The link between the shape of a distribution and the three measures of centre. A distribution is symmetric when a vertical line splits the histogram into mirror images, and then the mean and the median coincide; a distribution skewed to the left is pulled out to the left and its mean falls below its median; a distribution skewed to the right has its mean above its median. The mean reflects the skewing most strongly because it is the only one of the three that feels how far out a value sits, which makes the gap between the mean and the median a free measurement of asymmetry — subject to the book's own caveat that the rule is not true for every data set, with discrete data the most common exception.
Subject: Statistics · 64 slides · symbolic lesson
Open the interactive version of this deck
Title
Statistics · Chapter 2 — Descriptive Statistics
Read the shape of a distribution from its picture, then from two numbers
Objectives
Section 2.5 gave you three ways to say where the centre of a data set is. This section asks what happens to those three numbers when the data are lopsided, and turns the answer into a tool: two numbers you can already compute tell you which way a distribution leans.
Every claim on these slides rests on a picture drawn from the data and on arithmetic you can redo. The book's own data sets carry the lesson, along with its Example 2.31 on three authors and Try It 2.31 on gold medals, presidents and video games.
OpenStax Introductory Statistics 2e, §2.6 Skewness and the Mean, Median, and Mode §2.6, pp. 104-107 — the section this deck covers
Warm-up
Recall the two centres you will lean on most. The mean adds every value and shares the total equally; the median is the middle value once the data are in order.
Discussion prompt
Five friends report how many hours they slept last night: 2, 3, 3, 4 and 13. Find the mean and the median. Which one describes a typical friend better?
Hold on to that picture: one value far out on one side, the mean chasing it, the median staying with the crowd. This whole section is that observation, made precise and given names.
Notice that you could have predicted the order of the two answers without computing anything. The long value was on the high side, so the mean had to be the larger one. By the end of the deck you will run that reasoning in both directions.
OpenStax Introductory Statistics 2e, §2.6 Skewness and the Mean, Median, and Mode §2.5 — the mean and the median
Section
Part 1
Concept
A picture of data tells you three things: where it is centred, how spread out it is, and what shape it has. Sections 2.5 and 2.7 measure the first two with numbers. This section is about the third. Here is the book's first data set, sixteen values, drawn as a histogram.
\[ 4,\ 5,\ 6,\ 6,\ 6,\ 7,\ 7,\ 7,\ 7,\ 7,\ 7,\ 8,\ 8,\ 8,\ 9,\ 10 \]
Figure (svg): A histogram of the sixteen values from 4 to 10, one bar per value, each bar one unit wide. Heights from left to right are 1, 1, 3, 6, 3, 1, 1. A dashed yellow line stands at 7, through the middle of the tallest bar, and the bars on its two sides are mirror images.
Each bar is one unit wide and sits centred on its value, so the bar over 7 counts the six 7s. The tallest bar is in the middle, and the bars fall away by the same steps on both sides: one bar of three, then one of one, then one of one.
symmetrical distribution — A distribution whose histogram can be cut by a vertical line so that the shapes to the left and to the right of the line are mirror images of each other.
Symmetry is a statement about the whole picture, not about one number. You test it by folding: if the two halves land on top of each other, the distribution is symmetric about the fold.
OpenStax Introductory Statistics 2e, §2.6 Skewness and the Mean, Median, and Mode §2.6, p. 104 — Figure 2.16 and the definition of symmetry
Concept
Folding a histogram is a picture of a simpler test you can run on the data themselves. Pick the candidate fold, here 7, and ask of every value: is there another value exactly as far on the other side?
Figure (svg): A dot plot of the sixteen values on a line from 3 to 11. The six dots at 7 are yellow. Dashed arcs join 4 to 10, 5 to 9 and 6 to 8, each labelled with how far its two ends sit from 7: 3, 2 and 1.
Go outward from the fold. The values 6 and 8 are each one unit away, and there are three of each. The values 5 and 9 are two units away, one of each. The values 4 and 10 are three units away, one of each.
\[ 7 - 6 = 8 - 7 = 1, \quad 7 - 5 = 9 - 7 = 2, \quad 7 - 4 = 10 - 7 = 3 \]
Every value below 7 has a partner the same distance above it, and the six values sitting on 7 need no partner at all. That is the whole definition, checked. A single unmatched value anywhere breaks the symmetry, which is why the ends are the fastest place to look: the smallest and largest values have the least choice of partner.
Count the sides as a final check: five values below 7, five above, six on it, sixteen in all. Equal counts on the two sides are necessary for a mirror, though, as a later slide shows, they are not enough on their own.
OpenStax Introductory Statistics 2e, §2.6 Skewness and the Mean, Median, and Mode §2.6, p. 104 — a vertical line with mirror-image sides
Worked example
The book says that the mean, the median and the mode are each seven for these data. Check all three, because the point of this section is how those three numbers behave, and you should never take a claimed centre on trust.
\[ 4,\ 5,\ 6,\ 6,\ 6,\ 7,\ 7,\ 7,\ 7,\ 7,\ 7,\ 8,\ 8,\ 8,\ 9,\ 10 \qquad n = 16 \]
The plan: the mean needs the total, so group equal values and multiply; the median needs the middle position, so count along the ordered list; the mode needs only the tallest stack. Three different operations, and you will watch them agree here and disagree everywhere else.
Total the repeated values first
Why: three 6s, six 7s and three 8s; multiplying is faster than writing them out.
\[ 3(6) = 18, \quad 6(7) = 42, \quad 3(8) = 24 \]
Add every part
Why: the four single values and the three group totals.
\[ \sum x = 4 + 5 + 18 + 42 + 24 + 9 + 10 = 112 \]
Share the total equally
Why: the mean divides by the number of values.
\[ \bar{x} = \frac{112}{16} = 7 \]
Locate the median's position
Why: with sixteen values there is no single middle; it falls between two.
\[ \frac{n + 1}{2} = \frac{17}{2} = 8.5 \]
Average the 8th and 9th values
Why: counting along the ordered list, positions 6 to 11 all hold 7.
\[ M = \frac{7 + 7}{2} = 7 \]
Find the mode
Why: the value that occurs most often.
\[ 7 \text{ occurs 6 times} \quad \Rightarrow \quad \text{mode} = 7 \]
All three agree, and they agree at the fold you found by eye on the last slide. That is no coincidence, and the next slide shows why it has to happen. The book adds that this example is unimodal, with one mode, and its single mode lands on the same spot.
Check the mean by rebuilding the total
Why: sixteen equal shares of 7 must give back 112.
\[ 16 \times 7 = 112 \checkmark \]
OpenStax Introductory Statistics 2e, §2.6 Skewness and the Mean, Median, and Mode §2.6, p. 104 — Figure 2.16: mean, median and mode are each seven
Intuition
Why must the mean of a symmetric data set sit exactly on the fold? Take any mirror pair. Call the fold m and the distance d, so one value is m minus d and its partner is m plus d. Add them.
\[ (m - d) + (m + d) = m + m - d + d \]
\[ = 2m \]
The two distances cancel, so every pair adds to twice the fold, whatever d is. For the sixteen values the fold is 7, and indeed every pair adds to 14: four and ten, five and nine, six and eight, and even seven and seven.
\[ 4 + 10 = 5 + 9 = 6 + 8 = 7 + 7 = 14 \]
The sixteen values make eight pairs, each worth 14, so the total is eight fourteens and the mean is the fold:
\[ \bar{x} = \frac{8 \times 14}{16} = \frac{112}{16} = 7 \]
The median is forced there by counting rather than adding: a mirror puts as many values below the fold as above it, so the middle of the ordered list lands on the fold. That is the book's statement: in a perfectly symmetrical distribution the mean and the median are the same.
OpenStax Introductory Statistics 2e, §2.6 Skewness and the Mean, Median, and Mode §2.6, p. 104 — in a perfectly symmetrical distribution, the mean and the median are the same
Notation
The section talks about three centres at once, so fix the symbols. Hover over each piece; the book writes the median as M and the sample mean as x-bar, and the rest of the deck does the same.
Annotate
On: \( \bar{x} = \frac{\sum x}{n}, \quad M, \quad \text{symmetric} \Rightarrow \bar{x} = M \)
The arrow is the part to read carefully. It says that if a distribution is symmetric, then its mean equals its median. It does not say the reverse, and the most common misuse of this section is to run the arrow backwards.
Notice also that the mode is missing from the rule. A symmetric distribution pins the mean and the median to the fold, but it can put its peaks anywhere, as long as they come in mirror pairs. The next worked example shows exactly that.
OpenStax Introductory Statistics 2e, §2.6 Skewness and the Mean, Median, and Mode §2.6, p. 104 — the mean, the median and the mode
Worked example
The book adds a refinement: in a symmetrical distribution that has two modes, the two modes would be different from the mean and median. Test that on the smallest data set that can show it.
\[ 1,\ 1,\ 2,\ 3,\ 3 \qquad n = 5 \]
Check symmetry first, because the claim to test is about what symmetry does to the modes. Then compute the two centres symmetry controls, and finally the modes, which it does not.
Test the fold at 2
Why: pair the values outward from the middle value.
\[ 2 - 1 = 3 - 2 = 1 \quad \text{(twice)} \]
Total the values
Why: add all five.
\[ \sum x = 1 + 1 + 2 + 3 + 3 = 10 \]
Share the total
Why: five values.
\[ \bar{x} = \frac{10}{5} = 2 \]
Read the median
Why: with five values the third is the middle one.
\[ M = 2 \]
Find the modes
Why: 1 and 3 each occur twice; 2 occurs only once.
\[ \text{modes} = 1 \text{ and } 3 \]
Figure (svg): A dot plot of 1, 1, 2, 3, 3. Two dots stack at 1 and two at 3; one dot sits at 2. A dashed line at 2 is labelled mean and median 2. Labels above the two stacks read mode 1 and mode 3.
The mean and the median sit on the fold, exactly as symmetry promises. The two modes sit one unit either side of it, mirror images of each other. So a symmetric distribution need not have its peak in the middle: it can have a dip there, with two humps on the sides. Data like heights of a mixed group of adults can look like this.
Check the symmetry claim against the definition
Why: each mode has a partner mode the same distance on the other side.
\[ 2 - 1 = 3 - 2 = 1 \checkmark \]
OpenStax Introductory Statistics 2e, §2.6 Skewness and the Mean, Median, and Mode §2.6, p. 104 — a symmetrical distribution with two modes
Counterexample
The symmetry rule runs one way. Here is a data set that shows the other direction is false: its mean and median agree perfectly, but it is not symmetric.
\[ 0,\ 3,\ 3,\ 4,\ 5 \]
Figure (svg): A dot plot of 0, 3, 3, 4, 5. One dot at 0, two stacked at 3, one at 4, one at 5. A dashed yellow line at 3 marks the median; a wedge under the axis at 3 marks the mean. A red label at 6 reads no partner for 0.
Discussion prompt
Compute the mean and the median of 0, 3, 3, 4, 5. Then test the fold at 3. Is the data set symmetric?
What happened is a coincidence of totals. The long reach down to 0 is balanced, in the mean's arithmetic, by two shorter steps up to 4 and 5. The mean only sees totals of distances, so it cannot tell one long step from two short ones. The picture can.
The same thing can happen in a large data set. Two very different shapes can share a mean and a median; that is one reason Section 2.7 adds a measure of spread, and why statisticians still insist on drawing every data set before summarising it.
So whenever you see the mean and the median agree, the right conclusion is consistent with symmetry, never proof of it. Look at the picture before you say the word symmetric.
Prediction
The five values 10, 11, 12, 13, 14 are symmetric about 12: each pair, ten and fourteen, eleven and thirteen, is the same distance out, and 12 sits on the fold. A sixth value joins them.
Before you choose, think about what a single new value needs. A value away from the fold needs a partner on the other side, and there is no second new value to be that partner.
Predict first
Which sixth value keeps the six values symmetric about 12?
Correct: Another 12
Why: A value on the fold is zero distance from it, so it needs no partner. Any other single value would be left unmatched: a second 10 would need a second 14, and a 15 would need a 9. The new mean is 72 divided by 6, which is still 12.
\[ \frac{10 + 11 + 12 + 12 + 13 + 14}{6} = \frac{72}{6} = 12 \]
The lesson generalises: you can add values to a symmetric set and keep it symmetric only on the fold, or in mirror pairs. Add a 9 and a 15 together and the fold survives; add either one alone and it does not.
Section
Part 2
Concept
The book's second data set has ten values. Draw it the same way and the mirror fails at once.
\[ 4,\ 5,\ 6,\ 6,\ 6,\ 7,\ 7,\ 7,\ 7,\ 8 \]
Figure (svg): A histogram of 4, 5, 6, 6, 6, 7, 7, 7, 7, 8 with bar heights 1, 1, 3, 4, 1. The tallest bar is at 7 and the right side ends abruptly at 8, while short bars trail off to the left. Dashed lines mark the mean at 6.3 in pink, the median at 6.5 in orange and the mode at 7 in yellow.
In the book's words, the right-hand side seems chopped off compared with the left side. The tallest bar is at 7, the data stop one step later at 8, but on the left they straggle down through 6, then 5, then 4. The long thin side is called the tail.
skewed to the left — A distribution pulled out to the left: its long, thin tail stretches towards the small values. Also called negatively skewed.
Count the steps to check the picture. From the peak at 7 the data reach one unit to the right but three units to the left, and the left side thins out as it goes: three 6s, then one 5, then one 4. A side that is both longer and thinner is a tail.
The name follows the tail, not the bulk of the data. Most of these values are on the right, near 7, and yet the distribution is skewed left, because left is where it is stretched. If that sounds backwards, picture pulling on the thin end of the shape: you are pulling it to the left.
OpenStax Introductory Statistics 2e, §2.6 Skewness and the Mean, Median, and Mode §2.6, p. 104 — Figure 2.17: skewed to the left
Worked example
The book states that the mean is 6.3, the median 6.5 and the mode seven. Compute each, then put them in order, since the order is the whole point.
\[ 4,\ 5,\ 6,\ 6,\ 6,\ 7,\ 7,\ 7,\ 7,\ 8 \qquad n = 10 \]
Before computing, predict from the picture. The tail is on the left, so the mean should be dragged furthest left, the median less, and the mode should stay at the peak. Each line below either confirms that or does not.
Total the groups
Why: three 6s and four 7s.
\[ 3(6) = 18, \quad 4(7) = 28 \]
Add every part
Why: the singles 4, 5 and 8 and the two group totals.
\[ \sum x = 4 + 5 + 18 + 28 + 8 = 63 \]
Share the total
Why: ten values.
\[ \bar{x} = \frac{63}{10} = 6.3 \]
Locate the median
Why: ten values, so the middle falls between positions 5 and 6.
\[ \frac{10 + 1}{2} = 5.5 \]
Average the 5th and 6th values
Why: position 5 is the last 6, position 6 the first 7.
\[ M = \frac{6 + 7}{2} = 6.5 \]
Find the mode
Why: 7 occurs four times, more than any other value.
\[ \text{mode} = 7 \]
Put the three in order
Why: smallest first.
\[ \bar{x} = 6.3 < M = 6.5 < \text{mode} = 7 \]
The mean is less than the median, and both are less than the mode. Measure how far each is from the peak at 7: the median has moved 0.5 towards the tail and the mean 0.7. Both reflect the skewing, as the book puts it, but the mean reflects it more.
Check the mean by rebuilding the total
Why: ten shares of 6.3 must give back 63.
\[ 10 \times 6.3 = 63 \checkmark \]
OpenStax Introductory Statistics 2e, §2.6 Skewness and the Mean, Median, and Mode §2.6, pp. 104-105 — the mean is 6.3, the median is 6.5, and the mode is seven
Picture it
Why does the mean move further than the median? Think of each value as an equal weight sitting on a beam at its own position. The mean is the point where the beam balances.
Figure (svg): The left-skewed data as a dot plot resting on a beam. The orange dots at 4, 5 and the three 6s lie left of a pink wedge at 6.3; the blue dots at 7 and 8 lie right of it. Text on the left reads pull to the left, 2.3 plus 1.3 plus 3 times 0.3 equals 4.5; on the right, 4 times 0.7 plus 1.7 equals 4.5. A dashed yellow line at 6.5 is labelled median: five values each side.
Check the balance at 6.3. The values below it are 2.3, 1.3 and three times 0.3 away, a total pull of 4.5 to the left. The values above it are four times 0.7 and once 1.7 away, a total pull of 4.5 to the right. Equal pulls: the beam balances.
\[ 2.3 + 1.3 + 3(0.3) = 4.5 = 4(0.7) + 1.7 \]
Now see what the 4 does. It sits 2.3 from the balance point, further than any other value, so it has the most leverage. Slide it further left and the beam tips; the balance point must move left to compensate.
The median plays a different game. It only asks how many values are on each side, and five on each side of 6.5 stays five on each side however far the 4 slides. That is the mechanism behind every statement in the section: the mean feels distance, the median only counts.
Worked example
Eight students take a ten-point quiz that most of them found easy. One student missed most of it. The scores are below; find the three centres and the shape.
\[ 2,\ 7,\ 8,\ 8,\ 9,\ 9,\ 9,\ 10 \qquad n = 8 \]
Predict before computing. The scores pile up near the top of the scale, and the 2 straggles far down to the left, so the tail is on the left. Expect the mean below the median and the median below the mode.
Total the scores
Why: two 8s and three 9s.
\[ \sum x = 2 + 7 + 16 + 27 + 10 = 62 \]
Share the total
Why: eight scores.
\[ \bar{x} = \frac{62}{8} = 7.75 \]
Average the 4th and 5th scores
Why: eight values, so the middle falls between positions 4 and 5.
\[ M = \frac{8 + 9}{2} = 8.5 \]
Find the mode
Why: 9 occurs three times.
\[ \text{mode} = 9 \]
Order the three
Why: smallest first.
\[ 7.75 < 8.5 < 9 \]
In context: the typical student scored 8 or 9 out of 10, and the mean of 7.75 understates that because of a single score of 2. Here a quiz has a ceiling at 10 that nobody can pass, so the scores pile up against it and the only room to spread out is downwards. That is why easy tests are typically skewed left.
Check the prediction against the order
Why: mean below median below mode is the left-skewed pattern, and the tail was on the left.
\[ \bar{x} - M = 7.75 - 8.5 = -0.75 < 0 \checkmark \]
Trap
Asked to describe the quiz scores, most of which sit at 8, 9 and 10, a first attempt writes:
\[ \text{most values are high} \]
\[ \Rightarrow \text{ skewed right} \]
Wrong. The name follows the tail, not the pile.
Find the long thin side first. The scores stretch down to 2 on the left and stop at 10 on the right, so the tail points left.
\[ \text{tail on the left} \Rightarrow \text{skewed left} \]
\[ \bar{x} = 7.75 < M = 8.5 \]
The numbers confirm the name: the mean is dragged towards the tail, so it sits below the median. If the mean is below the median, the pull is to the left. Name the skew by where the tail points, and let the mean tell you if you are unsure.
OpenStax Introductory Statistics 2e, §2.6 Skewness and the Mean, Median, and Mode §2.6, p. 104 — skewed to the left because it is pulled out to the left
Ranking
Put the three centres of a typical left-skewed distribution in order from smallest to largest. Picture the left-skewed histogram: a peak on the right, a tail straggling left.
Put in order
Order the three centres of a left-skewed distribution, smallest first.
Why: The mode sits at the peak, on the right. The median has moved part of the way towards the tail, and the mean, which feels every distance, has moved furthest. For the book's data that is 6.3, then 6.5, then 7.
A way to remember it without memorising: start at the peak, where the mode is, and walk towards the tail. You meet the median first and the mean last, because the mean is the one dragged furthest out. For a left skew, walking towards the tail means walking down the number line, so the order from smallest is mean, median, mode.
The same walk works for right skew, where the tail is up the number line: mode, then median, then mean, from smallest. You never need two separate rules.
OpenStax Introductory Statistics 2e, §2.6 Skewness and the Mean, Median, and Mode §2.6, p. 105 — the mean and the median both reflect the skewing, but the mean reflects it more
Estimation
Two classes take different exams. On the hard one most students score between 30 and 60 out of 100 and a few score in the 90s. On the easy one most score between 85 and 100 and a few score below 40. Estimate before you compute anything.
Predict first
On the easy exam, how does the mean compare with the median?
Correct: The mean is below the median
Why: On the easy exam the scores pile up against the ceiling of 100 and the few low scores form a tail to the left. The tail drags the mean down, so the mean is below the median. The hard exam is the reverse: a tail to the right and a mean above the median.
Check the hard exam the same way. Its scores crowd between 30 and 60 with a few in the 90s, so the thin side is high and the mean sits above the median. A teacher quoting the mean of a hard exam makes the class look a little better than the typical student did.
Notice that you answered without a single number, from the shape alone. That is the practical value of the section: knowing the shape tells you which way the mean has been pulled and roughly how much to distrust it as a summary.
Here the ceiling did the work. A score cannot exceed 100, so on an easy exam there is no room to spread upwards and the only tail possible is downwards. Whenever a quantity is crowded against a limit, expect the tail on the side away from it.
Section
Part 3
Concept
The book's third data set is also not symmetrical, and this time it is skewed to the right.
\[ 6,\ 7,\ 7,\ 7,\ 7,\ 8,\ 8,\ 8,\ 9,\ 10 \]
Figure (svg): A histogram of 6, 7, 7, 7, 7, 8, 8, 8, 9, 10 with bar heights 1, 4, 3, 1, 1. The tallest bar is at 7 and short bars trail off to the right. Dashed lines mark the mode at 7 in yellow, the median at 7.5 in orange and the mean at 7.7 in pink.
The peak is again at 7, but now the data stop abruptly one step to its left, at 6, and straggle out to the right through 8, 9 and 10. The long thin side is on the right, so the tail points right.
skewed to the right — A distribution pulled out to the right: its long, thin tail stretches towards the large values. Also called positively skewed.
The book's sentence for this set is the mirror of the last one: of the three statistics, the mean is the largest, while the mode is the smallest. And again, the mean reflects the skewing the most.
Compare this picture with the last one. The bar heights read 1, 4, 3, 1, 1 from left to right, where the left-skewed set read 1, 1, 3, 4, 1. Read one list backwards and you get the other, which is the first hint that these two sets are reflections of each other.
Right skew is by far the more common of the two in real data, and Part 3 ends by asking why. First, the numbers.
OpenStax Introductory Statistics 2e, §2.6 Skewness and the Mean, Median, and Mode §2.6, p. 105 — Figure 2.18: skewed to the right
Worked example
Compute the three centres of the right-skewed set and order them. The book gives mean 7.7, median 7.5 and mode seven.
\[ 6,\ 7,\ 7,\ 7,\ 7,\ 8,\ 8,\ 8,\ 9,\ 10 \qquad n = 10 \]
Predict first again. This tail is on the right, so expect the order to reverse: the mode at the peak is now the smallest centre and the mean, dragged out along the tail, the largest.
Total the groups
Why: four 7s and three 8s.
\[ 4(7) = 28, \quad 3(8) = 24 \]
Add every part
Why: the singles 6, 9 and 10 and the two group totals.
\[ \sum x = 6 + 28 + 24 + 9 + 10 = 77 \]
Share the total
Why: ten values.
\[ \bar{x} = \frac{77}{10} = 7.7 \]
Average the 5th and 6th values
Why: position 5 is the last 7, position 6 the first 8.
\[ M = \frac{7 + 8}{2} = 7.5 \]
Find the mode
Why: 7 occurs four times.
\[ \text{mode} = 7 \]
Order the three
Why: smallest first.
\[ \text{mode} = 7 < M = 7.5 < \bar{x} = 7.7 \]
Walk from the peak towards the tail: mode at 7, median half a unit out, mean seven tenths out. The same walk as before, now heading up the number line. The mean is the furthest along it because the 9 and the 10 have the most leverage on the balance.
Check the median by counting
Why: five values must lie below 7.5 and five above.
\[ 6, 7, 7, 7, 7 \;\text{below}, \quad 8, 8, 8, 9, 10 \;\text{above} \checkmark \]
OpenStax Introductory Statistics 2e, §2.6 Skewness and the Mean, Median, and Mode §2.6, p. 105 — the mean is 7.7, the median is 7.5, and the mode is seven
Intuition
Compare the two lopsided sets value by value and something neat appears. Reflect each value of the left-skewed set through 7, which means replacing x by 14 minus x: the 4 becomes 10, the 5 becomes 9, each 6 becomes 8, each 7 stays 7 and the 8 becomes 6.
Figure (svg): Two dot plots on matching number lines from 3 to 11. The top one is the left-skewed set, the bottom one the right-skewed set. A dashed yellow line at 7 crosses both. Arrows run from 4 on the top line to 10 on the bottom, and from 5 to 9. The top mean is labelled 6.3 and the bottom mean 7.7.
\[ 14 - \{4, 5, 6, 6, 6, 7, 7, 7, 7, 8\} = \{10, 9, 8, 8, 8, 7, 7, 7, 7, 6\} \]
That is exactly the right-skewed set. Reflection turns a left tail into a right tail, and it carries every centre with it. The mean of the reflected values is 14 minus the old mean, because subtracting each value from 14 subtracts the total from ten fourteens:
\[ \frac{\sum (14 - x)}{10} = \frac{140 - 63}{10} = 14 - 6.3 = 7.7 \]
The median and the mode reflect the same way: 14 minus 6.5 is 7.5, and 14 minus 7 is 7. So the two sets are one example seen from both sides, which is why their gaps are equal and opposite, minus 0.2 and plus 0.2. Whatever you learn about one direction of skew, the mirror hands you for free about the other.
OpenStax Introductory Statistics 2e, §2.6 Skewness and the Mean, Median, and Mode §2.6, pp. 104-105 — Figures 2.17 and 2.18
Picture it
Real distributions with many values look less like stacks of dots and more like smooth curves. Here are the three shapes of this section as curves, with the three centres computed from each curve and marked on it.
Figure (svg): Three panels of smooth curves over 0 to 10. Left: a curve with a long tail running left and a peak near 8.4. Middle: a symmetric bell centred at 5. Right: a curve with its peak near 1.6 and a long tail running right. In each panel three dashed lines mark the mode in yellow, the median in orange and the mean in pink; in the middle panel all three coincide at 5.
In the right-skewed panel the peak, the mode, is at 1.60; the median, which splits the area under the curve in half, is at 2.14; the mean, the curve's balance point, is at 2.40. The left-skewed panel is its mirror image, with the same gaps the other way.
The order is the one you met with the dot plots, and the same walk from the peak towards the tail produces it. The median is the halfway point of the area, so it has to move into the tail a little to keep half the area on each side. The mean has to move further, because the far-out area has the most leverage.
Keep these three silhouettes in mind. Later chapters describe whole probability distributions by shape, and when they do, the first question to ask of any curve is the one this deck keeps asking: which side is the tail on?
OpenStax Introductory Statistics 2e, §2.6 Skewness and the Mean, Median, and Mode §2.6, p. 105 — skewness and symmetry become important when we discuss probability distributions
Worked example
Eight houses on one street sold last year for the prices below, in thousands of dollars. One is a large new build. Find the mean and the median, and say which you would quote to someone asking what a house on the street costs.
\[ 210,\ 230,\ 240,\ 250,\ 255,\ 260,\ 270,\ 950 \qquad n = 8 \]
Predict from the list before you add. Seven prices sit within 60 thousand dollars of each other and one is almost four times the rest, so the tail is on the right and the mean will sit well above the median.
Total the prices
Why: add all eight.
\[ \sum x = 210 + 230 + 240 + 250 + 255 + 260 + 270 + 950 \]
Simplify the sum
Why: the first seven come to 1,715.
\[ \sum x = 1715 + 950 = 2665 \]
Share the total
Why: eight houses.
\[ \bar{x} = \frac{2665}{8} = 333.125 \]
Average the 4th and 5th prices
Why: eight values, so the middle is between positions 4 and 5.
\[ M = \frac{250 + 255}{2} = 252.5 \]
Measure the gap
Why: mean minus median.
\[ \bar{x} - M = 333.125 - 252.5 = 80.625 \]
In context: the median house sold for 252,500 dollars, and seven of the eight houses sold for 270,000 dollars or less. The mean, 333,125 dollars, is above every price except the one new build. It answers the question how much money changed hands per house, not what a typical house costs. Quote the median.
Check the median by counting
Why: four prices must lie below 252.5 and four above.
\[ 210, 230, 240, 250 \;\text{below}, \quad 255, 260, 270, 950 \;\text{above} \checkmark \]
Real world
Incomes, house prices, waiting times, insurance claims, city populations and the number of followers on a social network are all famously skewed right. What do they have in common?
Discussion prompt
Name what incomes, waiting times and house prices share that pushes their tails to the right rather than the left.
The quiz was the mirror case: a ceiling at 10 and no floor worth mentioning, so the tail grew downwards. Easy tests, age at retirement and the lifetimes of well-made products tend to be skewed left for the same reason, crowding against an upper limit.
This is why news reports of income usually give the median household income rather than the mean. With a long right tail the mean is inflated by a few very large incomes and describes almost nobody, while the median is the income of the household in the middle.
Sorting
For each quantity, picture its histogram. Ask where the values crowd, whether there is a floor or a ceiling near them, and which side has room for a long tail.
Sort into buckets
Sort each quantity by the shape its distribution is most likely to have.
The dice total is the one exactly symmetric item here: the number of ways to roll each total climbs one, two, three up to six ways for a 7, then falls back down, a perfect mirror about 7. The heights are only approximately symmetric, which is what symmetric usually means for real data.
Age at death is worth a second look. Most deaths in a wealthy country come late in life, against a soft upper limit around 100, and early deaths form a thin tail down to young ages. Try It 2.31 at the end of the deck has real ages at death to test that idea on.
Section
Part 4
Concept
The book gives the reason in one sentence: the mean is affected by outliers that do not influence the median. Watch it happen. Take the right-skewed set and move its largest value from 10 out to 20.
Figure (svg): Two dot plots of the right-skewed set on lines from 5 to 21. In the top one the largest value is 10; in the bottom one it has been moved to 20, drawn in red, with an arrow from the old position. A dashed median line stands at 7.5 in both. The mean label moves from 7.7 to 8.7.
The median depends only on which values sit in positions 5 and 6. The 10 was in position 10 and the 20 is in position 10, so nothing in the middle changed:
\[ M_{\text{before}} = M_{\text{after}} = \frac{7 + 8}{2} = 7.5 \]
The mean adds every value, so the extra ten units go straight into the total:
\[ \bar{x}_{\text{before}} = \frac{77}{10} = 7.7, \qquad \bar{x}_{\text{after}} = \frac{87}{10} = 8.7 \]
Try the experiment in the other direction too. Move the 10 down to 8 instead and the total falls to 75, the mean to 7.5, and the median is still 7.5: the right tail is gone and the mean has landed back on the median.
The tail got longer, the median ignored it completely, and the mean moved a whole unit towards it. That is why, as the book says, when the distribution is skewed to the right the mean is often greater than the median, and the longer the tail the bigger the gap.
OpenStax Introductory Statistics 2e, §2.6 Skewness and the Mean, Median, and Mode §2.6, p. 105 — the mean is affected by outliers that do not influence the median
Intuition
The mean moved one unit when a value moved ten. That ratio is not a coincidence; it comes straight from the definition. Suppose one value changes by an amount D and nothing else changes.
\[ \bar{x}_{\text{new}} = \frac{\sum x + D}{n} \]
Split the fraction into the old mean and a new piece:
\[ \bar{x}_{\text{new}} = \frac{\sum x}{n} + \frac{D}{n} = \bar{x} + \frac{D}{n} \]
So the mean moves by the change divided by the number of values: the change is shared equally among all n of them. In the last slide D was 10 and n was 10, so the mean moved 1.
Two consequences are worth saying aloud. First, the mean always moves, however large n is; a far enough value can drag it anywhere. Second, in a big data set one value moves the mean only a little, which is why one outlier matters more in a sample of ten than in a sample of ten thousand. The median, in contrast, does not move at all as long as the value stays on the same side of the middle.
Worked example
A data set has twenty values. Its largest value, 30, is re-entered as 90 after a typing error is discovered. What happens to the mean and to the median?
You do not need the data. The change formula needs only the size of the change and the count, and the median needs only to know whether the middle positions are disturbed.
Find the change
Why: new value minus old value.
\[ D = 90 - 30 = 60 \]
Share it among the values
Why: the mean moves by the change over n.
\[ \frac{D}{n} = \frac{60}{20} = 3 \]
State the effect on the mean
Why: it rises by that amount.
\[ \bar{x}_{\text{new}} = \bar{x} + 3 \]
Locate the median's positions
Why: with twenty values the median averages positions 10 and 11.
\[ \frac{20 + 1}{2} = 10.5 \]
Decide whether they change
Why: the value was largest before and is largest after, so it stays in position 20.
\[ M_{\text{new}} = M \]
In context: one corrected entry moves the mean up by three units and leaves the median untouched. If the mean and the median were equal before, the data now look right-skewed by the gap test, purely because of one value; that is how a single typing error can change the description of a whole data set.
Check through the total
Why: the total rises by 60, and 60 over 20 values is 3 per value.
\[ \frac{\sum x + 60}{20} = \frac{\sum x}{20} + 3 \checkmark \]
Error analysis
Back to the symmetric sixteen values, with mean 7. A student is asked what happens to the mean if the 10 is replaced by 14. The working below looks reasonable and gets a badly wrong answer.
Annotate
On: \( \bar{x}_{\text{new}} = 7 + (14 - 10) = 7 + 4 = 11 \)
\[ \bar{x}_{\text{new}} = 7 + \frac{4}{16} = 7.25, \qquad \frac{112 + 4}{16} = \frac{116}{16} = 7.25 \]
The quick sense check would have caught this. A mean is a balance point, so it must lie between the smallest and largest values and somewhere near the crowd. A mean of 11 for data that are mostly 6, 7 and 8 fails that test at a glance.
Notice the size of the true effect: a value moved four units and the mean moved a quarter of a unit. The data set is now very slightly skewed right, with mean 7.25 and median still 7. Small, but the direction is already the one the rule predicts.
Fill the middle
Same sixteen values, mean 7 and median 7. This time the 10 is replaced by 26. Fill in the change, the new mean and the new median.
Fill in the blanks
D = 26 - 10 = 16, \quad \bar16_8} = 7 + \frac7}___ = ___, \quad M____} = ___
Why: The value changes by 16, and 16 shared among 16 values raises the mean by exactly 1, to 8. The 26 is still the largest value, so positions 8 and 9 still hold 7s and the median stays at 7. Mean 8 against median 7: the data now lean right.
A gap of one whole unit has appeared between the mean and the median, produced by one value. The picture shows why: the 26 sits sixteen units out on its own, a tail of one, and every bit of that distance goes into the mean.
Try the reverse in your head: replace the 4 by negative 12. The change is minus 16, the mean falls to 6, the median stays at 7, and the data now lean left. The size of a single value's pull depends only on how far it travels, and its direction only on which way.
Worked example
Nine people at a pharmacy counter wait the times below, in minutes. One of them arrived during a delivery. Find the mean and the median, and say what a newcomer should expect to wait.
\[ 2,\ 2,\ 3,\ 3,\ 3,\ 4,\ 4,\ 5,\ 18 \qquad n = 9 \]
Total the waits
Why: add all nine.
\[ \sum x = 2 + 2 + 3 + 3 + 3 + 4 + 4 + 5 + 18 = 44 \]
Share the total
Why: nine waits.
\[ \bar{x} = \frac{44}{9} \approx 4.89 \text{ minutes} \]
Read the median
Why: nine values, so the 5th is the middle.
\[ M = 3 \text{ minutes} \]
Count the waits below the mean
Why: every wait except 5 and 18 is under 4.89.
\[ 7 \text{ of } 9 \text{ waits} < 4.89 \]
Figure (svg): A dot plot of nine waiting times in minutes: 2, 2, 3, 3, 3, 4, 4, 5 and 18. Eight dots crowd between 2 and 5; one red dot sits alone at 18. A dashed line at 3 marks the median; a pink wedge under the axis at 4.89 marks the mean.
In context: a newcomer should expect about three minutes, the median. The mean of nearly five minutes is larger than seven of the nine actual waits, because the one 18-minute wait supplies more than a third of the total time on its own. Reporting the mean alone would make the counter look slower than it usually is.
Check the 18's share of the total
Why: its distance above the median is what lifted the mean.
\[ \frac{18}{44} \approx 0.41, \quad \frac{44 - 18}{8} = \frac{26}{8} = 3.25 \]
Without the long wait the other eight average 3.25 minutes, close to the median. One value moved the mean by more than a minute and a half.
Tweak it
Here is a family of right-skewed curves controlled by one number, k. The peak, the mode, is at k minus 1; the mean is at k; the median sits between them, at about k minus a third. Before you move the slider, predict what happens to the shape as k grows.
Parameter explorer
Slide k from 1.5 up to 12. Watch where the peak is and how lopsided the curve looks. Does the curve become more or less skewed as k grows, and what happens to the gap between the mode and the mean compared with the width of the curve?
\[ \text{mean} = {k}, \quad \text{mode} = {k} - 1, \quad \text{median} \approx {k} - \tfrac{1}{3} \]
What to notice: the gap between the peak and the mean is always one unit, yet the curve looks far less lopsided at k equal to 12 than at k equal to 2. A gap means more or less depending on how spread out the data are. A gap of 0.2 in the book's ten values, which span six units, is noticeable; the same 0.2 in data spanning six hundred units would be nothing.
So the sign of the mean minus the median tells you the direction of skew, but its size only means something next to the spread. Section 2.7's standard deviation is the yardstick that makes that comparison precise.
Concept
Put the section's findings together and you have a test that needs only two numbers you already know how to compute. The book states it carefully: when the distribution is skewed to the left the mean is often less than the median, and when it is skewed to the right the mean is often greater.
\[ \bar{x} < M: \text{ suggests skewed left} \qquad \bar{x} \approx M: \text{ suggests symmetric} \]
\[ \bar{x} > M: \text{ suggests skewed right} \]
Figure (svg): A dot chart with one row per data set. Each dot sits at that set's mean minus median, on a scale from minus 1 to 2 with a dashed line at zero. The left-skewed set sits at minus 0.2, Davis at minus 0.3, the symmetric set at 0, the right-skewed set at plus 0.2, Maris at plus 0.6, Terry at plus 0.7 and the nine waits at plus 1.89.
Here is every data set in this deck, placed by its mean minus its median. The left-skewed ones sit left of zero and the right-skewed ones right of it, and the nine waits, with their single long tail, sit furthest out. The test works for all of them except one: Maris, which Part 5 will look at closely.
For a quick reading of a news report this is often all you have. If a report gives a mean of 42 and a median of 55, the mean is 13 below the median, so the distribution has a tail to the left.
Read the word often in the book's statement seriously. The gap points the direction of skew in most data sets, but Part 6 builds one where it points the wrong way, and the reason will tell you when to distrust it.
OpenStax Introductory Statistics 2e, §2.6 Skewness and the Mean, Median, and Mode §2.6, p. 105 — the mean is often less than, or greater than, the median
Notation
The diagnostic is one subtraction, but each piece of it carries a condition. Hover over each part before you use it on real data, because the mistakes in this section come from forgetting one of them.
Annotate
On: \( \bar{x} - M > 0 \;\Rightarrow\; \text{tail probably on the right} \)
Read it as a sentence: if the mean is above the median, the distribution probably has its tail on the right. The words that do the work are the order of the subtraction and the word probably. Write the gap with its sign every time, as the deck has done, rather than just saying the two are different.
The size of the gap needs a yardstick. A gap of 0.75 in the gold medals, which run from 0 to 14, is a clear lean; a gap of 2 years in the presidents' ages, which run over almost fifty years, is a mild one. Section 2.7's standard deviation is the yardstick statisticians use.
OpenStax Introductory Statistics 2e, §2.6 Skewness and the Mean, Median, and Mode §2.6, p. 105 — when the distribution is skewed to the right, the mean is often greater than the median
Concept
It is tempting to think skew always means one wild value, like the 18-minute wait. Often it does not. Davis's letter counts are skewed left, yet none of them is unusual: the smallest, 1, is only two letters below the peak.
\[ \text{Davis: } 1,\ 1,\ 2,\ 3,\ 3,\ 3,\ 3,\ 3,\ 4,\ 4 \]
What makes the skew is the asymmetry of the two sides. Below the peak at 3 the counts thin out over two units, through 2 down to 1; above it they stop after one unit, at 4. A thin side on the left and an abrupt edge on the right is a left skew, with or without anything you would call an outlier.
The mean still feels it. Measured from the median of 3, the values below sit a total of 2 plus 2 plus 1, which is 5 units down, while the two 4s sit only 2 units up. The net pull of 3 units, shared among ten values, is exactly the 0.3 by which the mean falls below the median.
\[ \frac{2(1) - (2 + 2 + 1)}{10} = \frac{-3}{10} = -0.3 = 2.7 - 3 \]
So outliers are one way to make a tail, but a tail can be built from ordinary values thinning out on one side. Look at the shape of the whole side, not just at its most extreme point.
OpenStax Introductory Statistics 2e, §2.6 Skewness and the Mean, Median, and Mode §2.6, p. 106 — Figure 2.20: Davis' distribution has a left skew
Matching
Each summary below gives a data set's centres and nothing else. Match each one to the description it implies. Use the sign of the mean minus the median first, and the modes second.
Match the pairs
Why: A mean well below the median points to a tail on the left. A mean above the median, with the mode lowest, points to a tail on the right. Equal mean and median are consistent with symmetry: two modes placed evenly either side suggest two humps, one mode on the centre suggests a single peak.
Notice the careful words probably and consistent with. Two or three numbers can never prove a shape, as the counterexample in Part 1 showed: the mean and the median of 0, 3, 3, 4, 5 agree and the data are still lopsided. The summary narrows the possibilities; the picture settles them.
The pair of modes 5 and 9 around a centre of 7 is the bimodal pattern from Part 1: a symmetric distribution whose peaks are not at its centre.
Section
Part 5
Step zero
Statistics are used to compare, and sometimes identify, authors. Example 2.31 takes a simple random sample of ten words from each of three authors and records the number of letters in each word.
| author | letter counts |
|---|---|
| Terry | 7, 9, 3, 3, 3, 4, 1, 3, 2, 2 |
| Davis | 3, 3, 3, 4, 1, 4, 3, 2, 3, 1 |
| Maris | 2, 3, 4, 4, 4, 6, 6, 6, 8, 3 |
Figure (svg): Three dot plots of letter counts on matching lines from 1 to 10. Terry: one at 1, two at 2, four at 3, one at 4, one at 7, one at 9. Davis: two at 1, one at 2, five at 3, two at 4. Maris: one at 2, two at 3, three at 4, three at 6, one at 8.
Discussion prompt
Part (a) asks for dot plots and a comparison of the shapes. Before computing a single mean, describe each author's shape and predict whether the mean will land above or below the median.
Use the same scale for all three, as the figure does. Comparing shapes on different scales is how a mild lean comes to look dramatic, or a real tail comes to look flat.
This is step zero for every problem in the section: draw it and find the tail before you compute. The arithmetic on the next three slides then either confirms the shape or tells you something is odd, and one of these three authors does turn out to be odd.
OpenStax Introductory Statistics 2e, §2.6 Skewness and the Mean, Median, and Mode §2.6, p. 105 — Example 2.31(a) and Figure 2.19
Worked example
Find the mean and the median of Terry's ten letter counts, and compare them with the shape from the dot plot.
The dot plot predicted a mean above the median for Terry, because the thin side reaches out to 7 and 9. Sorting comes first, since the median is read from positions and the book lists the counts in the order the words were sampled.
Sort the counts
Why: the median needs the values in order.
\[ 1,\ 2,\ 2,\ 3,\ 3,\ 3,\ 3,\ 4,\ 7,\ 9 \]
Total them
Why: group the 2s and the 3s.
\[ \sum x = 1 + 2(2) + 4(3) + 4 + 7 + 9 = 37 \]
Share the total
Why: ten words.
\[ \bar{x} = \frac{37}{10} = 3.7 \text{ letters} \]
Average the 5th and 6th values
Why: both are 3.
\[ M = \frac{3 + 3}{2} = 3 \text{ letters} \]
Find the mode
Why: 3 occurs four times.
\[ \text{mode} = 3 \]
Figure (svg): Terry's ten counts as a dot plot from 0 to 10: four dots stacked at 3, smaller stacks at 1, 2 and 4, and single dots far out at 7 and 9. A dashed yellow line at 3 marks the median; a pink wedge at 3.7 marks the mean.
In context: Terry's typical word has three letters, and half the sampled words have three or fewer. The mean of 3.7 is pulled up by just two words, of seven and nine letters. The book's caption agrees with the picture: Terry's distribution has a right, or positive, skew, and mean above median is what that predicts.
Check the gap against the shape
Why: mean minus median should be positive for a right skew.
\[ \bar{x} - M = 3.7 - 3 = 0.7 > 0 \checkmark \]
OpenStax Introductory Statistics 2e, §2.6 Skewness and the Mean, Median, and Mode §2.6, pp. 105-106 — Example 2.31(b)-(c): Terry's mean is 3.7 and median three
Worked example
Now Davis. The dot plot showed a tall stack at 3, nothing above 4, and a thin straggle down to 1.
The prediction from the picture is the reverse of Terry's: a thin side on the left means a mean below the median. The arithmetic follows the same four moves: sort, total, share, then read the middle.
Sort the counts
Why: in order.
\[ 1,\ 1,\ 2,\ 3,\ 3,\ 3,\ 3,\ 3,\ 4,\ 4 \]
Total them
Why: two 1s, five 3s and two 4s.
\[ \sum x = 2(1) + 2 + 5(3) + 2(4) = 27 \]
Share the total
Why: ten words.
\[ \bar{x} = \frac{27}{10} = 2.7 \text{ letters} \]
Average the 5th and 6th values
Why: both are 3.
\[ M = \frac{3 + 3}{2} = 3 \text{ letters} \]
Find the mode
Why: 3 occurs five times.
\[ \text{mode} = 3 \]
Figure (svg): Davis's ten counts as a dot plot from 0 to 10: five dots stacked at 3, two at 4, one at 2 and two at 1. A dashed yellow line at 3 marks the median; a pink wedge at 2.7 marks the mean.
In context: Davis also writes mostly three-letter words, but his short words, a 2 and two 1s, form the long side. The book calls it a left, or negative, skew. Notice that Davis's tail is not long in absolute terms; it reaches only two units below the peak. What makes it a tail is that it is thin and on one side only, while the other side stops one unit above the peak.
Check the gap against the shape
Why: mean minus median should be negative for a left skew.
\[ \bar{x} - M = 2.7 - 3 = -0.3 < 0 \checkmark \]
OpenStax Introductory Statistics 2e, §2.6 Skewness and the Mean, Median, and Mode §2.6, pp. 105-106 — Example 2.31(b)-(c) and Figure 2.20
Worked example
Finally Maris. The book's caption calls her distribution symmetrically shaped, so the rule predicts a mean close to the median. Compute and see.
If the book's description is right, the gap should be close to zero. Keep that prediction in view while you compute, because a prediction that fails is more informative than one that succeeds.
Sort the counts
Why: in order.
\[ 2,\ 3,\ 3,\ 4,\ 4,\ 4,\ 6,\ 6,\ 6,\ 8 \]
Total them
Why: two 3s, three 4s and three 6s.
\[ \sum x = 2 + 2(3) + 3(4) + 3(6) + 8 = 46 \]
Share the total
Why: ten words.
\[ \bar{x} = \frac{46}{10} = 4.6 \text{ letters} \]
Average the 5th and 6th values
Why: position 5 is the second 4 and position 6 the third.
\[ M = \frac{4 + 4}{2} = 4 \text{ letters} \]
Find the modes
Why: 4 and 6 each occur three times, more than any other value.
\[ \text{modes} = 4 \text{ and } 6 \]
Figure (svg): Maris's ten counts as a dot plot from 0 to 10: one at 2, two at 3, three at 4, none at 5, three at 6 and one at 8. A dashed yellow line at 4 marks the median; a pink wedge at 4.6 marks the mean.
The book's answers, 4.6 and four, are right. But a gap of 0.6 is almost as large as Terry's 0.7, and Terry was clearly skewed. So either Maris is not really symmetric, or the rule is misleading here. The anomaly slide after the next one sorts out which.
Check the median by counting
Why: five values must lie at or below it and five at or above.
\[ 2, 3, 3, 4, 4 \;\text{and} \; 4, 6, 6, 6, 8 \checkmark \]
OpenStax Introductory Statistics 2e, §2.6 Skewness and the Mean, Median, and Mode §2.6, p. 106 — Example 2.31(b)-(c) and Figure 2.21: Maris' mean is 4.6, median four
Comparison
Collect the three authors' results in one table. Fill in the blanks from the last three slides; then read the table down its last two columns.
Comparison matrix
| author | mean | median | mode | mean − median | shape (book) |
|---|---|---|---|---|---|
| Terry | 3.7 | 3 | 3 | 0.7 | right skew |
| Davis | 2.7 | 3 | 3 | −0.3 | left skew |
| Maris | 4.6 | 4 | 4 and 6 | +0.6 | symmetric |
Down the gap column, the signs match the shapes for Terry and Davis: positive for the right skew, negative for the left. Maris breaks the pattern, with a positive gap almost as large as Terry's but a shape the book calls symmetric.
Down the mode column, notice that for Terry and Davis the median equals the mode. That is the observation the book draws in part (d), and it is the next slide's subject.
OpenStax Introductory Statistics 2e, §2.6 Skewness and the Mean, Median, and Mode §2.6, p. 106 — Example 2.31(b)-(c)
Concept
Part (d) asks you to describe any pattern between the shapes and the measures of centre. Put each author's three centres on one strip and the pattern is visible at a glance.
Figure (svg): Three horizontal strips from 1 to 7, one for each author. On each strip a yellow triangle marks the mode, a hollow orange ring marks the median and a pink dot marks the mean, with a pink bar underneath showing the gap from median to mean. Terry: mode 3, median 3, mean 3.7, gap plus 0.7. Davis: mode 3, median 3, mean 2.7, gap minus 0.3. Maris: modes 4 and 6, median 4, mean 4.6, gap plus 0.6.
The book's answer: it appears that the median is always closest to the high point, the mode, while the mean tends to be farther out on the tail. For Terry and Davis the median sits exactly on the mode, at 3, and the mean has moved off towards each author's tail, up for Terry and down for Davis.
Why would the median stick to the peak? The peak is where the values pile up, and a big pile in the middle of the ordered list is very likely to contain the middle positions. Terry's four 3s fill positions 4 to 7, and Davis's five 3s fill positions 4 to 8. Positions 5 and 6 are inside both piles.
The book also says that in a symmetrical distribution the mean and the median are both centrally located close to the high point. Maris is meant to illustrate that, and the strip shows the trouble: she has two high points, and the mean sits between them while the median sits on one.
OpenStax Introductory Statistics 2e, §2.6 Skewness and the Mean, Median, and Mode §2.6, p. 106 — Example 2.31(d)
Anomaly
Test Maris against the mirror-line test from Part 1. The smallest and largest values, 2 and 8, put the only possible fold halfway between them, at 5.
\[ \frac{2 + 8}{2} = 5: \quad 5 - 4 = 6 - 5 = 1 \;\text{(three each)}, \quad 5 - 3 = 2 \text{ but no } 7 \]
Predict first
Which statement best explains why Maris's mean, 4.6, is so far from her median, 4?
Correct: Her counts are not exactly symmetric, and the median must land on a stack because the counts are whole numbers
Why: The 4s and 6s pair up around 5, and the 2 pairs with the 8, but the two 3s have no partners at 7, so the data lean slightly left of 5 and the mean is 4.6 rather than 5. The median is pushed even further, all the way to 4, because positions 5 and 6 both fall inside the stack of 4s and there are no values at 5 for it to land on.
So the book's word symmetrically is a loose description of the shape, not an exact property, and two things combine. The unpartnered 3s pull the mean a little below the fold, to 4.6. The empty gap at 5 means the median cannot sit at the centre at all; with whole-number data it has to land on a stack, and it lands on the 4s.
This is exactly the book's warning that the mean-median rule is not true for every data set, with the most common exceptions in discrete data. With only a few possible values, the median jumps from stack to stack instead of sliding smoothly, so the gap can say more about where the stacks are than about the shape.
OpenStax Introductory Statistics 2e, §2.6 Skewness and the Mean, Median, and Mode §2.6, pp. 105-106 — the most common exceptions occur in sets of discrete data
Section
Part 6
Concept
The book is careful: the link between shape and the mean-median gap is an important connection, but it is not true for every data set. Here is a small, honest data set where it points the wrong way.
\[ 1,\ 2,\ 2,\ 2,\ 3,\ 3,\ 3,\ 3,\ 3,\ 6 \]
Figure (svg): A dot plot of 1, 2, 2, 2, 3, 3, 3, 3, 3, 6. Five dots stack at 3, three at 2, one at 1 and one alone at 6. A dashed yellow line at 3 marks the median; a pink wedge at 2.8 marks the mean, to the left of the median.
The long thin side is clearly on the right: the data reach three units above the peak, to 6, but only two below it, to 1. By eye this is skewed right. Now the numbers:
\[ \bar{x} = \frac{1 + 3(2) + 5(3) + 6}{10} = \frac{28}{10} = 2.8, \qquad M = \frac{3 + 3}{2} = 3 \]
The mean is below the median, which the rule would read as a left skew. What went wrong is the big block of five 3s. It fills positions 5 to 9, so the median is stuck at 3 no matter what. Meanwhile the four values below 3 together sit 4 units under it, more than the single 6 sits above it, so the mean ends up just below 3.
Move the 6 out to 10 and the mean climbs to 3.2, above the median, and the rule works again. The failure was not about the direction of the tail but about its weight: one value three units out could not outweigh four values pulling the other way.
Discrete data, with a few possible values and large stacks on them, are where this happens most. When the picture and the gap disagree, believe the picture.
OpenStax Introductory Statistics 2e, §2.6 Skewness and the Mean, Median, and Mode §2.6, p. 105 — it is not, however, true for every data set
Trap
Given only the summary of the lumpy data set, a first attempt concludes:
\[ \bar{x} = 2.8 < M = 3 \]
\[ \Rightarrow \text{ skewed left} \]
Wrong. The dot plot has its tail on the right.
The rule says the mean is often below the median in a left skew; it does not say a mean below the median proves a left skew. Look at the picture.
\[ \text{tail: } 3 \to 6 \text{ on the right} \]
\[ \text{five 3s pin } M = 3 \]
The giveaway is a big stack of one value covering the middle positions. Then the median cannot move and the gap measures the stacks, not the shape. Always draw the data before naming its shape from two numbers.
OpenStax Introductory Statistics 2e, §2.6 Skewness and the Mean, Median, and Mode §2.6, p. 105 — the most common exceptions occur in sets of discrete data
Worked example
The dot plot in Figure 2.22 shows the number of gold medals won by each of the top 20 medal-winning countries at the 2010 Winter Olympics. Read off the data, then discuss the mean, the median and the mode and whether the pattern matches the shape. The book prints no answer; these are computed from the plot.
\[ 0,0,1,1,1,1,2,2,2,3,4,4,5,5,6,6,9,9,10,14 \]
Total the medals
Why: two 0s, four 1s, three 2s, then the rest in pairs.
\[ \sum x = 0 + 4 + 6 + 3 + 8 + 10 + 12 + 18 + 10 + 14 = 85 \]
Share the total
Why: twenty countries.
\[ \bar{x} = \frac{85}{20} = 4.25 \text{ gold medals} \]
Average the 10th and 11th values
Why: twenty values, so the middle falls between positions 10 and 11.
\[ M = \frac{3 + 4}{2} = 3.5 \text{ gold medals} \]
Find the mode
Why: 1 occurs four times.
\[ \text{mode} = 1 \]
Order the three
Why: smallest first.
\[ 1 < 3.5 < 4.25 \]
Figure (svg): A dot plot of gold medals won by the top 20 medal-winning countries at the 2010 Winter Olympics, on a line from 0 to 14: two at 0, four at 1, three at 2, one at 3, two each at 4, 5 and 6, two at 9, one at 10 and one at 14. The mode 1 is marked in yellow above its stack, a dashed orange line marks the median 3.5 and a pink wedge marks the mean 4.25.
In context: half of these twenty countries won three or fewer golds, and the most common result was a single gold. A few countries won nine, ten and fourteen, and those carry the mean up to 4.25. The data have a floor at zero and a long tail to the right, and the centres come in the right-skew order: mode, median, mean.
The shape has a real cause. A country cannot win fewer than zero golds, and most of the twenty sit near that floor; only a handful of large winter-sport nations win many. A floor with no ceiling is the same pattern as incomes and waits.
Check the total by counting the dots
Why: twenty countries must appear, and the shape and the gap must agree.
\[ 2 + 4 + 3 + 1 + 2 + 2 + 2 + 2 + 1 + 1 = 20, \quad 4.25 - 3.5 = 0.75 > 0 \checkmark \]
OpenStax Introductory Statistics 2e, §2.6 Skewness and the Mean, Median, and Mode §2.6, p. 106 — Try It 2.31(a) and Figure 2.22
Worked example
Table 2.29 is a stem-and-leaf plot of the ages at which 39 former U.S. presidents died, with stems for the tens digit: 8 and 0 means 80. Find the three centres and discuss the shape. The book prints no answer.
| stem | leaves | count |
|---|---|---|
| 4 | 6 9 | 2 |
| 5 | 3 6 7 7 7 8 | 6 |
| 6 | 0 0 3 3 4 4 5 6 7 7 7 8 | 12 |
| 7 | 0 1 1 2 3 4 7 8 8 9 | 10 |
| 8 | 0 1 3 5 8 | 5 |
| 9 | 0 0 3 3 | 4 |
A stem-and-leaf plot is already a sideways histogram and already sorted, which makes the median easy: count leaves from the top. For the mean you need every age, so total row by row; 46 plus 49 is 95 for the first row, and so on.
Total each stem's row
Why: add the ages in each row.
\[ 95 + 338 + 774 + 743 + 417 + 366 \]
Add the rows
Why: the six row totals.
\[ \sum x = 2733 \]
Share the total
Why: 39 presidents.
\[ \bar{x} = \frac{2733}{39} \approx 70.08 \text{ years} \]
Find the median's position
Why: 39 values, so one middle value.
\[ \frac{39 + 1}{2} = 20 \]
Count to the 20th age
Why: 2 + 6 = 8 in the 40s and 50s, so the 20th is the 12th leaf in the 60s.
\[ M = 68 \text{ years} \]
Find the modes
Why: 57 and 67 each occur three times.
\[ \text{modes} = 57 \text{ and } 67 \]
Figure (svg): A histogram of the 39 ages at death in classes ten years wide: 40s 2, 50s 6, 60s 12, 70s 10, 80s 5, 90s 4. A dashed orange line marks the median at 68 and a dashed pink line the mean at 70.1.
In context: the middle president died at 68, and the mean age at death is about 70.1. The mean is two years above the median, a mild lean to the right: the ages reach 25 years above the median, to 93, but only 22 below it, to 46. Against a spread of almost fifty years, a two-year gap is small, so roughly symmetric with a slight right lean is the fair description.
Check the median by counting both sides
Why: 19 ages must lie below the 20th and 19 above.
\[ 2 + 6 + 11 = 19 \text{ below}, \quad 10 + 5 + 4 = 19 \text{ above} \checkmark \]
OpenStax Introductory Statistics 2e, §2.6 Skewness and the Mean, Median, and Mode §2.6, pp. 106-107 — Try It 2.31(b) and Table 2.29
Worked example
Figure 2.23 is a histogram of the hours 25 students spent playing video games on weekends, in classes five hours wide. Grouped data have lost the individual values, so estimate the mean from class midpoints, as in Section 2.5, and locate the class that holds the median.
| hours | midpoint | students | midpoint × students |
|---|---|---|---|
| 0–4.99 | 2.5 | 2 | 5 |
| 5–9.99 | 7.5 | 3 | 22.5 |
| 10–14.99 | 12.5 | 4 | 50 |
| 15–19.99 | 17.5 | 7 | 122.5 |
| 20–24.99 | 22.5 | 9 | 202.5 |
Total the midpoint products
Why: each midpoint stands in for every student in its class.
\[ 5 + 22.5 + 50 + 122.5 + 202.5 = 402.5 \]
Share among the students
Why: 2 + 3 + 4 + 7 + 9 = 25.
\[ \bar{x} \approx \frac{402.5}{25} = 16.1 \text{ hours} \]
Find the median's position
Why: 25 values, so the 13th.
\[ \frac{25 + 1}{2} = 13 \]
Run the cumulative counts
Why: the first three classes hold 9 students, the first four hold 16.
\[ 2 + 3 + 4 = 9 < 13 \le 16 = 9 + 7 \]
Name the median and modal classes
Why: the 13th student is in the fourth class; the tallest bar is the fifth.
\[ M \in [15, 20), \quad \text{modal class } [20, 25) \]
Figure (svg): A histogram of hours spent playing video games on weekends, 25 students, classes five hours wide: 0 to 4.99 has 2, 5 to 9.99 has 3, 10 to 14.99 has 4, 15 to 19.99 has 7, 20 to 24.99 has 9. The class 15 to 19.99 is shaded orange and labelled median class. A dashed pink line marks the estimated mean at 16.1.
In context: the typical student played somewhere between 15 and 20 hours, the most common class was 20 to 25 hours, and the estimated mean, 16.1 hours, sits at the bottom of the median class, pulled down by the few students who played very little. The tail is on the left and the order mean, median, mode is the left-skew order.
Remember that 16.1 is an estimate. Using midpoints assumes the students in each class are spread evenly across it; if the nine heavy players mostly played 24 hours, the true mean is a little higher. The direction of the skew does not depend on that assumption, because it is read from the bars.
Check the mean is plausible
Why: it must sit between the smallest and largest midpoints, and on the tail side of the bulk.
\[ 2.5 < 16.1 < 22.5, \quad 16.1 < 17.5 \checkmark \]
OpenStax Introductory Statistics 2e, §2.6 Skewness and the Mean, Median, and Mode §2.6, p. 107 — Try It 2.31(c) and Figure 2.23
Concept
The shape of a distribution decides which centre is honest. In a roughly symmetric distribution the mean and the median nearly agree, and the mean is usually preferred because it uses the size of every value and behaves well in the calculations of later chapters.
\[ \text{symmetric: } \bar{x} \approx M \Rightarrow \text{ either; the mean uses every value} \]
In a skewed distribution they disagree, and the median is the better description of a typical value, because it is not dragged towards the tail. That is why Section 2.5's town of fifty, with one very rich resident, was summarised by its median.
\[ \text{skewed: } \bar{x} \neq M \Rightarrow \text{ the median for a typical value} \]
The best report often gives both. The pair is itself information: a mean well above the median tells your reader there is a right tail without your having to draw it. The pharmacy should report a median wait of 3 minutes, a mean of 4.9, and the longest wait.
The book closes the section with a promise: skewness and symmetry become important when we discuss probability distributions in later chapters. The normal curve of Chapter 6 is the great symmetric shape, with its mean and median together; the exponential distribution of Chapter 5 is a standard right-skewed one, with its mean well above its median.
OpenStax Introductory Statistics 2e, §2.6 Skewness and the Mean, Median, and Mode §2.6, p. 105 — skewness and symmetry become important in later chapters
Discrimination
For each situation, decide whether the median is the better single summary, or whether the mean is fine because the data are roughly symmetric. Think about the shape first.
Sort into buckets
Sort each situation by the centre that best describes a typical value.
The founder's salary is the clearest case. With one value a hundred times the others, the mean salary may be higher than every employee's actual pay, while the median is what a typical employee earns.
The exam item depends on the word bell-shaped. A very easy or very hard exam would move to the median bucket, because it would be skewed, as the estimation slide in Part 2 showed.
Notice that in every skewed item, someone has a reason to quote the mean: it is the larger number. Knowing the shape lets you see when a reported average is flattering.
Section
Part 7
Pattern
Every problem in this section ran through the same short sequence. The arithmetic is Section 2.5's; what is new is the order in which you look and the conclusion you draw.
Figure (svg): A flow diagram of five boxes joined by arrows. Draw it: a dot plot or histogram. Find the tail: which side is long and thin. Compute: the mean and the median. Compare: the sign of mean minus median. Then two exits: if the sign agrees with the picture, name the shape and report; if it disagrees, trust the picture and look for lumps of repeated values.
\[ \text{left: } \bar{x} < M < \text{mode} \qquad \text{right: } \text{mode} < M < \bar{x} \]
Two numbers summarise; a picture describes. When you have the raw data, the picture is always available and always worth the minute it takes. When you have only a reported mean and median, as in a news story, the gap is the best evidence you have, so use it, and say probably.
Most wrong answers in this section come from skipping the first box: naming a shape from where the bulk is, or from two numbers alone. The picture takes a minute to draw and settles both.
Check
Find the tail before you look at the options. Picture the dot plot from the description.
The mean-median gap is a useful confirmation here, but start from the shape: ask where the values crowd and which side stretches out.
Check your understanding
Commute times for a company's staff crowd between 15 and 30 minutes, with a few people commuting 90 minutes or more. Which description is right?
Answer: A
Why: The few very long commutes form a thin tail stretching to the right, so the distribution is skewed right. Those long commutes have the most leverage on the balance point, which pulls the mean above the median.
In context: a typical employee commutes 15 to 30 minutes, and the mean commute overstates that. The floor at zero with no ceiling is the same pattern as incomes and waiting times.
The tempting wrong answer is the one that names the skew after the bulk. Remember the walk from the peak: the tail is where you end up, and that side gives the name.
Check
Decide, for each centre separately, whether it can move at all.
The mean shares any change among all n values. The median only notices a change that crosses the middle positions.
Check your understanding
A data set has 25 values with mean 40 and median 38. The largest value is increased by 50 and nothing else changes. What are the new mean and median?
Answer: A
Why: The change of 50 is shared among 25 values, so the mean rises by 50 divided by 25, which is 2, to 42. The largest value stays the largest, in position 25, so the 13th value is unchanged and the median stays 38.
Check through the total: 25 values with mean 40 total 1,000; adding 50 gives 1,050, and 1,050 over 25 is 42. The gap between mean and median grows from 2 to 4, so the data look more right-skewed than before.
Notice which option is ruled out instantly: a mean of 90 would sit above essentially every value in the data set, which no balance point can do.
Check
This one tests the book's caveat, not the rule itself.
Think about which kind of data set has stacks of identical values big enough to pin the median.
Check your understanding
In which situation is the mean-median rule most likely to give the wrong direction of skew?
Answer: A
Why: The book says the most common exceptions occur in discrete data. A big stack on one value can cover the middle positions and pin the median there, so the gap reflects where the stacks sit rather than which side has the tail, as in the set with five 3s and a lone 6.
The incomes option is the instructive contrast. With thousands of nearly distinct values there are no big stacks, the median can sit anywhere, and the gap tracks the shape faithfully; that is why reported mean and median incomes are such a reliable sign of right skew.
Maris's letter counts and the lumpy set with five 3s are both instances: few possible values, big stacks, and a median that jumps from stack to stack.
The practical habit that follows: when the data are small whole-number counts, always draw the dot plot before using the gap.
Explain it to yourself
This idea came up in every part of the deck. Explaining it in your own words is the best test that it has landed.
Discussion prompt
Using Terry's letter counts, mean 3.7 and median 3, explain in two or three sentences why the mean moves towards the tail and the median does not.
A strong answer names the two different operations: adding, which feels distance, and counting positions, which does not. A very strong one adds the size of the effect: moving one value by D moves the mean by D over n and leaves the median alone, unless the value crosses the middle.
If your explanation mentioned outliers, good, but notice that Terry's 7 and 9 are not especially extreme. Skew does not need dramatic outliers; a thin tail on one side is enough.
Exit ticket
A class's quiz scores have mean 6.3, median 6.5 and mode 7. A second class's scores are 6, 7, 7, 7, 7, 8, 8, 8, 9, 10.
Discussion prompt
Name the shape of the first class's scores and order its centres. Then find the mean and the median of the second class's scores and name its shape. Say in one sentence what each shape tells you about the class.
These are the book's own two lopsided sets in disguise, mirror images of each other through 7. If you named the first class's skew right because most scores are high, go back to the trap in Part 2: the tail names the skew, and the tail is wherever the mean has been dragged.
If your mean for the second class came out different from 7.7, total again in groups: one 6, four 7s, three 8s, a 9 and a 10 make 6 plus 28 plus 24 plus 9 plus 10, which is 77.
OpenStax Introductory Statistics 2e, §2.6 Skewness and the Mean, Median, and Mode §2.6, pp. 104-105 — Figures 2.17 and 2.18
Recap
Shape is the third thing a picture of data shows you, after centre and spread. A distribution is symmetric when a vertical line splits it into mirror images, and then its mean and median coincide on that line.
| shape | the tail | typical order of the centres | mean − median |
|---|---|---|---|
| symmetric | none: the sides mirror | mean = median (modes in mirror pairs) | about 0 |
| skewed left | long and thin on the left | mean < median < mode | negative |
| skewed right | long and thin on the right | mode < median < mean | positive |
The mean reflects skew the most because it adds every value, so a value far out on the tail pulls it; one value changing by D moves the mean by D over n and leaves the median unmoved. The median only counts positions and stays with the bulk, near the peak.
The gap between the mean and the median is a quick diagnostic of the direction of skew, but it is a tendency, not a law. It fails most often for discrete data with big stacks of repeated values, as Maris's letter counts and the set with five 3s showed. Draw the data first; let the numbers confirm.
The habits that carry forward matter more than any one number: find the tail before naming the shape, show the total and the middle positions when you compute, write the gap with its sign, and finish every answer with a sentence about the data in context, in their units.
The Introductory Business Statistics text covers the same section with business data, if you want more practice. Section 2.7 now measures the spread that this deck kept saying a gap should be compared with.
OpenStax Introductory Statistics 2e, §2.6 Skewness and the Mean, Median, and Mode §2.6, pp. 104-107 — the whole section
OpenStax Introductory Business Statistics 2e, §2.6 Skewness and the Mean, Median, and Mode §2.6 Skewness and the Mean, Median, and Mode — the same ideas in business contexts
Want this taught 1-on-1? Alexander tutors Statistics — $55/session, free consultation.