the reverse direction. The 90 percent is already below k, an area to the left, so nothing needs subtracting before you look it up.
Find the z-score with left area 0.90
Why: table inside out, or invNorm without μ and σ.
\[ z = 1.2816 \]
Walk from the mean
Why: k equals μ plus zσ.
\[ k = 63 + (1.2816)(5) \]
Multiply
Why: 1.28 standard deviations of 5 points.
\[ k = 63 + 6.408 \]
Add and round
Why: the book reports one decimal.
\[ k = 69.408 \approx 69.4 \]
Figure (svg): The exam-score curve with score ticks and a z row beneath. The area left of a yellow line at 69.4 is shaded and labelled 0.90. An orange arrow runs from the mean 63 to the line, labelled 1.28 times 5 equals 6.41.
\[ \text{invNorm}(0.90,\ 63,\ 5) = 69.4 \]
Interpret it: the 90th percentile is 69.4, so 90 percent of the test scores fall at or below 69.4 and 10 percent fall at or above. The book stresses that k separates the scores into those at or below it and those at or above it; on a continuous curve the boundary itself carries no area, as Part 1 showed.
Check by running it forwards
Why: the left area at the unrounded k must be 0.90; at the rounded 69.4 it is a hair less.
\[ P(X < 69.408) = 0.9000, \qquad P(X < 69.4) = 0.8997 \]
OpenStax Introductory Statistics 2e, §6.2 Using the Normal Distribution §6.2, pp. 342-343 — Example 6.8(c)
Worked example
Same scores, X ~ N(63, 5). Find the 70th percentile: the score k with 70 percent of scores below it and 30 percent above.
Write the equation for k
Why: 70 percent to the left.
\[ P(X < k) = 0.70 \]
Predict before computing. Seventy percent is more than half, so k is above the mean; and 70 is much less extreme than 90, so k should be much closer to 63 than the 69.4 you just found.
Find the z-score with left area 0.70
Why: a little above the mean.
\[ z = 0.5244 \]
Walk from the mean
Why: 0.52 standard deviations of 5.
\[ k = 63 + (0.5244)(5) = 63 + 2.622 \]
Add and round
Why: one decimal.
\[ k = 65.622 \approx 65.6 \]
Figure (svg): The exam-score curve with the area left of a yellow line at 65.6 shaded blue, labelled 0.70, and the area right of it orange, labelled 0.30.
\[ \text{invNorm}(0.70,\ 63,\ 5) = 65.6 \]
Interpret it: 70 percent of the test scores fall at or below 65.6 and 30 percent fall at or above. The book's interpretation sentence says 65.5; that is a misprint, since its own calculation, and node, give 65.6.
Compare with the 90th percentile. Going from 70 to 90 percent moved the cut-off from 65.6 to 69.4, almost four points, while going from 50 to 70 percent moved it only 2.6. Near the peak a little distance buys a lot of area; out in the tail you have to walk much further for each extra percent.
Check by running it forwards
Why: the left area at 65.622 must be 0.70.
\[ P(X < 65.622) = 0.7000 \]
OpenStax Introductory Statistics 2e, §6.2 Using the Normal Distribution §6.2, p. 343 — Example 6.8(d)
Picture it
Collect every left area of the exam scores into one graph: for each score x, plot the area to its left. The result climbs from 0 on the far left to 1 on the far right, and it is the single object behind both kinds of question in this section.
Figure (svg): An S-shaped curve rising from 0 to 1 over exam scores from 46 to 80, the area to the left of x for N of 63 and 5. A dashed yellow line goes up from 65 to the curve and an arrow runs left to the height 0.6554. A dashed orange line runs across from the height 0.90 to the curve and an arrow runs down to 69.4. Notes read normalcdf reads up, invNorm reads across.
Follow the yellow path first. Start at the score 65 on the axis, go up to the curve and across to the vertical axis: the height is 0.6554, the left area of Example 6.8(a). That is a probability question, and normalcdf is a machine for reading upward.
Now the orange path. Start at the height 0.90, go across to the curve and down to the axis: the score is 69.4, the 90th percentile of Example 6.8(c). That is a percentile question, and invNorm is a machine for reading across.
\[ 65 \xrightarrow{\text{ up }} 0.6554, \qquad 0.90 \xrightarrow{\text{ across }} 69.4 \]
The curve is steepest over the mean, 63, and nearly flat far from it. So near the mean a small change in score changes the left area a lot, and far out a big change in score changes it hardly at all. The next slide lets you move this curve yourself.
Tweak it
Plot the left area P(X < x) against x and you get an S-shaped curve that climbs from 0 to 1. Reading it up from the x-axis is a probability question; reading it across from the y-axis and down is a percentile question. The slider draws this curve for any normal distribution.
Parameter explorer
Set the mean to 63 and the standard deviation to 5. Read the height of the curve above x equal to 65, and then find where the curve crosses the height 0.9. Then double the standard deviation and say what happens to the 90th percentile.
\[ P(X < x), \quad X \sim N({m},\ {s}) \]
\[ \sigma = 5: k_{90} = 69.4, \qquad \sigma = 10: k_{90} = 63 + 12.8 = 75.8 \]
The curve is steepest at the mean, where the density is tallest, which is why a small step near the mean changes the left area so much. Doubling σ stretches the whole S sideways, so every percentile moves twice as far from the mean, but the 50th percentile stays put at μ.
This curve is exactly what normalcdf and invNorm compute: one reads it upward, the other reads it sideways. Chapter 5 called it the cumulative distribution function.
Prediction
Try It 6.10 asks for the 30th percentile of smartphone users' ages, X ~ N(36.9, 13.9). Before any calculation, decide where it must lie.
Predict first
Where does the 30th percentile of X ~ N(36.9, 13.9) lie?
Correct: below the mean, a little over half a standard deviation down
Why: Only 30 percent of the area is to its left, less than half, so it must be left of the mean, where half the area lies. Its z-score is minus 0.5244, so it is a bit more than half a standard deviation below 36.9.
\[ k = 36.9 + (-0.5244)(13.9) = 36.9 - 7.289 \approx 29.61 \]
Interpret it: the 30th percentile is 29.6 years, so 30 percent of smartphone users aged 13 to 55 and over are 29.6 or younger. That it is close to 30 is a coincidence of these numbers, which is exactly why the option saying exactly 30 is a trap: a percentile's value has nothing to do with its label.
The last option confuses the area with the distance. Thirty percent of the area and 0.3 standard deviations are different things, and the table of z-scores for percentiles is how you move between them.
OpenStax Introductory Statistics 2e, §6.2 Using the Normal Distribution §6.2, p. 345 — Try It 6.10(a)
Worked example
Household entertainment use is X ~ N(2, 0.5) hours per day. Find the maximum number of hours per day that the bottom quartile of households uses a computer for entertainment.
Translate the phrase
Why: the bottom quartile is the lowest 25 percent; its maximum is the 25th percentile.
\[ P(X < k) = 0.25 \]
Quartile means quarter, and bottom means the left end. So the bottom quartile is the lowest quarter of households, and its largest value is the line with a quarter of the area to its left. The area is already on the left; no subtraction.
Find the z-score with left area 0.25
Why: below the mean.
\[ z = -0.6745 \]
Walk from the mean
Why: 0.67 standard deviations of half an hour, to the left.
\[ k = 2 + (-0.6745)(0.5) \]
Multiply and add
Why: about a third of an hour below 2.
\[ k = 2 - 0.3372 = 1.6628 \approx 1.66 \]
Figure (svg): The entertainment-hours curve for mean 2 and standard deviation 0.5. The left tail below a yellow line at 1.66 hours is shaded pink and labelled bottom quarter 0.25; the rest is labelled other 75 percent.
\[ \text{invNorm}(0.25,\ 2,\ 0.5) = 1.66 \]
Interpret it: the bottom quarter of households use their computer for entertainment at most 1.66 hours per day. The question says maximum because the bottom quartile is a whole group, everyone from zero up to the line, and the line is its largest member.
The number 25 never appears in the question; the word quartile carries it. Part 3 is about exactly this: turning words into a left area before any technology is touched.
Check by running it forwards and against the mean
Why: a quarter of the area lies left of 1.6628, and a bottom quartile must sit below the mean of 2.
\[ P(X < 1.6628) = 0.2500, \qquad 1.66 < 2 \]
OpenStax Introductory Statistics 2e, §6.2 Using the Normal Distribution §6.2, p. 344 — Example 6.9(b)
Trap
Practice Exercise 52: X ~ N(54, 8), find the 80th percentile. A first attempt feeds the 80 in as a value:
\[ P(X < 80) \]
\[ = 0.9994 \]
and reports 0.9994. Wrong: the question asked for a value of X.
The 80 is an area, 0.80, and the answer is a point on the axis. Run the question backwards.
\[ P(X < k) = 0.80 \]
\[ z = 0.8416 \]
\[ k = 54 + (0.8416)(8) \]
\[ k = 60.73 \]
A percentile is always in the units of X. If your answer to a percentile question is between 0 and 1 and X is measured in something else, you computed a probability. Check forwards: P(X < 60.73) is 0.80.
OpenStax Introductory Statistics 2e, §6.2 Using the Normal Distribution §6.2 practice, p. 354 — Exercise 52
Fill the middle
Practice Exercise 53 uses the same distribution, X ~ N(54, 8), and asks for the 60th percentile. The z-score with left area 0.60 is 0.2533. Fill in the walk from the mean and the answer, to two decimal places.
Fill in the blanks
k = 54 + (0.2533)(8) = 54 + 2.03 = 56.03
Why: Sixty percent is a little more than half, so the percentile is a little above the mean: 0.2533 standard deviations of 8 is about 2.03, and 54 plus 2.03 is 56.03, the book's answer.
So 60 percent of values of X are at or below 56.03. The step from the 50th to the 60th percentile is only a quarter of a standard deviation, because the ten percent nearest the peak is packed into a narrow strip.
If you got 51.97, you used minus 0.2533: that is the 40th percentile, the mirror image. Ask which side of the mean before you walk.
\[ P(X < 56.03) = 0.6000 \]
OpenStax Introductory Statistics 2e, §6.2 Using the Normal Distribution §6.2 practice, p. 354 — Exercise 53
Section
Part 3
Concept
The real difficulty in this section is not the arithmetic or the technology. It is reading a sentence and knowing which area it describes. Each phrase below appears in the book's examples or exercises; the right-hand column is the left area it becomes.
Figure (svg): Four small bell curves in a row. At most 50.8: the left part is shaded blue, note left area. At least k: the right part is shaded orange, note 1 minus left area. Top 10 percent: a thin right tail is shaded orange, note left area 0.90. Bottom quartile: a left tail is shaded pink, note left area 0.25.
| the question says | the region | area to the left |
|---|---|---|
| at most 50.8, less than 50.8 | left of 50.8 | P(X < 50.8), as it is |
| more than 65, at least 8 | right of the value | 1 − P(X < value) |
| between 23 and 64.7 | a band | P(X < 64.7) − P(X < 23) |
| the bottom quartile | lowest 25% | 0.25 |
| the top 2% | highest 2% | 0.98 |
| 40% are at least k | right of k is 0.40 | 0.60 |
| the fastest 3% (times) | shortest 3% | 0.03 |
Two phrases give you the complement of the number they mention: at least and top both describe an area on the right, so the left area is one minus the number. The book's Example 6.11(b) spells it out: at least translates to greater than or equal to.
The last row is a warning that the word must be read in context. Fastest sounds like the top of something, but fast laps take the fewest seconds, so the fastest 3 percent are on the left.
OpenStax Introductory Statistics 2e, §6.2 Using the Normal Distribution §6.2, pp. 344-346 — Examples 6.9-6.11; homework pp. 356-358
Matching
Each phrase comes from a homework exercise of this section. Match it with the area to the left that you would enter into invNorm or read from a table. Sketch each one; it takes seconds.
Match the pairs
Why: Top and more than name the area on the right, so the left area is one minus the number: 0.98 and 0.30. Within, under and the lower quarter already name the area on the left: 0.60, 0.03 and 0.25.
Notice that the phrase more than appears twice in this section with different jobs. In Example 6.8(a), more than 65 names a region beyond a known value. Here, more than k minutes names a region beyond an unknown value, with its area given. Both are right tails.
A reliable test for any phrase: imagine one person, or one trial, that fits it. Would they sit left of the line or right of it? A trial completed within k days is short, so left; an IQ in the top 2 percent is high, so right; a lap among the fastest is short, so left.
Within k days means the trial is over by day k, so it is a left area. If you matched it with 0.40, you read within as the complement; nothing in the phrase points to the right.
OpenStax Introductory Statistics 2e, §6.2 Using the Normal Distribution §6.2 homework, pp. 357-358 — Exercises 72, 74, 75, 79 and 80
Worked example
Smartphone users aged 13 to 55 and over have ages X ~ N(36.9, 13.9). Forty percent of them are at least what age? Round to one decimal place.
Translate the phrase
Why: at least means greater than or equal to: the 40 percent is on the right.
\[ P(X \ge k) = 0.40 \]
Turn it into a left area
Why: the rest of the area is to the left of k.
\[ P(X < k) = 1 - 0.40 = 0.60 \]
This is the step the section is really about. The question mentions 40 percent, but the number you look up is 60 percent, because a look-up always wants the area to the left and at least puts the 40 percent on the right.
Find the z-score with left area 0.60
Why: just above the mean.
\[ z = 0.2533 \]
Walk from the mean
Why: a quarter of a standard deviation of 13.9 years.
\[ k = 36.9 + (0.2533)(13.9) = 36.9 + 3.5215 \]
Add and round
Why: one decimal.
\[ k = 40.4215 \approx 40.4 \]
Figure (svg): The smartphone-age curve for mean 36.9 and standard deviation 13.9, ticks at 9.1, 23, 36.9, 50.8 and 64.7. A yellow line at 40.4 years splits it: the left part is shaded blue and labelled left 0.60, the right part orange and labelled at least k 0.40.
\[ \text{invNorm}(0.60,\ 36.9,\ 13.9) = 40.4215 \]
Interpret it: forty percent of smartphone users from 13 to 55 and over are at least 40.4 years old. The 40 in the question and the 40.4 in the answer are unrelated; one is a share of people and the other is an age.
Forty percent is less than half, so the region at least k covers less than half the curve, and k must sit above the mean. Entering 0.40 instead would put k below the mean, at 33.4, where 60 percent are at least that old.
Check the right tail directly
Why: one minus the left area at 40.4215 must give back 0.40.
\[ 1 - P(X < 40.4215) = 1 - 0.6000 = 0.40 \]
OpenStax Introductory Statistics 2e, §6.2 Using the Normal Distribution §6.2, pp. 345-346 — Example 6.11(b)
Worked example
The time to find a parking space at 9 a.m. is normal with mean five minutes and standard deviation two minutes. Seventy percent of the time, it takes more than how many minutes to find a space?
Set up
Why: X is minutes to find a space.
\[ X \sim N(5, 2) \]
Translate
Why: more than k, seventy percent of the time, is a right area.
\[ P(X > k) = 0.70 \]
Turn it into a left area
Why: the other 30 percent is to the left.
\[ P(X < k) = 1 - 0.70 = 0.30 \]
Sketch it: a line with 70 percent of the area to its right must be left of the mean, since more than half the curve is on its right. So the z-score will be negative and k will come out under 5 minutes.
Find the z-score with left area 0.30
Why: below the mean.
\[ z = -0.5244 \]
Walk from the mean
Why: half a standard deviation down.
\[ k = 5 + (-0.5244)(2) = 5 - 1.0488 \]
Subtract and round
Why: two decimals, to match the book's choices.
\[ k = 3.9512 \approx 3.95 \]
Figure (svg): A bell curve of minutes to find a parking space, mean 5 and standard deviation 2, x ticks minus 1 to 11 and z values minus 3 to 3. A yellow line at 3.95 minutes has 0.30 of the area to its left, shaded blue, and 0.70 to its right, shaded orange and labelled more than k 0.70.
Interpret it: seventy percent of the time, finding a parking space takes more than 3.95 minutes. Most mornings take longer than that, which is why the line sits below the mean.
The book offers four choices, and the one at 6.05 is the mirror image of the right answer: it comes from entering 0.70 into invNorm. It is the most common error on this exercise, and a sketch shows the fault instantly, since only 30 percent of the time does it take longer than 6.05 minutes.
Check by running it forwards
Why: the right tail above 3.9512 must be 0.70.
\[ 1 - P(X < 3.9512) = 1 - 0.3000 = 0.70 \]
OpenStax Introductory Statistics 2e, §6.2 Using the Normal Distribution §6.2 homework, p. 357 — Exercise 72
Worked example
A motorcycle racer averages 129.71 seconds per 2.5-mile lap, with a standard deviation of 2.28 seconds, and lap times are normal. The fastest 3 percent of the racer's laps are under how many seconds?
Set up
Why: X is the time of one lap, in seconds.
\[ X \sim N(129.71, 2.28) \]
Translate
Why: fast laps are short times, so the fastest 3 percent is a left tail.
\[ P(X < k) = 0.03 \]
Picture the laps as a curve of times. The quickest laps are the smallest times, at the left end, and 3 percent of them is a thin sliver out there. That sliver is already an area to the left, so 0.03 goes straight into the look-up.
Find the z-score with left area 0.03
Why: far below the mean.
\[ z = -1.8808 \]
Walk from the mean
Why: nearly two standard deviations down.
\[ k = 129.71 + (-1.8808)(2.28) \]
Multiply
Why: 1.88 standard deviations of 2.28 seconds.
\[ k = 129.71 - 4.2882 \]
Subtract and round
Why: two decimals.
\[ k = 125.4218 \approx 125.42 \]
Figure (svg): A bell curve of lap times with mean 129.71 seconds and standard deviation 2.28, ticks one standard deviation apart. The thin left tail below a yellow line at 125.42 seconds is shaded green and labelled fastest 3 percent; a note on the right says slow laps are long times.
Interpret it: the fastest 3 percent of the racer's laps are completed in under 125.42 seconds. The word fastest pulls many students to the right tail, because top performances usually sit there. Here a top performance is a small number, so it is on the left.
Always ask what a large value of X means in the problem. For scores and heights, large is high; for times, weights lost as negative numbers, or golf scores, large can be bad.
Check against the Empirical Rule
Why: 3 percent is a bit more than the 2.5 percent beyond two standard deviations, so the answer should be a bit less than two standard deviations below the mean.
\[ 129.71 - 2(2.28) = 125.15 < 125.42 < 129.71 \]
OpenStax Introductory Statistics 2e, §6.2 Using the Normal Distribution §6.2 homework, p. 358 — Exercise 80(d)
Step zero
Four questions from the homework, all needing a different first move. For each, before any arithmetic, decide whether it asks for a probability or a value, and write the area to the left that you would use.
Discussion prompt
For each question, say probability or value, and give the left area or left areas involved. (a) IQ is N(100, 15); MENSA takes the top 2 percent: find the minimum IQ. (b) Trials last N(21, 7) days; find the probability a trial lasts at least 24 days. (c) Children spend N(3, 1.5) hours alone; seventy percent spend at least how long? (d) Pregnancy lengths are N(280, 13); find the probability of a length below 240 or above 306.
The first move is always the same pair of questions: is a line given or an area given, and which side of the line is the region? Everything after that is mechanical.
Part (b) is worth a note on rounding. The book prints 0.3336, which is what a table gives with the z-score 3 over 7 rounded to 0.43; the unrounded answer is 0.3341. Neither is wrong, but know which one you are reporting.
OpenStax Introductory Statistics 2e, §6.2 Using the Normal Distribution §6.2 homework, pp. 357-360 — Exercises 74(c), 77(e), 79(c) and 84
Trap
Mandarin diameters are N(5.85, 0.24) cm. What diameter marks the top 10 percent? A first attempt enters the 10 percent directly:
\[ \text{invNorm}(0.10,\ 5.85,\ 0.24) \]
\[ = 5.54 \text{ cm} \]
Wrong: 5.54 cm is below the mean, so 90 percent of oranges are bigger.
invNorm always wants the area to the left. The top 10 percent is on the right, so 90 percent is to its left.
\[ P(X < k) = 0.90 \]
\[ k = 5.85 + (1.2816)(0.24) \]
\[ k = 6.16 \text{ cm} \]
A top anything must sit above the mean. That one sentence catches the error before the sketch does. This value is Example 6.12(c)'s 90th percentile: ninety percent of the oranges have a diameter of at most 6.16 cm.
OpenStax Introductory Statistics 2e, §6.2 Using the Normal Distribution §6.2, p. 346 — Example 6.12(c)
Discrimination
One last sort before the middle regions. For each phrase, decide whether the percentage it mentions is already the area to the left, or whether you must subtract it from one first.
Sort into buckets
Sort each phrase by what you enter as the area to the left.
Two of these end at the same left area, 0.60, from opposite directions: 60 percent within k, and 40 percent at least k. The numbers in the phrases differ, the regions are identical.
If you are unsure, sketch the region the phrase describes and ask whether its area is on the left of the line. That question never fails.
Section
Part 4
Concept
Symmetry gives a shortcut that the rest of this part leans on. On the standard normal curve, the area to the left of minus z equals the area to the right of z: the two tails are reflections of each other.
\[ P(Z < -z) = P(Z > z) = 1 - P(Z < z) \]
Read that as a statement about percentiles. If z is the pth percentile, with left area p, then minus z has left area one minus p. So the percentiles p and one minus p have z-scores that are negatives of each other.
\[ z_{0.10} = -1.2816, \qquad z_{0.90} = 1.2816 \]
\[ z_{0.25} = -0.6745, \qquad z_{0.75} = 0.6745 \]
Translate back to X with k equals μ plus zσ. The two percentiles sit the same distance from the mean, one below and one above, so their average is the mean itself.
\[ k_{p} = \mu - |z|\sigma, \quad k_{1-p} = \mu + |z|\sigma, \quad \frac{k_{p} + k_{1-p}}{2} = \mu \]
That is why only one inverse calculation is ever needed for a middle region, and why every middle-region answer in this deck is checked by averaging its two ends. It also explains Example 6.8(a)'s 0.3446 reappearing as Example 6.9(a)'s lower left area: z of 0.4 and minus 0.4 are mirror images.
OpenStax Introductory Statistics 2e, §6.2 Using the Normal Distribution §6.2, pp. 344-346 — Examples 6.9, 6.11 and 6.12
Concept
Example 6.12(b) asks for the diameters between which the middle 20 percent of mandarins lie. A middle region is centred on the mean, so the rest of the area splits equally between the two tails. That gives both cut-offs as percentiles.
\[ \text{tails together} = 1 - 0.20 = 0.80 \]
\[ \text{each tail} = \frac{0.80}{2} = 0.40 \]
Figure (svg): The mandarin-diameter curve for mean 5.85 and standard deviation 0.24. A narrow yellow band in the middle between pink lines at 5.79 and 5.91 is labelled 0.20; each side of it is shaded pale blue and labelled tail 0.40.
The lower cut-off has one tail, 0.40, to its left: it is the 40th percentile. The upper cut-off has that tail plus the middle, 0.40 plus 0.20, to its left: it is the 60th percentile.
\[ k_{1}: \ P(X < k_{1}) = 0.40, \qquad k_{2}: \ P(X < k_{2}) = 0.60 \]
In general, a middle share p has cut-offs at the percentiles one minus p over two, and one plus p over two. For the middle 50 percent those are the 25th and 75th percentiles, the quartiles; for the middle 80 percent, the 10th and 90th.
The picture shows why the answer is such a narrow band. The middle is where the curve is tallest, so 20 percent of the area fits into barely half a standard deviation.
OpenStax Introductory Statistics 2e, §6.2 Using the Normal Distribution §6.2, p. 346 — Example 6.12(b)
Worked example
A citrus farmer finds mandarin diameters are normal with mean 5.85 cm and standard deviation 0.24 cm. The middle 20 percent of mandarins have diameters between which two values?
Set up
Why: X is the diameter of one mandarin, in cm.
\[ X \sim N(5.85, 0.24) \]
Split the leftover area
Why: 80 percent outside the middle, 40 percent in each tail.
\[ \frac{1 - 0.20}{2} = 0.40 \]
Name the two cut-offs as percentiles
Why: the 40th, and the 40th plus the middle.
\[ k_{1} = 40\text{th}, \qquad k_{2} = 0.40 + 0.20 = 60\text{th} \]
Write down both percentiles before computing either. A common slip is to use 0.20 and 0.80, the numbers that appear in the problem; that band would hold the middle 60 percent, three times too much. The tails come first, then the middle is added on top.
Find their z-scores
Why: mirror images: 0.2533 either side of zero.
\[ z_{1} = -0.2533, \qquad z_{2} = 0.2533 \]
Walk from the mean
Why: 0.2533 standard deviations of 0.24 cm is 0.0608 cm.
\[ k_{1} = 5.85 - 0.0608 = 5.7892 \]
And the other way
Why: the same distance above.
\[ k_{2} = 5.85 + 0.0608 = 5.9108 \]
Round
Why: two decimals, as the book does.
\[ k_{1} \approx 5.79 \text{ cm}, \qquad k_{2} \approx 5.91 \text{ cm} \]
\[ \text{invNorm}(0.40,\ 5.85,\ 0.24) = 5.79, \quad \text{invNorm}(0.60,\ 5.85,\ 0.24) = 5.91 \]
Interpret it: the middle 20 percent of mandarins from this farm have diameters between 5.79 cm and 5.91 cm. Only the first cut-off needs a calculation; the second is its mirror image across the mean.
Check the area and the centre
Why: the band must hold 0.20 and be centred on 5.85.
\[ P(X < 5.9108) - P(X < 5.7892) = 0.6 - 0.4 = 0.20, \quad \frac{5.79 + 5.91}{2} = 5.85 \]
OpenStax Introductory Statistics 2e, §6.2 Using the Normal Distribution §6.2, p. 346 — Example 6.12(b)
Worked example
Smartphone users' ages are X ~ N(36.9, 13.9). Calculate the interquartile range, rounding the answer to one decimal place.
Recall the definition
Why: the IQR is the width of the middle 50 percent.
\[ \text{IQR} = Q_{3} - Q_{1} \]
Name the quartiles as percentiles
Why: the middle 50 percent leaves 25 percent in each tail.
\[ Q_{1} = 25\text{th}, \qquad Q_{3} = 75\text{th} \]
Find the z-scores
Why: mirror images.
\[ z = -0.6745, \qquad z = 0.6745 \]
Because the quartiles are mirror images, one multiplication gives the distance for both: 0.6745 standard deviations of 13.9 years. Everything after that is adding and subtracting the same number from the mean.
Walk to the third quartile
Why: 0.6745 standard deviations of 13.9 years is 9.3754.
\[ Q_{3} = 36.9 + 9.3754 = 46.2754 \]
And to the first
Why: the same distance below.
\[ Q_{1} = 36.9 - 9.3754 = 27.5246 \]
Subtract
Why: the IQR, rounded to one decimal.
\[ \text{IQR} = 46.2754 - 27.5246 = 18.7508 \approx 18.8 \]
Figure (svg): The smartphone-age curve cut by orange lines at Q1 equals 27.52 and Q3 equals 46.28 years, plus the mean at 36.9, into four regions each labelled 0.25: two pink tails and two yellow middle pieces. A bracket below the axis from Q1 to Q3 is labelled IQR equals 18.75 years.
Interpret it: the middle half of smartphone users, from 27.5 to 46.3 years old, span an age range of about 18.8 years. The book prints 18.8, the one-decimal rounding of 18.7508 that node gives.
Look at how the IQR compares with the standard deviation: 18.75 against 13.9. The next slide shows that ratio is the same for every normal distribution.
Check with the standard-normal shortcut
Why: the IQR is twice 0.6745 standard deviations.
\[ 2(0.6745)(13.9) = 1.349(13.9) = 18.75 \]
OpenStax Introductory Statistics 2e, §6.2 Using the Normal Distribution §6.2, p. 345 — Example 6.11(a)
Picture it
The quartiles of the standard normal are at minus 0.6745 and 0.6745. Every normal curve is that curve with a different ruler under it, so every normal distribution has its quartiles 0.6745 standard deviations either side of its mean.
Figure (svg): The standard normal curve with the band from minus 0.6745 to 0.6745 shaded and labelled middle 50 percent. Under the axis, a bracket over the band reads IQR equals 1.349 sigma, and a wider bracket from minus 2 to 2 reads 95 percent band equals 4 sigma.
\[ Q_{1} = \mu - 0.6745\sigma, \qquad Q_{3} = \mu + 0.6745\sigma \]
\[ \text{IQR} = 2(0.6745)\sigma \approx 1.349\sigma \]
So for a normal distribution the IQR carries the same information as σ, just scaled. Knowing one gives the other. Try It 6.11 checks this: exam scores with σ equal to 15 have an IQR of 20.23 points, and 1.349 times 15 is 20.23.
That ratio is also a quick test of whether data look normal. Chapter 2's box plots gave you a sample IQR and a sample standard deviation; if their ratio is far from 1.35, the bell curve is probably the wrong model. A skewed data set can have a much smaller or larger ratio.
Compare the two brackets in the picture: half the data fit within about two thirds of a standard deviation of the mean, while 95 percent need two whole standard deviations. The tails are long and thin.
Worked example
Two thousand students took an exam whose scores are approximately normal with mean 81 points and standard deviation 15 points. Find (a) the first and third quartile scores, and (b) the two values between which the middle 50 percent of scores lie.
Set up
Why: X is one student's score.
\[ X \sim N(81, 15) \]
(a) First quartile: walk 0.6745 standard deviations down
Why: 0.6745 times 15 is 10.1173.
\[ Q_{1} = 81 - 10.1173 = 70.8827 \approx 70.88 \]
Third quartile: the same distance up
Why: the mirror image.
\[ Q_{3} = 81 + 10.1173 = 91.1173 \approx 91.12 \]
So a quarter of the students scored 70.88 or less, and a quarter scored 91.12 or more.
(b) Middle 50 percent
Why: 25 percent in each tail: exactly the quartiles.
\[ 70.88 < X < 91.12 \]
Part (b) is part (a) in different words. The middle half of any distribution runs from the first quartile to the third; the question is testing whether you recognise the phrase, not asking for new work.
The number of students, two thousand, plays no part in the answer. The percentiles depend only on the mean and standard deviation of the distribution; the count tells you roughly how many students are in each quarter, about 500.
Check with the IQR shortcut
Why: the width must be 1.349 standard deviations.
\[ 91.12 - 70.88 = 20.24 \approx 1.349(15) = 20.23 \]
OpenStax Introductory Statistics 2e, §6.2 Using the Normal Distribution §6.2, p. 346 — Try It 6.11
Worked example
Using the mandarins of Example 6.12, X ~ N(5.85, 0.24): (a) the middle 40 percent are between which diameters? (b) Find the 16th percentile and interpret it.
(a) Split the leftover area
Why: 60 percent outside, 30 percent in each tail.
\[ \frac{1 - 0.40}{2} = 0.30 \]
The cut-offs are the 30th and 70th percentiles
Why: z of minus and plus 0.5244.
\[ 0.5244 \times 0.24 = 0.1259 \]
Walk both ways from the mean
Why: round to two decimals.
\[ 5.85 - 0.1259 \approx 5.72, \qquad 5.85 + 0.1259 \approx 5.98 \]
So the middle 40 percent of mandarins have diameters between 5.72 and 5.98 cm, a wider band than the middle 20 percent, as it must be.
Part (b) returns to a single percentile. Sixteen percent is less than half, so expect a diameter below 5.85, and 16 is a share you have met before: it is roughly the area below one standard deviation under the mean.
(b) The 16th percentile
Why: left area 0.16, z of minus 0.9945.
\[ k = 5.85 + (-0.9945)(0.24) = 5.85 - 0.2387 \approx 5.61 \]
Interpret it: 16 percent of the mandarins from this farm have a diameter of 5.61 cm or less. And 5.61 is almost exactly the mean minus one standard deviation, 5.85 minus 0.24. That is no accident: below one standard deviation is the Empirical Rule's 16 percent, so the 16th percentile sits at a z-score of about minus 1.
Check (a) by its centre and (b) by the Empirical Rule
Why: the band is centred on 5.85, and the 16th percentile is close to one standard deviation down.
\[ \frac{5.72 + 5.98}{2} = 5.85, \qquad 5.85 - 0.24 = 5.61 \]
OpenStax Introductory Statistics 2e, §6.2 Using the Normal Distribution §6.2, pp. 346-347 — Try It 6.12
Comparison
Each row is a middle region from this section or its homework. Fill in the area in each tail and the lower percentile it gives; the upper percentile is filled in for you. Every row uses the same two moves: halve the leftover, then add the middle.
Comparison matrix
| middle share | each tail | lower cut-off | upper cut-off | example |
|---|---|---|---|---|
| 20% | 0.40 | 40th percentile | 60th percentile | mandarins, Ex. 6.12(b) |
| 40% | 0.30 | 30 | 70th percentile | mandarins, Try It 6.12(a) |
| 50% | 0.25 | 25 | 75th percentile | IQs, Homework 74(d) |
| 80% | 0.10 | 10 | 90th percentile | lap times, Homework 80(e) |
Check one row with numbers. For the lap times the cut-offs are the 10th and 90th percentiles, with z-scores minus and plus 1.2816. That is 2.92 seconds either side of the mean, so the middle 80 percent of laps run from 126.79 to 132.63 seconds.
The pattern: a middle share p leaves one minus p over two in each tail, and that tail is the lower percentile. The upper percentile is the lower one plus p, which is also one minus the tail.
Homework 74(d) gives IQs between 89.88 and 110.12 for the middle 50 percent of N(100, 15), and Homework 80(e) gives lap times from 126.79 to 132.63 seconds for the middle 80 percent. Both are the mean plus and minus the z-score of the upper percentile times σ.
OpenStax Introductory Statistics 2e, §6.2 Using the Normal Distribution §6.2 homework, pp. 357-358 — Exercises 74(d) and 80(e)
Counterexample
For a normal curve, the middle 20 percent is the narrowest band holding 20 percent of the area, because it sits under the peak. Push on that claim: when would a band holding 20 percent not be centred on the mean, and would the answer to a middle question still be symmetric?
Discussion prompt
For the mandarins, X ~ N(5.85, 0.24), compare the width of the middle 20 percent with the width of the band from the 50th to the 70th percentile, which also holds 20 percent. Then say why a middle-percent answer for a skewed distribution would not be symmetric about the mean.
The shortcut of computing one cut-off and reflecting it is a normal-distribution fact, not a general one. It is safe here because every normal curve is symmetric about μ; Part 5 shows real data that are only roughly so.
The definition of middle p percent, though, never changes: it is always the band from the percentile one minus p over two to the percentile one plus p over two.
OpenStax Introductory Statistics 2e, §6.2 Using the Normal Distribution §6.2, p. 346 — Example 6.12(b) and Try It 6.12(a)
Section
Part 5
Concept
Example 6.10(c) asks you to find the 80th percentile and interpret it in a complete sentence. The book's answer is two sentences long, and both matter: the number, and what it means for the people in the problem.
\[ k = 36.9 + (0.8416)(13.9) = 36.9 + 11.6985 = 48.5985 \approx 48.6 \]
The 80th percentile is 48.6 years. Eighty percent of the smartphone users in the age range 13 to 55 and over are 48.6 years old or less.
| kind of answer | the sentence names | example |
|---|---|---|
| a probability | the event, in context, and the number | the probability that a random user is between 23 and 64.7 years old is 0.8186 |
| a percentile | the share, the direction, the value with units | 80% of users are 48.6 years old or less |
| a middle region | the share and both ends with units | the middle 20% of mandarins are 5.79 to 5.91 cm across |
A bare 48.6 answers nothing. Is it an age, a probability, a z-score? The sentence settles it by naming the variable and the units, and it forces you to notice if your number makes no sense, such as a probability of 48.6 or an age of 0.80.
Every worked example in this deck ends its reasoning with such a sentence. Make it a habit; it is where the marks for interpretation are, and later chapters build their conclusions from exactly this kind of sentence.
OpenStax Introductory Statistics 2e, §6.2 Using the Normal Distribution §6.2, p. 345 — Example 6.10(c)
Worked example
Ages of smartphone users aged 13 to 55 and over are approximately X ~ N(36.9, 13.9). Find (a) the probability that a random user is between 23 and 64.7 years old and (b) the probability that a user is at most 50.8 years old.
The two parts need different moves. Part (a) names a band, between 23 and 64.7, so it needs two left areas and a subtraction. Part (b) says at most, which is already an area to the left, so it needs a single look-up.
(a) Write the band as a difference
Why: upper minus lower.
\[ P(23 < X < 64.7) = P(X < 64.7) - P(X < 23) \]
Standardise both edges
Why: both land exactly on whole z-scores.
\[ \frac{64.7 - 36.9}{13.9} = 2, \qquad \frac{23 - 36.9}{13.9} = -1 \]
Read the left areas and subtract
Why: z of 2 and z of minus 1.
\[ 0.97725 - 0.15866 = 0.81859 \approx 0.8186 \]
The unrounded areas are 0.97725 and 0.15866, whose difference is 0.81859: rounding each area first gives 0.8185, so carry four or more decimals and round at the end. The book's 0.8186 is right.
(b) At most is a left area
Why: no subtraction; 50.8 is one standard deviation up.
\[ P(X \le 50.8) = P(Z < 1) = 0.8413 \]
Figure (svg): The smartphone-age curve with ticks at 9.1, 23, 36.9, 50.8 and 64.7 years and z values minus 2, minus 1, 0, 1 and 2 beneath. The region from 23 to 64.7 is shaded in three pieces labelled 34 percent, 34 percent and 13.5 percent, with a note 0.8186 shaded.
Interpret them: the probability that a random smartphone user in this range is between 23 and 64.7 years old is 0.8186, and the probability that a user is at most 50.8 years old is 0.8413.
Check both with the Empirical Rule
Why: minus 1 to 2 is two 34 percent pieces and one 13.5; below 1 is 50 plus 34.
\[ 0.34 + 0.34 + 0.135 = 0.815, \qquad 0.50 + 0.34 = 0.84 \]
OpenStax Introductory Statistics 2e, §6.2 Using the Normal Distribution §6.2, pp. 344-345 — Example 6.10(a)(b)
Worked example
With the same ages, X ~ N(36.9, 13.9), what is the probability that a randomly selected smartphone user in the range 13 to 55 and over is less than 27 years old?
Write the probability
Why: less than is a left area.
\[ P(X < 27) \]
Standardise
Why: 9.9 years below the mean.
\[ z = \frac{27 - 36.9}{13.9} = \frac{-9.9}{13.9} = -0.7122 \]
Read the left area
Why: node, from the unrounded z.
\[ P(Z < -0.7122) = 0.2382 \]
\[ \text{normalcdf}(-10^{99},\ 27,\ 36.9,\ 13.9) = 0.2382 \]
Interpret it: the probability that a randomly selected smartphone user in this age range is younger than 27 is 0.2382, a little under one in four.
Compare with Try It 6.10(a), the 30th percentile, 29.6 years. Since 27 is below the 30th percentile, fewer than 30 percent of users are younger than 27; and since 27.5 is the first quartile from Example 6.11, slightly fewer than 25 percent are. The answer 0.2382 fits both.
Checking one answer against another answer you already trust is a powerful habit. Percentiles and probabilities of the same distribution must line up, in order, like marks on the same ruler.
Check against the quartile
Why: 27 is just below Q1 of 27.52, so the left area must be just under 0.25.
\[ 27 < 27.52 \Rightarrow P(X < 27) < 0.25, \quad 0.2382 < 0.25 \]
OpenStax Introductory Statistics 2e, §6.2 Using the Normal Distribution §6.2, p. 345 — Try It 6.10(b)
Estimation
Homework 71: parking takes X ~ N(5, 2) minutes. Find the probability that it takes at least eight minutes. The book offers four answers; use the Empirical Rule to decide which is plausible before computing anything.
Predict first
Which is the probability that it takes at least eight minutes to find a space?
Correct: 0.0668
Why: Eight minutes is one and a half standard deviations above the mean. Beyond one standard deviation lies about 16 percent and beyond two about 2.5 percent, so the answer must be between those, well under 16 percent. Only 0.0668 fits.
\[ z = \frac{8 - 5}{2} = 1.5, \qquad 1 - P(Z < 1.5) = 1 - 0.9332 = 0.0668 \]
The Empirical Rule bracket did all the work: 0.0668 is the only choice between 0.025 and 0.16. The value 0.9270 is roughly the left area, the classic missed subtraction; 0.0001 would need z near 3.7; and 0.1862 sits just above the one-standard-deviation tail.
You could also answer Homework 70 this way. Less than one minute is two standard deviations below the mean, so its probability is about 2.5 percent, exactly 0.0228: rare enough that you would be surprised.
OpenStax Introductory Statistics 2e, §6.2 Using the Normal Distribution §6.2 homework, p. 357 — Exercises 70 and 71
Worked example
Mandarin diameters are X ~ N(5.85, 0.24) cm. Find (a) the probability that a random mandarin has a diameter larger than 6.0 cm and (c) the 90th percentile of diameters, interpreted in a sentence.
(a) Larger than 6 is a right tail
Why: one minus the left area.
\[ P(X > 6) = 1 - P(X < 6) \]
Standardise
Why: 0.15 cm above the mean.
\[ z = \frac{6 - 5.85}{0.24} = 0.625 \]
Read the left area and subtract
Why: node.
\[ 1 - 0.7340 = 0.2660 \]
Part (c) flips direction: an area is given, 90 percent below, and a diameter is wanted. It uses the same z-score, 1.2816, as the exam scores' 90th percentile, because the 90th percentile of every normal curve is 1.28 standard deviations above its mean.
(c) The 90th percentile
Why: z of 1.2816 from Part 2.
\[ k = 5.85 + (1.2816)(0.24) = 5.85 + 0.3076 = 6.1576 \]
Figure (svg): The mandarin-diameter curve. The region right of a yellow line at 6.0 centimetres is shaded orange and labelled P of X greater than 6 equals 0.2660. Further right, a dashed pink line marks the 90th percentile at 6.16.
Interpret both: the probability that a randomly selected mandarin from this farm is wider than 6.0 cm is 0.2660, about one in four; and 90 percent of the mandarins have a diameter of at most 6.16 cm.
The two answers must be consistent, and the picture shows it. About 26.6 percent are wider than 6.0, and only 10 percent are wider than the 90th percentile, so the 90th percentile must lie beyond 6.0. It does.
Check (a) against the Empirical Rule and (c) against (a)
Why: 0.625 standard deviations up leaves between 16 and 50 percent; and a smaller right tail needs a larger cut-off.
\[ 0.16 < 0.2660 < 0.50, \qquad 6.16 > 6.0 \]
OpenStax Introductory Statistics 2e, §6.2 Using the Normal Distribution §6.2, p. 346 — Example 6.12(a) and (c)
Worked example
A De Anza student study found heights of Asian adult males normal with mean 66 inches and standard deviation 2.5 inches. Find (b) the probability that a randomly chosen man is between 65 and 69 inches, and (c) whether you would expect to meet many over 72 inches, justified numerically.
Define X
Why: height of one randomly chosen man, in inches.
\[ X \sim N(66, 2.5) \]
(b) Standardise both edges
Why: one below the mean, three above.
\[ z = \frac{65 - 66}{2.5} = -0.4, \qquad z = \frac{69 - 66}{2.5} = 1.2 \]
Subtract the left areas
Why: upper minus lower.
\[ P(65 < X < 69) = 0.88493 - 0.34458 = 0.54035 \approx 0.5404 \]
Part (c) is a yes-or-no question that must be justified with a number. Turn would you meet many into a probability: what share of men are taller than 72 inches? If that share is tiny, the answer is no.
(c) Standardise 72
Why: 6 inches above the mean.
\[ z = \frac{72 - 66}{2.5} = 2.4 \]
The right tail
Why: one minus the left area.
\[ P(X > 72) = 1 - 0.9918 = 0.0082 \]
Figure (svg): A bell curve of heights with mean 66 inches and standard deviation 2.5, ticks from 58.5 to 73.5 and z values minus 3 to 3 beneath. The band from 65 to 69 inches is shaded yellow and labelled 0.5404. The tiny tail beyond 72 inches is shaded red and labelled 72 inches, 0.0082.
Interpret it: the probability that a randomly chosen man is between 65 and 69 inches tall is 0.5404. The probability that he is taller than 72 inches is 0.0082, fewer than one man in a hundred, so no, you would not expect to meet many.
The justify numerically instruction is the important part of (c). An answer of no, because 72 is tall, earns nothing; an answer that names the probability, or says 72 inches is 2.4 standard deviations above the mean and beyond the 95 percent band, earns full marks.
Check (c) against the Empirical Rule
Why: beyond two standard deviations is about 2.5 percent, and 2.4 is further out, so the tail must be smaller.
\[ 0.0082 < 0.025 \]
OpenStax Introductory Statistics 2e, §6.2 Using the Normal Distribution §6.2 homework, p. 357 — Exercise 73
Real world
Homework 84 is a real kind of courtroom argument. An expert witness testifies that pregnancy length is normal with mean 280 days and standard deviation 13 days. The man being sued was out of the country from 240 to 306 days before the birth, so he could be the parent only if the pregnancy lasted less than 240 days or more than 306.
Figure (svg): A bell curve of pregnancy length with mean 280 days and standard deviation 13, ticks from 241 to 319. The region between red lines at 240 and 306 is shaded pale blue and labelled away 0.9762. The left tail below 240, labelled 0.0010, and the right tail above 306, labelled 0.0228, are shaded red.
Discussion prompt
Calculate the z-scores of 240 and 306 first, then the probability that he could be the parent and the probability that he was not. How strong is this evidence, and what assumption is the whole argument resting on?
Notice the structure: two separate tails, one on each side, added together. It is the first time in this deck that the region of interest is the outside of a band rather than the inside, and it is the shape every two-sided hypothesis test in Chapter 9 will have.
The asymmetry matters too: 306 is only two standard deviations from the mean, but 240 is more than three. Almost all of the 0.0238 comes from the right tail.
OpenStax Introductory Statistics 2e, §6.2 Using the Normal Distribution §6.2 homework, p. 360 — Exercise 84
Edge cases
Homework 81 uses real data: fifty customers at a supermarket timed how long they waited in the checkout line. The exercise asks you to fit a normal model and see how well it matches. This is the boundary of everything in the section: the normal curve is a model, and a model can be tested.
Figure (svg): A histogram of the 50 checkout-line waiting times in one-minute bins from 0 to 11 minutes, highest between 4 and 7 minutes, with bin counts 1, 1, 5, 3, 8, 10, 12, 4, 2, 3, 1. A yellow normal curve with mean 5.51 and standard deviation 2.15 is drawn over it, and a dashed pink line marks 6.1 minutes.
\[ \bar{x} = 5.51, \quad s = 2.1458, \quad X \approx N(5.51, 2.15) \]
Discussion prompt
Under the model, what is the probability of waiting fewer than 6.1 minutes? What fraction of the fifty customers actually waited fewer than 6.1 minutes? Why are the two not equal, and would they be closer or further apart with ten customers?
The book's printed answer for the model is 0.6029, but that cannot be reproduced from N(5.51, 2.15): node gives 0.6081, or 0.6083 with the unrounded standard deviation. Treat 0.6029 as a misprint. The conclusion is unchanged: model and data agree to within about 0.03.
The histogram shows why the fit is decent but not perfect: one peak, roughly symmetric, but with a flatter top than a bell. Before you trust any normal probability, look at the data it claims to describe.
OpenStax Introductory Statistics 2e, §6.2 Using the Normal Distribution §6.2 homework, pp. 358-359 — Exercise 81 and Table 6.3
Section
Part 6
Pattern
Every question in this section followed the same procedure. What changed was only whether the line or the area was given, and which side of the line the region lay.
Figure (svg): A flow diagram. Top: draw the curve and shade the region. Then branch: if the region's boundary is given, it is a probability question, so find left areas and subtract; if the area is given, it is a percentile question, so turn the area into a left area and go backwards. Both lead to: check against the Empirical Rule and say it in context.
\[ P(a < X < b) = P(X < b) - P(X < a) \qquad k = \mu + z\sigma \]
Notice how little there is to memorise: the subtraction for a band, and the walk k equals μ plus zσ. Everything else is decided by the picture, which is why the book tells you to draw a new graph for every part of every problem.
The common errors in this deck, reporting the left area for a right tail, entering the top 10 percent as 0.10, standardising a band's width, answering a percentile with a probability, are all caught at the first two steps. The shading and the decision are where the marks are won.
Check
Homework Exercise 68. Shade the region before you choose.
Check your understanding
Recovery time from a surgical procedure is normal with mean 5.3 days and standard deviation 2.1 days. What is the probability of spending more than two days in recovery?
Answer: A
Why: Two days is 1.57 standard deviations below the mean, so almost all of the curve lies to its right. The left area at 2 is 0.0580, and the right tail is one minus that, 0.9420.
\[ z = \frac{2 - 5.3}{2.1} = -1.571, \qquad 1 - 0.0580 = 0.9420 \]
The sketch settles it in a second: two days is well left of the mean, so the region to its right is nearly everything. Any choice below one half is out at once.
The distractor 0.0580 is the left area itself, the error from the first error-analysis slide. Reading a question for the word more, and shading to the right, is the whole defence.
OpenStax Introductory Statistics 2e, §6.2 Using the Normal Distribution §6.2 homework, p. 356 — Exercise 68
Check
Homework Exercise 69. A value is asked for, not a probability.
Check your understanding
Recovery time is normal with mean 5.3 days and standard deviation 2.1 days. What is the 90th percentile for recovery times?
Answer: A
Why: The 90th percentile has left area 0.90, which is a z-score of 1.2816. Walk 1.2816 standard deviations of 2.1 days above the mean: 5.3 plus 2.69 is 7.99 days.
\[ k = 5.3 + (1.2816)(2.1) = 5.3 + 2.6913 = 7.9913 \approx 7.99 \]
Interpret it: 90 percent of patients recover within 7.99 days. The answer below the mean is ruled out without arithmetic, and the others are ruled out by knowing the z-score 1.28 for the 90th percentile.
Check forwards: the left area at 7.9913 days is 0.9000.
OpenStax Introductory Statistics 2e, §6.2 Using the Normal Distribution §6.2 homework, pp. 356-357 — Exercise 69
Check
Homework Exercise 79(d). Translate the phrase before anything else.
Check your understanding
The duration of a particular type of criminal trial is normal with mean 21 days and standard deviation 7 days. Sixty percent of these trials are completed within how many days?
Answer: A
Why: Completed within k days is a left area: sixty percent of trials last k days or less. The z-score with left area 0.60 is 0.2533, so k is 21 plus 0.2533 times 7, which is 22.77 days.
\[ k = 21 + (0.2533)(7) = 21 + 1.7734 = 22.77 \]
Interpret it: sixty percent of trials of this type are completed within about 22.8 days. Sixty percent is a little more than half, so the answer is a little more than the mean, which the correct choice is and the 19.23 choice is not.
The number-in-a-probability-range distractor is the percentile trap from Part 2. A value of X is always in the units of X, here days.
OpenStax Introductory Statistics 2e, §6.2 Using the Normal Distribution §6.2 homework, p. 358 — Exercise 79(d)
Explain it to yourself
The section has two kinds of question that use the same curve in opposite directions. Explaining how they fit together is the best test of whether it has landed.
Discussion prompt
In a few sentences, explain the difference between finding P(X > 65) and finding the 90th percentile for X ~ N(63, 5). Say what is given and what is found in each, which way you read the left-area curve, and why top 10 percent and at least both force a subtraction from one.
A strong answer names the given and the unknown in each case, and ties both to the single left-area curve. If you also said that normalcdf and invNorm undo each other, you have seen why every percentile can be checked by running it forwards.
If your answer only listed calculator keystrokes, try again without them. The reasoning survives without any particular tool; the keystrokes do not survive without the reasoning.
Exit ticket
Homework Exercise 75. The percent of fat calories a person in America consumes each day is normal with mean about 36 and standard deviation 10.
Discussion prompt
State the distribution. Find the probability that a person's percent of fat calories is more than 40, and the maximum for the lower quarter. Interpret each answer in a sentence, and check each against the side of the mean it must fall on.
The first part is Example 6.8(a) in disguise: 40 is 0.4 standard deviations above 36, just as 65 was above 63, so the answer is the same 0.3446. Spotting that a z-score repeats is itself a check.
The second part is the bottom quartile of Example 6.9(b). If you wrote 0.25 as your answer, you gave the area back instead of the value; the question asks for a percent of calories.
OpenStax Introductory Statistics 2e, §6.2 Using the Normal Distribution §6.2 homework, p. 357 — Exercise 75
Recap
Normal tables, calculators and computers all report one number, the area to the left of a line. Every question in this section is built from it, in one direction or the other.
| the question | the move | example |
|---|---|---|
| less than, at most | the left area itself | P(X < 65) = 0.6554 |
| more than, at least | 1 − left area | P(X > 65) = 0.3446 |
| between a and b | left area at b − left area at a | P(1.8 < X < 2.75) = 0.5886 |
| kth percentile | left area k/100, z, then μ + zσ | 90th of N(63, 5) = 69.4 |
| top p, at least k with share p | left area 1 − p | 40% at least 40.4 years |
| middle p | percentiles (1 − p)/2 and (1 + p)/2 | middle 20%: 5.79 to 5.91 cm |
| quartiles, IQR | μ ± 0.6745σ; IQR = 1.349σ | IQR of ages = 18.8 years |
The hard part is never the arithmetic. It is shading the right region, deciding whether a line or an area was given, and translating words like at least, top and within. Then check every answer against the Empirical Rule and the side of the mean, and finish with a sentence in context.
The Introductory Business Statistics text covers the same section with business contexts if you want more practice. Chapter 7 uses exactly these tools on the distribution of a sample mean, where the standard deviation shrinks as the sample grows.
OpenStax Introductory Statistics 2e, §6.2 Using the Normal Distribution §6.2, pp. 340-347 — the whole section
OpenStax Introductory Business Statistics 2e, §6.2 Using the Normal Distribution §6.2 Using the Normal Distribution — the same tools in business contexts
Want this taught 1-on-1? Alexander tutors Statistics — $55/session, free consultation.