The vocabulary the whole course is written in: descriptive statistics as organising the data you have against inferential statistics as arguing about data you do not, with probability as the bridge; population and sample and why sampling is the only practical choice; parameter and statistic as the fixed unknown and the computed known, and the gap between them that every later chapter measures; numerical and categorical variables and the data that are their values; and the two summaries this section names, the mean for numerical data and the proportion for categorical data.
Subject: Statistics · 68 slides · symbolic lesson
Open the interactive version of this deck
Title
Statistics · Section 1.1
Population and sample, parameter and statistic, variable and data, mean and proportion
Objectives
Every later chapter of this course is written in a small vocabulary: population, sample, parameter, statistic, variable, data. This section introduces those words, and the rest of the course assumes you can use them without hesitating.
They look like definitions to memorise. They are really a way of asking, about any number you meet, who it describes and how it was found. Those two questions decide whether the number can be trusted.
Keep one habit running through every slide: whenever a number appears, ask which group of people or things it is about. Almost every mistake in this section is an answer to that question given to the wrong group.
OpenStax Introductory Statistics 2e, §1.1 Definitions of Statistics, Probability, and Key Terms §1.1, pp. 5-7 — the chapter introduction and the key terms
Warm-up
The book opens with a class exercise. Everyone writes down the average time, to the nearest half-hour, that they sleep per night, and the instructor records the answers. Here are the book's fourteen.
\[ 5, \; 5.5, \; 6, \; 6, \; 6, \; 6.5, \; 6.5, \; 6.5, \; 6.5, \; 7, \; 7, \; 8, \; 8, \; 9 \]
Discussion prompt
Before any formula: where do these answers cluster, what single number would you give as typical, and would an English class of the same size give the same list? Say why.
You have just done the two things this section is about. Summarising the fourteen answers you have is descriptive statistics; wondering what another class would say is the first step towards inferential statistics.
OpenStax Introductory Statistics 2e, §1.1 Definitions of Statistics, Probability, and Key Terms §1.1, pp. 5-6 — Collaborative Exercise and Figure 1.2
Section
Part 1
Concept
The book defines the subject in one sentence: the science of statistics deals with the collection, analysis, interpretation and presentation of data. Data are the recorded values, such as the fourteen sleep times.
The first tool it gives you is a picture. A dot plot is a number line with one dot above it for each value; repeated values stack.
Figure (svg): A dot plot of fourteen sleep times on a number line from 4.5 to 9.5 hours. Dots stack at 5 (one), 5.5 (one), 6 (three), 6.5 (four), 7 (two), 8 (two) and 9 (one). A yellow marker shows the mean, about 6.68 hours.
Look at what the picture does that the list did not. The cluster around 6 and 6.5 jumps out, the lone 9 looks unusual, and you can see the spread at a glance, from 5 hours to 9.
Notice also the book's warning about why you are learning this. The goal of statistics is not to perform many calculations; a calculator or a computer does those. The understanding of what the data say has to come from you, and the rest of the course is built to give you that.
OpenStax Introductory Statistics 2e, §1.1 Definitions of Statistics, Probability, and Key Terms §1.1, pp. 5-6 — the definition of statistics and the dot plot
Worked example
Turn the picture into numbers. Two summaries will do: how many people answered, and the average answer. Each line below is one move.
Count the dots
Why: each dot is one person, so the count is the number of answers.
\[ n = 1 + 1 + 3 + 4 + 2 + 2 + 1 = 14 \]
The letter n is the standard name for the number of values in a sample, and you will see it in nearly every formula from here on. Reading the count off the stacks is quicker than counting the list, and it is a check on the list too: fourteen values, fourteen dots.
Add the values
Why: multiply each value by how many dots sit on it, then add; repeated values are just repeated addition.
\[ 5 + 5.5 + 3(6) + 4(6.5) + 2(7) + 2(8) + 9 \]
Evaluate each piece
Why: write the products out so that nothing is skipped.
\[ 5 + 5.5 + 18 + 26 + 14 + 16 + 9 = 93.5 \]
Divide the total by the count
Why: the mean shares the total equally among the fourteen people.
\[ \bar{x} = \dfrac{93.5}{14} = 6.678\ldots \approx 6.68 \text{ hours} \]
The answer is in hours, because every value was in hours. And it sits inside the cluster you saw in the picture, a little above 6.5, because the three long sleepers at 8 and 9 pull it up.
Check by multiplying back
Why: fourteen people each sleeping the mean must use up the same total; then compare with the picture.
\[ 14 \times 6.678571 = 93.5 \]
So these fourteen people sleep about 6.68 hours a night on average. That sentence is descriptive: it is about exactly these fourteen people and nobody else.
OpenStax Introductory Statistics 2e, §1.1 Definitions of Statistics, Probability, and Key Terms §1.1, p. 6 — Figure 1.2 data
Concept
descriptive statistics — Organising and summarising data, by graphing or by using numbers such as an average.
The book names two ways to summarise: graphs, like the dot plot, and numbers, like the mean of 6.68 hours. Both describe the data actually in front of you.
\[ \text{descriptive: } \text{the 14 answers} \;\longrightarrow\; \text{graph, mean } 6.68 \]
The important property of a descriptive statement is that it is certain, apart from arithmetic slips. If you say that these fourteen people averaged 6.68 hours, you can check it against the list, and it is either right or wrong.
That certainty has a price: the statement says nothing about anyone outside the list. It cannot tell you how much college students in general sleep, or how much the class next door sleeps. For that you need the second branch of the subject, and a tool to go with it.
Chapter 2 is entirely about descriptive statistics: histograms, box plots, means, medians, standard deviations. Every one of those is a way of organising the data you already have.
OpenStax Introductory Statistics 2e, §1.1 Definitions of Statistics, Probability, and Key Terms §1.1, p. 6 — descriptive statistics
Concept
inferential statistics — Formal methods for drawing conclusions from good data about a larger group, using probability to judge how confident you can be that the conclusions are correct.
Figure (svg): A flow diagram. A large box on the left labelled population, everyone the question is about. An arrow labelled sampling leads to a smaller box, the sample. From the sample an arrow leads to a box of data, then to a box labelled descriptive statistics: graphs and numbers. A curved arrow labelled inferential statistics, using probability, leads from the descriptive box back to the population, with a question mark.
Follow the picture. You choose a sample from the population, record data, and describe them; that is the left-to-right path. Inference is the yellow arrow back: a claim about the whole population, based on the part you measured.
That arrow can miss. Your sample might be unusual, by bad luck or by a bad choice of who to ask. Inference does not pretend otherwise: it uses probability to say how far off you are likely to be, which is why the book puts probability between the two branches.
The book adds that effective inference depends on good procedures for producing data. No formula can rescue data gathered badly, and Section 1.2 is about exactly those procedures.
OpenStax Introductory Statistics 2e, §1.1 Definitions of Statistics, Probability, and Key Terms §1.1, p. 6 — inferential statistics
Sorting
Six statements about a study in which a college times 30 of its 900 gym members on a treadmill, chosen at random. For each one, ask: is it about the 30 people timed, or about people who were not timed?
Sort into buckets
Sort each statement as descriptive or inferential.
The test is not whether the sentence contains a number or the word average. Statements about the 30 and statements about all 900 can use the very same number, 21 minutes, and still be different kinds of claim.
Words like about, probably and all members are the usual signals of inference, because a claim about people you did not measure should carry some hedge. A careful writer says how big the hedge is, and that is what the later chapters teach you to compute.
Concept
probability — A mathematical tool for studying randomness: it deals with the chance, or likelihood, of an event occurring.
The book's example is a fair coin. Toss it four times and you may well not get two heads and two tails. Toss it four thousand times and the share of heads will be close to one half.
\[ P(\text{heads}) = \dfrac{1}{2} = 0.5 \]
Figure (svg): A line graph of the share of heads after each toss in a simulated run of 2,000 fair-coin tosses. The horizontal axis is the number of tosses on a log scale from 1 to about 30,000. The line swings wildly over the first ten tosses, between 0 and 1, then settles close to a dashed line at 0.5. Two extra points mark the author's 2,000 tosses at 0.498 and Pearson's 24,000 tosses at 0.5005.
The picture is a simulation of 2,000 tosses, drawn by a computer; the two extra points are the real experiments the book reports. Over the first few tosses the share of heads jumps all over the place. After hundreds it barely moves from 0.5.
That is the sense of the book's sentence: even though the outcomes of a few repetitions are uncertain, there is a regular pattern of outcomes when there are many repetitions. Probability describes that long-run pattern, not any single toss.
OpenStax Introductory Statistics 2e, §1.1 Definitions of Statistics, Probability, and Key Terms §1.1, p. 6 — Probability
Worked example
The book reports two real experiments. The statistician Karl Pearson tossed a coin 24,000 times and got 12,012 heads. One of the authors tossed a coin 2,000 times and got 996 heads.
A count of heads cannot be compared across experiments of different lengths: 12,012 heads sounds enormous beside 996 only because Pearson tossed twelve times as often. The first move is therefore to put both on the same scale, the share of tosses that came up heads.
Turn Pearson's count into a share
Why: divide the heads by the tosses, so that experiments of different sizes can be compared.
\[ \dfrac{12012}{24000} = 0.5005 \]
Do the same for the author's coin
Why: same move, same reason.
\[ \dfrac{996}{2000} = 0.498 \]
Measure each against one half
Why: subtract 0.5 to see how far each landed from the theoretical probability.
\[ 0.5005 - 0.5 = 0.0005, \qquad 0.5 - 0.498 = 0.002 \]
Say it in counts as well
Why: half of 24,000 is 12,000 and half of 2,000 is 1,000.
\[ 12012 - 12000 = 12, \qquad 1000 - 996 = 4 \]
Notice the two ways of looking. In counts, Pearson missed by more, 12 heads against 4. In shares, he came much closer, 0.0005 against 0.002. Long runs let the count drift while the share settles, and the share is what probability is about.
Hold the coin experiments up against the vocabulary of the next parts. The true probability of heads, one half, plays the role of a parameter; the observed share in a run of tosses plays the role of a statistic, and a longer run is a larger sample.
Check against the theoretical value
Why: multiply each share back by its number of tosses to recover the counts, and confirm both shares round to 0.50.
\[ 0.5005 \times 24000 = 12012, \quad 0.498 \times 2000 = 996 \]
Both experiments agree with a probability of one half for heads, to two decimal places. Neither came out exactly one half, and the book does not expect it to: exact agreement is not what probability promises.
OpenStax Introductory Statistics 2e, §1.1 Definitions of Statistics, Probability, and Key Terms §1.1, p. 6 — Pearson's 24,000 tosses and the author's 2,000
Prediction
The book says that four tosses may not give two heads and two tails. Before you work out how likely two-and-two actually is, commit to a guess.
Predict first
You toss a fair coin four times. How likely is it that you get exactly two heads?
Correct: Less likely than not
Why: There are 16 equally likely sequences of four tosses, and exactly 6 of them have two heads, so the chance is 6 out of 16, which is 0.375. More often than not, four tosses do not split evenly, even though the coin is perfectly fair.
\[ \dfrac{6}{16} = 0.375 \]
The six sequences are HHTT, HTHT, HTTH, THHT, THTH and TTHH. The other ten sequences have more heads than tails or fewer.
This is the reason the book separates a few repetitions from many. A probability of one half does not mean that every short run splits in half; it means that the share of heads settles near one half as the runs grow long. Chapter 3 builds this counting into a method.
Tweak it
The coin picture showed the share of heads settling down. This slider shows how fast. The horizontal axis is the number of tosses; the height is the typical distance between the share of heads you observe and the coin's true probability.
Parameter explorer
Start with a fair coin, probability 0.5, and read the typical distance at 100 tosses and at 2,000 tosses. Then change the coin's probability and see whether the shape of the curve changes.
\[ \text{typical distance} = \sqrt{\dfrac{{p}(1 - {p})}{n}} \]
For a fair coin the curve gives 0.05 at 100 tosses and about 0.011 at 2,000. The author's 0.498 missed by 0.002, well inside that typical distance, so it is just what a fair coin does.
The formula comes much later in the course; for now read the shape. The distance falls quickly at first and then slowly, so quadrupling the tosses only halves it. More data always helps, but each extra toss helps a little less than the one before.
Keep this curve in mind for the rest of the lesson. A sample statistic behaves exactly like the share of heads: close to its target when the sample is large and fairly chosen, scattered when it is small.
Real world
The book lists where probability shows up outside a classroom: predicting earthquakes and rain, a doctor weighing the chance that a vaccine causes the disease it prevents, a stockbroker judging a rate of return, your own decision whether to buy a lottery ticket.
Discussion prompt
Pick one of those examples. Say what the repeated event is, what a probability of, say, 0.2 would mean in the long run, and why a single outcome cannot prove the probability wrong.
This is the same lesson as the coin, in clothes you will meet again. A probability is a statement about the long run, so it is checked against long runs, never against one result.
It also explains why the book says statistics and probability work together. Probability tells you what samples tend to look like when you know the population; inferential statistics runs that reasoning backwards, from one sample you have to the population you do not.
OpenStax Introductory Statistics 2e, §1.1 Definitions of Statistics, Probability, and Key Terms §1.1, p. 6 — uses of probability
Section
Part 2
Concept
In statistics you generally want to study a population. To study it, you select a sample, a part of it, and use the sample to learn about the whole.
population — The whole collection of persons, things or objects under study: everyone or everything the question is about.
sample — A portion, or subset, of the population that is selected and studied to gain information about the population.
Figure (svg): A rectangle holding 200 small circles in a grid, labelled population. Twenty circles scattered across the grid are filled orange and labelled sample. To the right, two boxes: the parameter, a number describing all 200, unknown; and the statistic, the same number computed from the 20 orange circles, known. An arrow from the statistic to the parameter is labelled estimates.
The orange circles in the picture were chosen at random, which is why they are scattered across the whole grid rather than bunched in one corner. A sample bunched in one corner would only be able to speak for that corner.
Two words in the picture carry the weight. The population is defined by the question, not by who you managed to reach; the sample is defined by who you actually measured.
The book adds that data are the result of sampling from a population. That is the order of events: first decide the population, then choose the sample, and only then are there data. Getting the order wrong, by calling whoever you happened to measure the population, is a mistake you will see later in this lesson.
OpenStax Introductory Statistics 2e, §1.1 Definitions of Statistics, Probability, and Key Terms §1.1, p. 6 — Key Terms: population and sample
Worked example
The book's reason for sampling is plain: it takes a lot of time and money to examine an entire population. Make that concrete with an imaginary college of 2,000 first-years and a five-minute interview each.
Price the census in minutes
Why: a census measures every member, so multiply members by minutes.
\[ 2000 \times 5 = 10000 \text{ minutes} \]
Convert to hours
Why: divide by 60 minutes per hour, so the workload is readable.
\[ \dfrac{10000}{60} = 166.7 \text{ hours} \]
At a forty-hour working week that is more than four weeks of one person doing nothing but interviews, before anyone has entered or analysed a single answer. Now price the alternative the book recommends.
Price a sample of 100
Why: same two moves for the smaller group.
\[ 100 \times 5 = 500 \text{ minutes} = \dfrac{500}{60} \approx 8.3 \text{ hours} \]
Figure (svg): Two horizontal bars on a scale of hours. Interviewing all 2,000 first-years at five minutes each takes about 166.7 hours, a long blue bar. Interviewing a sample of 100 takes about 8.3 hours, a short orange bar.
A census here is about four working weeks; the sample is one working day. The book's own examples have the same shape: a school's grade point average estimated from a sample of students, and national opinion polls of 1,000 to 2,000 people standing for the whole country.
Check the ratio
Why: the sample is one twentieth of the population, so its cost should be one twentieth of the census.
\[ \dfrac{166.7}{8.3} \approx 20 = \dfrac{2000}{100} \]
And sometimes a census is not merely expensive but self-defeating. The book's drink manufacturer checks whether a 16-ounce can holds 16 ounces; measuring every can means opening every can, leaving nothing to sell.
OpenStax Introductory Statistics 2e, §1.1 Definitions of Statistics, Probability, and Key Terms §1.1, pp. 6-7 — why sampling is practical
Concept
A sample is only useful if it looks like the population in the ways that matter. The book's phrase is that the sample must contain the characteristics of the population in order to be a representative sample.
representative sample — A sample that has the same characteristics as the population it is drawn from, so that what is true of the sample is roughly true of the population.
To see what can go wrong, imagine a college so small you can see everyone: Pine College, with 20 first-years. The numbers are made up, so that the true answer is known and you can watch samples miss it.
Figure (svg): A dot plot of an imaginary college's 20 first-years by dollars spent on supplies: three at 50, five at 100, five at 150, four at 200, two at 250 and one at 300. The four students first in the bookstore line, at 200, 250, 250 and 300, are drawn in orange; the rest are blue. A yellow line marks the population mean of 150 and an orange marker the line's mean of 250.
In the picture every first-year is a dot, placed at the dollars they spent on supplies. Seven of the twenty spent 200 dollars or more, a share of 0.35. The four orange students, the first four in the bookstore line in the first week of term, all spent 200 or more.
\[ \text{population: } \dfrac{7}{20} = 0.35 \qquad \text{the line: } \dfrac{4}{4} = 1 \]
So the line is not shaped like the population. It is not anybody's fault, and nothing about the four students is unusual for them; it is simply who stands first in a bookstore queue.
OpenStax Introductory Statistics 2e, §1.1 Definitions of Statistics, Probability, and Key Terms §1.1, p. 7 — representative samples
Worked example
Because Pine College is made up, you can compute the answer a real study never sees: the mean of the whole population. Then put the line's mean beside it.
Total the population by stacks
Why: multiply each amount by how many students spent it.
\[ 3(50) + 5(100) + 5(150) + 4(200) + 2(250) + 300 \]
Evaluate each stack
Why: write out every product.
\[ 150 + 500 + 750 + 800 + 500 + 300 = 3000 \]
Divide by the 20 students
Why: the population mean shares the total equally.
\[ \mu = \dfrac{3000}{20} = 150 \text{ dollars} \]
The Greek letter mu marks this as the population's mean, the number a real study is trying to find. Here you can compute it only because the college is imaginary and every student is on the page. In real life this line is the one nobody gets to write.
Average the four in line
Why: add their amounts and divide by four.
\[ \dfrac{200 + 250 + 250 + 300}{4} = \dfrac{1000}{4} = 250 \text{ dollars} \]
Measure the miss
Why: subtract the true mean from the sample's.
\[ 250 - 150 = 100 \text{ dollars} \]
Put the miss in proportion: the line's average is 250 over 150, about 1.67 times the truth, so a college budgeting from it would overestimate its students' spending by two thirds. The students in line were real and their answers were honest; the error came entirely from who got asked.
Check both means by multiplying back
Why: each mean times its group size must return its group's total.
\[ 20 \times 150 = 3000, \qquad 4 \times 250 = 1000 \]
The arithmetic is right twice over, and the line still overstates what a Pine first-year spends by 100 dollars. Nothing in the arithmetic could have warned you; only the question of who was asked does.
Trap
Worried that four is too few, you ask the next four in line as well. They spent 200, 200, 200 and 150.
\[ \dfrac{1000 + 750}{8} = 218.75 \]
The reasoning: twice the data, so the answer must be closer to the truth.
\[ 218.75 - 150 = 68.75 \]
Closer, but still far above 150, and seven of the eight spent 200 or more.
More people chosen the same way share the same lean. The line attracts big spenders whether you take four of them or forty.
\[ \text{same method} \Rightarrow \text{same lean} \]
The fix is to change how people are chosen, so that every first-year has the same chance: names drawn from a hat.
\[ \text{fair choice} \Rightarrow \text{representative} \]
Size controls how much a sample scatters; the method controls where it is aimed.
Socratic
The book says presidential opinion polls take samples of 1,000 to 2,000 people and that the poll is supposed to represent the views of the entire country. That sounds absurd at first, since the poll asks a tiny fraction of voters.
\[ \dfrac{1500}{150000000} = 0.00001 \]
The voter count above is a round illustrative figure, not the book's. Even so, the fraction is about one voter in a hundred thousand.
Discussion prompt
If 1,500 people out of millions can represent a country, what must be true of how they were chosen? And why would 1,500 fans leaving one sports stadium not do the same job?
A cook tasting soup is the standard picture. One spoonful tells you about the whole pot, provided you stir first, and a bigger pot does not need a bigger spoon. Stirring is the fair choice; the spoonful is the sample. Section 1.2 shows how to stir.
OpenStax Introductory Statistics 2e, §1.1 Definitions of Statistics, Probability, and Key Terms §1.1, p. 7 — opinion polls of 1,000 to 2,000 people
Section
Part 3
Concept
Once you have a population and a sample, there are two kinds of summary number: one that describes the whole population, and one computed from the sample.
parameter — A numerical characteristic of the whole population, which can be estimated by a statistic.
statistic — A number that represents a property of the sample, computed from the sample data.
The book's example takes one math class as a sample of the population of all math classes. The average points earned by students in that one class is a statistic; the average points per student over all math classes is the parameter it estimates.
\[ \text{statistic (one class)} \;\longrightarrow\; \text{estimates} \;\longrightarrow\; \text{parameter (all classes)} \]
The book also says what the pair is used for: in inferential statistics you care about both, and in a later chapter you will use a sample statistic to test whether a claimed population parameter is believable. That is a hypothesis test, and it starts here.
A memory aid that works: parameter and population both begin with p, statistic and sample both begin with s. More important than the letters is the asymmetry. You know the statistic, because you computed it; you want the parameter, because it is the truth about the whole population, and you usually never learn it exactly.
OpenStax Introductory Statistics 2e, §1.1 Definitions of Statistics, Probability, and Key Terms §1.1, p. 7 — statistic and parameter
Notation
The book does not introduce the symbols in this section, but every later chapter uses them, and they follow the same rule: Greek or plain letters for the population, marked letters for the sample. Take the pair for means apart now.
Annotate
On: \( \mu = \dfrac{\text{sum of all values}}{N} \qquad \bar{x} = \dfrac{\sum x}{n} \)
The same pattern holds for proportions in later chapters: p for the proportion in the population, p-hat, a p with a little roof, for the proportion in the sample. The roof is the reminder that it was estimated.
When you read a formula, ask of every letter whether it has a bar or a hat on it. If it does, you know it from the data; if not, it is a fact about the population that the formula is trying to reach.
Worked example
Now sample Pine College fairly: write the 20 names on slips, mix them in a hat, and draw four. Do it four separate times. The draws below are made up but entirely possible.
\[ \text{A: } 50, 100, 150, 250 \quad \text{B: } 100, 200, 250, 300 \]
\[ \text{C: } 50, 150, 200, 200 \quad \text{D: } 50, 100, 100, 150 \]
A hat draw gives every first-year the same chance of being picked on each draw. That is the difference from the bookstore line, where the big spenders were far more likely to be first. Watch what the fairness does and does not buy you.
Average draw A
Why: add the four amounts and divide by four.
\[ \bar{x}_A = \dfrac{550}{4} = 137.5 \]
Average draw B
Why: same two moves.
\[ \bar{x}_B = \dfrac{850}{4} = 212.5 \]
Average draw C
Why: same again.
\[ \bar{x}_C = \dfrac{600}{4} = 150 \]
Average draw D
Why: and once more.
\[ \bar{x}_D = \dfrac{400}{4} = 100 \]
Figure (svg): A number line of dollars from 50 to 300 with a yellow vertical line at the parameter, 150. Orange dots mark the two bookstore-line statistics, 250 and 218.75. Blue dots mark four hat-draw statistics: 137.5, 212.5, 150 and 100.
Four fair samples, four different statistics, from 100 dollars to 212.5. The parameter did not move at all: it is 150 no matter which slips came out of the hat. Draw C hit 150 exactly, but no one without the full list could have known that it had.
This is what the book means when it says the accuracy of a statistic depends on how well the sample represents the population. A small fair sample can still be unlucky, as draw B was; what fairness removes is the tendency to be wrong in the same direction every time.
Check the totals and the centre
Why: each mean times four rebuilds its draw's total, and the four statistics average to the parameter.
\[ \dfrac{137.5 + 212.5 + 150 + 100}{4} = \dfrac{600}{4} = 150 \]
That the four happen to centre exactly on 150 is luck, but the tendency is not. Fair draws scatter on both sides of the parameter; the bookstore line landed on one side every time.
Comparison
Fill in the blanks to fix the differences between the two kinds of number. Each row asks one question about each.
Comparison matrix
| question | parameter | statistic |
|---|---|---|
| describes which group? | the population | sample |
| usually known? | no | yes, computed from the data |
| changes from sample to sample? | no, it is fixed | yes |
| symbol for a mean | μ | x̄ |
| Pine College example | 150 | 137.5, 212.5, 150, 100 |
The third row is the one that surprises people. The parameter is a fixed number, even though nobody knows it. The statistic is the thing that varies, because a different sample gives a different value.
That is why the book says one of the main concerns of statistics is how accurately a statistic estimates a parameter. The accuracy depends on how well the sample represents the population, which is the bookstore-line lesson in its general form.
In later chapters this table becomes the backbone of inference: a confidence interval is a range built around a statistic, and a hypothesis test checks a claimed parameter against a statistic.
OpenStax Introductory Statistics 2e, §1.1 Definitions of Statistics, Probability, and Key Terms §1.1, p. 7 — how accurately a statistic estimates a parameter
Sorting
Suppose a study is repeated next week with a fresh random sample of the same size from the same population. Some of the six key terms would come out differently; some would not.
Sort into buckets
Sort each term by whether it would change with a new sample.
The split is clean: the three terms on the population side stay put, and the three on the sample side move. If you ever find yourself saying that the parameter changed because you took a new sample, you have mixed the two sides.
This is exactly what the slider on coin tosses showed, in another setting. The probability of heads, a parameter, never changes; the share of heads in a run, a statistic, wobbles from run to run and wobbles less as the runs grow.
Counterexample
The distinction between parameter and statistic rests on the sample being only part of the population. Push on that assumption and see what happens at the edge.
Discussion prompt
A teacher records the final mark of every one of the 28 students in her class and computes their mean, 71.4. If the question is about this class, is 71.4 a parameter or a statistic? And if the question is about all students who will ever take the course, which is it then?
So the label belongs to the relationship between the number and the question, not to the number itself. The same 71.4 is a parameter for one question and a statistic for another, depending on which group the question is about.
This is why every example in this section starts with the question's own words. Change the population in the question and the same data can change their role. Pine College's mean of 150 was a parameter only because the question was about those 20 first-years.
Trap
In the book's math class of 40, 22 students are men. A report says:
\[ \text{55\% of math students are men} \]
The reasoning: the class's proportion was computed exactly, so the claim is exact.
\[ \dfrac{22}{40} = 0.55 \]
The 0.55 is exact for this class, a statistic. The claim is about all math students, a parameter.
\[ \text{this class: } 0.55 \text{ (known)} \]
\[ \text{all classes: about } 0.55 \text{ (estimated)} \]
Draw B at Pine College had a statistic 62.5 dollars from its parameter, by luck alone. A single class can miss by as much.
Say which group a number describes, and hedge any claim beyond it.
Section
Part 4
Concept
variable — A characteristic or measurement that can be determined for each member of a population, usually named by a capital letter such as X or Y.
The book's two examples set the pattern. Let X be the number of points earned by one math student at the end of a term. Let Y be a person's party affiliation, with values such as Republican, Democrat and Independent.
\[ X = \text{points earned by one student} \qquad Y = \text{party of one person} \]
Notice the phrasing, one student and one person. A variable is defined for a single member, because it is the question you put to each member in turn. That is why the book's examples always write it as the amount spent by one first-year, or whether one doctor has been sued.
A common slip is to define the variable as a summary, for instance the average amount spent. The average is a parameter or a statistic; the variable is what each individual answers. Keep them apart and the six terms fall into place.
OpenStax Introductory Statistics 2e, §1.1 Definitions of Statistics, Probability, and Key Terms §1.1, p. 7 — variables
Concept
Variables come in two kinds, and the kind decides what arithmetic is allowed.
numerical variable — A variable whose values are numbers with equal units, such as weight in pounds or time in hours.
categorical variable — A variable that places the person or thing into a category, such as party affiliation.
Figure (svg): Two panels. On the left, a number line of exam points from 70 to 95 with tick marks every 5 points, the scores 75, 86 and 92 dotted, and a bracket showing that the gap from 86 to 92 is 6 points. On the right, three separate unconnected boxes labelled Republican, Democrat and Independent, with no line joining them.
Read the two panels of the picture as two different kinds of question. On the left, how much: points, hours, pounds. On the right, which one: which party, which team, which colour.
Equal units is the phrase to hold on to. Between 86 and 92 points there are six points, and each of those points is the same size as any other point on the scale. That is what makes sums, differences and means meaningful.
The book puts the consequence bluntly. You could do math with values of X, such as averaging the points. It makes no sense to do math with values of Y: calculating an average party affiliation makes no sense.
Section 1.2 calls these two kinds quantitative and qualitative and splits the numerical kind further, into counted and measured. The idea underneath is the one on this slide.
OpenStax Introductory Statistics 2e, §1.1 Definitions of Statistics, Probability, and Key Terms §1.1, p. 7 — numerical and categorical variables
Worked example
Suppose you ignore the warning. Four voters are Republican, Democrat, Independent and Republican. A spreadsheet wants numbers, so you code the parties. These four voters and codes are made up.
Coding categories as numbers is not wrong in itself; spreadsheets and survey software do it all the time. The question is whether arithmetic on the codes means anything, and the way to find out is to try two codings and compare.
Code A: Republican 1, Democrat 2, Independent 3
Why: a natural-looking choice, in the order the book lists them.
\[ 1 + 2 + 3 + 1 = 7 \]
Average under code A
Why: divide by the four voters, as for any mean.
\[ \dfrac{7}{4} = 1.75 \]
Code B: Independent 1, Republican 2, Democrat 3
Why: just as legitimate, since the order of parties is arbitrary.
\[ 2 + 3 + 1 + 2 = 8 \]
Average under code B
Why: same division.
\[ \dfrac{8}{4} = 2 \]
Figure (svg): Two number lines from 1 to 3. On the top line, coding A puts Republican at 1, Democrat at 2 and Independent at 3; four voters (Republican, Democrat, Independent, Republican) give a mean of 1.75. On the bottom line, coding B puts Independent at 1, Republican at 2 and Democrat at 3; the same four voters give a mean of 2.
Under code A the average, 1.75, sits between Republican and Democrat; under code B it is exactly 2, which is Republican. Nothing about the voters changed. The answer was decided by an arbitrary labelling, so it describes the labelling, not the voters.
Check with a summary that does not depend on the code
Why: count the voters in each category; the result is the same under any coding.
\[ \text{Republican: } \dfrac{2}{4} = 0.5 \text{ under code A and code B} \]
That is the right summary for categorical data: the proportion in each category. Half of these four voters are Republicans, whichever numbers you wrote in the spreadsheet.
Concept
data — The actual values of the variable. They may be numbers or they may be words.
datum — A single value of the variable.
So the variable is the question, and the data are the answers. For the exam variable X, the data might be 86, 75 and 92; one of those, say 86, is a datum. For party affiliation Y, the data are words, such as Democrat, and a single word is a datum.
\[ X = \text{exam points}: \quad 86, \; 75, \; 92 \quad \leftarrow \text{ data} \]
Notice that data is grammatically plural in the book: data are, not data is. The book writes it that way throughout, and the plural is a useful reminder that the data are many separate values, one per member of the sample.
Notice also what is not data. The mean of the three scores, 84.3, is not one of the values anyone recorded. It is a statistic computed from the data, which is why the six terms need to be kept apart: data, statistic and variable are three different things in every study.
OpenStax Introductory Statistics 2e, §1.1 Definitions of Statistics, Probability, and Key Terms §1.1, p. 7 — data and datum
Sorting
Sort each variable. Ask one question: do the values have equal units, so that an average of them would mean something? Watch the ones written in digits.
Sort into buckets
Sort each variable into numerical or categorical.
The jersey number and the postal code are the reason this slide exists. They look like numbers and behave like names. Player 23 is not more of anything than player 7, and the gap of 16 between them measures nothing.
The yes or no answer is categorical too, even though you will soon code it as 1 and 0. The codes are a bookkeeping trick, and the next part shows why that particular trick happens to be safe.
Section
Part 5
Step zero
Every example in this part asks you to name six terms. Before you name any of them, there is a step that decides most of the answers, and it is the step people skip.
Here is a study written the way news reports write them: a transit authority wants to know the average commute time of people who ride its buses on weekdays, and it hands a questionnaire to every tenth rider boarding at the central station one Tuesday morning.
Discussion prompt
Before naming anything, underline the words that fix the population and the summary. Then say what the question itself requires that the method may not deliver.
The question decides the population, the parameter and the variable; the method decides the sample, the statistic and the data. Reading the question first gives you half the answers, and comparing it with the method shows you where the study is weak.
That comparison is what the book's careful answers to Examples 1.1 and 1.4 are quietly doing, as you will see in a moment.
Worked example
The book's first full example. We want the average amount of money first-year college students spend at ABC College on school supplies that do not include books. We randomly survey 100 first-year students; three of them spent 150, 200 and 225 dollars.
Work from the question, not from the people who answered. The question names its own population and its own summary, an average, and both of those decide four of the six answers before you look at who was surveyed.
Find the population
Why: ask who the question is about: every first-year at the college this term.
\[ \text{population: all first-years at ABC College this term} \]
Find the sample
Why: ask who was actually measured: the students surveyed.
\[ \text{sample: the 100 first-years surveyed} \]
Find the parameter
Why: the summary the question asks for, computed over the population.
\[ \text{parameter: mean spent by all first-years} \]
Find the statistic
Why: the same summary, computed over the sample.
\[ \text{statistic: mean spent by the 100} \]
Define the variable for one member
Why: the question each student answers.
\[ X = \text{amount spent, no books, by one first-year} \]
Name the data
Why: the answers, in dollars.
\[ \text{data: } 150, \; 200, \; 225, \; \ldots \text{ dollars} \]
One discrepancy in the printed solution: it says the sample could be all students in one section of a beginning statistics course, which may not represent the population. The problem as stated surveyed 100 students at random, so the 100 are the sample here. The book's alternative is a warning about a convenience sample, not the answer to this problem.
Check the pairing
Why: the parameter and the statistic must be the same kind of summary of the same variable, one for each group; both are means of X.
\[ \mu = \text{mean of } X \text{ (all)}, \quad \bar{x} = \text{mean of } X \text{ (the 100)} \]
OpenStax Introductory Statistics 2e, §1.1 Definitions of Statistics, Probability, and Key Terms §1.1, pp. 7-8 — Example 1.1
Picture it
Example 1.1 is worth drawing once, because the same diagram fits every study in this section. The six terms are not a list; they are three pairs.
Figure (svg): A diagram of Example 1.1. A large box labelled population, all first-year students at ABC College this term, contains a smaller box labelled sample, the 100 surveyed. Beside the population box sits the parameter, the mean spent by all first-years; beside the sample box sits the statistic, the mean spent by the 100. Below, a box for the variable X, the amount one first-year spent, with an arrow to the data, 150, 200, 225 dollars and so on.
The first pair is two groups of people: the population and, inside it, the sample. The second pair is two numbers summarising those groups: the parameter for the population, the statistic for the sample.
The third pair is the question and its answers: the variable, asked of each member, and the data, the values it took. The three amounts printed in the problem are three data values, one datum per student.
Once you see the pairs, each answer can be checked against its partner. If your parameter mentions the sample, or your statistic mentions all first-years, something is crossed. If your variable is an average, it has wandered into the second pair, where it does not belong.
For the three printed amounts, the mean is 191.67 dollars. That is not the statistic of the study, which uses all 100 students; it is just the mean of the three you were shown.
\[ \dfrac{150 + 200 + 225}{3} = \dfrac{575}{3} \approx 191.67 \]
Worked example
The same structure, a new school. We want the average amount spent on school uniforms each year by families with children at Knoll Academy. We randomly survey 100 families; three spent 65, 75 and 95 dollars.
Population
Why: who the question is about.
\[ \text{all families with children at Knoll Academy} \]
Sample
Why: who was asked.
\[ \text{the 100 families surveyed} \]
Parameter and statistic
Why: the mean amount spent, over each group in turn.
\[ \mu: \text{all families}, \qquad \bar{x}: \text{the 100} \]
Variable
Why: for one family, so it is a question, not a summary.
\[ X = \text{amount one family spends on uniforms per year} \]
Data
Why: the dollar amounts the families gave.
\[ 65, \; 75, \; 95, \; \ldots \text{ dollars} \]
Notice that the variable is numerical: dollars have equal units, so the mean the question asks for is a sensible summary. If the question had asked what share of families spend more than 100 dollars, the same data would feed a proportion instead.
The unit of the study is the family, not the child. A family with three children is one member of the population and gives one datum, so the variable must be written per family.
Check against the three-pair diagram
Why: two groups, two means, one question and its answers, with nothing crossed between the pairs; the three printed values average 78.33 dollars, a summary of those three only.
\[ \dfrac{65 + 75 + 95}{3} = \dfrac{235}{3} \approx 78.33 \]
OpenStax Introductory Statistics 2e, §1.1 Definitions of Statistics, Probability, and Key Terms §1.1, p. 8 — Try It 1.1
Matching
A study at a local college analyses the average cumulative GPA of students who graduated last year. Match each term to the phrase that describes it. One phrase from the book is left out here on purpose and discussed below.
Match the pairs
Why: The population is every graduate from last year, and the randomly selected group is the sample. The two averages differ only in the words in the study: with them it is the statistic, without them the parameter. The GPA of one student is the variable, and the four GPAs are data.
The phrase left out is all students who attended the college last year. It is the tempting wrong population: it includes students who did not graduate, and the question is only about graduates. The population is fixed by the question, so read the question's own words.
The two averages are the pair to slow down on. They differ by three words, in the study. Those three words move a number from the population to the sample, and that is the whole difference between a parameter and a statistic.
OpenStax Introductory Statistics 2e, §1.1 Definitions of Statistics, Probability, and Key Terms §1.1, p. 8 — Example 1.2
Error analysis
Four students each fill in one line of Example 1.2. Two lines are sound and two contain a classic slip. Find the slips before reading the notes.
Annotate
On: \( \text{pop.} = \text{attended},\; \text{var.} = \text{mean GPA},\; \text{data} = \text{GPAs} \)
Both slips have the same cure: anchor every answer to the question's own words. The question names last year's graduates, so anything broader is the wrong population; the question asks about one GPA per student, so anything that averages is not the variable.
If you want a number to go with the four data values, their mean is 2.96. That is still not the study's statistic, which averages every graduate in the sample, not just the four shown.
\[ \dfrac{3.65 + 2.80 + 1.50 + 3.90}{4} = \dfrac{11.85}{4} \approx 2.96 \]
OpenStax Introductory Statistics 2e, §1.1 Definitions of Statistics, Probability, and Key Terms §1.1, p. 8 — Example 1.2
Matching
A survey of athletes at a university studies their heights, in metres. Match each term to its description. The book's list has seven phrases; the extra one, all students in the university, is the wrong population, since the study is about athletes only.
Match the pairs
Why: The population is all the university's athletes and the randomly selected group is the sample. The average height in the survey is the statistic; the average height of all the university's athletes is the parameter. The height of one athlete is the variable and the four heights are data.
The book's list writes Statistics, plural, where it means the term statistic. It is a small misprint, but the plural is genuinely a different word: statistics is the subject, a statistic is one number from a sample.
The variable here is numerical, height in metres, with equal units. The four printed heights average 1.80 metres, again a summary of those four values only.
\[ \dfrac{1.82 + 1.76 + 1.69 + 1.93}{4} = \dfrac{7.20}{4} = 1.80 \text{ m} \]
OpenStax Introductory Statistics 2e, §1.1 Definitions of Statistics, Probability, and Key Terms §1.1, p. 8 — Try It 1.2
Worked example
The National Transportation Safety Board crashes cars with dummies in the front seats into a wall at 35 miles per hour. We want the proportion of dummies in the driver's seat that would have had head injuries, had they been real drivers. We start with a simple random sample of 75 cars.
The book's Table 1.1 fixes the conditions: a crash speed of 35 miles per hour and dummies in the front seat. Those conditions belong inside the population's definition, because a result for 35 miles per hour says nothing about crashes at 60.
Population
Why: the question is about all such cars, not just the 75.
\[ \text{all cars containing dummies in the front seat} \]
Sample
Why: the cars actually crashed.
\[ \text{the 75 cars, a simple random sample} \]
Parameter
Why: the question asks for a proportion, so the parameter is a proportion.
\[ \text{share of all driver dummies with head injuries} \]
Statistic
Why: the same proportion, over the 75.
\[ \text{share of the 75 driver dummies with head injuries} \]
Variable, for one dummy
Why: the book's printed sentence is garbled; what it means is a yes or no question.
\[ X = \text{whether one driver dummy would have had a head injury} \]
Data
Why: each answer is a word, not an amount.
\[ \text{yes, head injury} \quad \text{or} \quad \text{no, no head injury} \]
Why dummies? Real drivers cannot be crashed on purpose, so the dummies stand in for them, and the question is phrased as what would have happened had they been real. The variable is still a question asked of one member, here one dummy.
This is the first example with a categorical variable. The data are yes and no, so the right summary is a proportion, not a mean; that is why the question asks for a proportion in the first place.
Check that the summary fits the variable
Why: a categorical variable is summarised by the share in a category, and both the parameter and the statistic are such shares.
\[ \text{categorical } X \;\Rightarrow\; \text{proportion} \]
OpenStax Introductory Statistics 2e, §1.1 Definitions of Statistics, Probability, and Key Terms §1.1, p. 9 — Example 1.3 and Table 1.1
Worked example
The book gives no count of injuries. Suppose, for practice, that 33 of the 75 driver dummies would have had head injuries. The number is made up; the method is the point.
Compute the statistic
Why: a proportion is the count in the category over the whole sample.
\[ \hat{p} = \dfrac{33}{75} = 0.44 \]
Compute the share without injury
Why: the rest of the sample.
\[ \dfrac{75 - 33}{75} = \dfrac{42}{75} = 0.56 \]
Notice the hat on the p. It marks the proportion as one computed from the sample, a statistic, in the same way the bar on x marks a sample mean. The population's proportion, the parameter, is plain p and stays unknown.
Code each yes as 1 and each no as 0
Why: now every datum is a number, and you can add them.
\[ \underbrace{1 + \cdots + 1}_{33} + \underbrace{0 + \cdots + 0}_{42} = 33 \]
Take the mean of the codes
Why: the total of the codes over the 75 dummies.
\[ \dfrac{33}{75} = 0.44 \]
The mean of the zeros and ones is exactly the proportion of yeses. That is why the yes or no variable can safely be coded: unlike party codes, the 1 and 0 mean something, namely has the property and does not.
Check the two shares and the interpretation
Why: the two shares must cover the whole sample, so they add to 1.
\[ 0.44 + 0.56 = 1 \]
In context: in this sample, 44 percent of driver dummies would have suffered head injuries. That is the statistic; the parameter, for all such cars, is estimated by it, not equal to it.
Worked example
A survey checks how long phones take to charge from 50 percent to 100 percent, using a 30-watt charger on Android phones. We want the proportion of Android phones that charge to 100 percent within 30 minutes, starting with a simple random sample of 200 phones.
As with the crash test, the book's table of criteria, a 30-watt charger and an Android phone, becomes part of the population. A result for these phones on this charger cannot be stretched to other phones or slower chargers.
Population
Why: every phone the question covers: Android phones charged this way.
\[ \text{all Android phones charged on a 30 W charger} \]
Sample
Why: the phones tested.
\[ \text{the 200 phones, a simple random sample} \]
Parameter and statistic
Why: the question asks for a proportion, so both are proportions.
\[ p: \text{share of all such phones}, \quad \hat{p}: \text{share of the 200} \]
Variable
Why: one yes or no question per phone.
\[ X = \text{whether one phone reaches 100\% within 30 minutes} \]
Data
Why: words, not amounts.
\[ \text{yes or no, one per phone} \]
A tempting alternative is to take the variable as the charging time in minutes, which is numerical. That would be a perfectly good variable for a different question, the average charging time. The question asked here is a proportion, so the variable has to be the yes or no answer.
Check the variable against the question
Why: the question's summary is a proportion, and a proportion is a share of a category, so the variable must be categorical.
\[ \text{asked: a proportion} \;\Rightarrow\; X \text{ is yes or no} \]
OpenStax Introductory Statistics 2e, §1.1 Definitions of Statistics, Probability, and Key Terms §1.1, p. 9 — Try It 1.3 and Table 1.2
Worked example
An insurance company wants the proportion of all medical doctors who have been involved in one or more malpractice lawsuits. It selects 500 doctors at random from a professional directory and counts how many in the sample have been involved in a lawsuit.
Population, as the book gives it
Why: the doctors the company could actually have drawn.
\[ \text{all medical doctors listed in the directory} \]
Figure (svg): Three nested regions. The outer dashed region is all medical doctors. Inside it, a solid region is the doctors listed in the professional directory, which the book names as the population. Inside that, a small orange region holds the 500 doctors selected at random. Some doctors sit in the outer region but outside the directory, marked as unlisted.
Notice the gap. The question says all medical doctors; the book's solution says the doctors listed in the directory. The solution is the careful one. Doctors missing from the directory had no chance of being drawn, so the sample can only speak for the directory's list.
A list that a sample is drawn from is called a sampling frame in later courses. Whatever the frame leaves out, the study leaves out, and no amount of careful random drawing inside the frame can reach the doctors outside it.
Sample
Why: the doctors drawn.
\[ \text{the 500 doctors selected at random} \]
Parameter and statistic
Why: a proportion, over each group.
\[ \text{share sued: directory } (p), \quad \text{the 500 } (\hat{p}) \]
Variable and data
Why: one yes or no question per doctor.
\[ X = \text{whether one doctor has been sued}; \quad \text{yes or no} \]
Check that the conclusion stays inside the list
Why: the sample was drawn from the directory, so any claim built on its statistic is a claim about listed doctors.
\[ \hat{p} \;\longrightarrow\; p_{\text{directory}}, \text{ not } p_{\text{all doctors}} \]
If the directory leaves out, say, retired doctors, they may differ in lawsuits. The insurer can only call its estimate a figure for all doctors if the directory is nearly complete.
OpenStax Introductory Statistics 2e, §1.1 Definitions of Statistics, Probability, and Key Terms §1.1, pp. 9-10 — Example 1.4
Comparison
A news agency wants the proportion of all truck drivers who have no points on their licence. It selects 1,000 truck drivers at random from the directory of truck drivers. Fill in the blanks, using Example 1.4 as the model.
Comparison matrix
| term | Try It 1.4 |
|---|---|
| population | all truck drivers listed in the directory |
| sample size | 1000 |
| parameter | share of all listed drivers with no points |
| statistic | share of the sampled drivers with no points |
| type of variable | categorical |
| summary used | proportion |
The variable is whether one driver has no points on their licence, so the data are yes and no. That makes the variable categorical and the summary a proportion, exactly as for the doctors.
The population follows Example 1.4's reasoning: drivers in the directory, because those are the only drivers the agency could have drawn. If you wrote all truck drivers, you have matched the question's words; the directory version is the more careful answer, and you should be able to say why.
If 830 of the 1,000 had no points, the statistic would be 830 over 1,000, which is 0.83, and the agency would estimate that about 83 percent of listed drivers have clean licences. That count is invented for illustration.
OpenStax Introductory Statistics 2e, §1.1 Definitions of Statistics, Probability, and Key Terms §1.1, p. 10 — Try It 1.4
Worked example
The book ends the section with a group exercise. You want the average number of glasses of milk college students drink per day. Yesterday, in your English class, you asked five students how many glasses they drank the day before: 1, 0, 1, 3 and 4.
Population and sample
Why: the question is about college students; you asked five.
\[ \text{all college students} \quad \supset \quad \text{the 5 asked} \]
Variable and data
Why: one student's count of glasses; the five answers.
\[ X = \text{glasses one student drank}; \quad 1, 0, 1, 3, 4 \]
Parameter
Why: the mean over the population, unknown.
\[ \mu = \text{mean glasses per day, all college students} \]
The parameter is written with mu and described in words because no one knows its value. The whole purpose of the exercise is to estimate it, and the next line is the estimate.
Statistic
Why: add the five answers and divide by five.
\[ \bar{x} = \dfrac{1 + 0 + 1 + 3 + 4}{5} = \dfrac{9}{5} = 1.8 \]
Figure (svg): A dot plot of glasses of milk drunk the day before by five students: one at 0, two stacked at 1, one at 3 and one at 4. A yellow marker shows the mean, 1.8 glasses.
The variable is numerical, since glasses have equal units, so a mean is the right summary. But look at the sample: five people who happened to sit near you in one English class. That is a tiny sample chosen for convenience, and Section 1.2 will name both problems.
Check the statistic and its reach
Why: five times 1.8 gives back the total of 9 glasses; the claim must stay with the five.
\[ 5 \times 1.8 = 9 \]
Interpretation: the five students asked drank 1.8 glasses of milk on average yesterday. Whether college students in general drink that much, this sample cannot tell you.
OpenStax Introductory Statistics 2e, §1.1 Definitions of Statistics, Probability, and Key Terms §1.1, p. 10 — Collaborative Exercise
Section
Part 6
Concept
The book names two words that come up constantly: mean and proportion. Take three exam scores, 86, 75 and 92. To find your mean score, add the three and divide by three.
\[ \dfrac{86 + 75 + 92}{3} = \dfrac{253}{3} = 84.33\ldots \]
Figure (svg): A number line of exam scores from 70 to 95 with dots at 75, 86 and 92. A yellow triangle at the mean, 84.33, acts as a balance point. Arrows show each score's distance from the mean: 75 is 9.33 below, 86 is 1.67 above, 92 is 7.67 above.
One way to see the mean: if the three exams had all scored the same, and the total stayed 253, each would have scored 84.33. Another way is in the picture. The mean is the balance point of the data, where the distances below and the distances above cancel.
\[ -9.33 + 1.67 + 7.67 = 0 \]
A note from the book about words. Mean and average are used interchangeably in practice, and this course does too. Technically, the arithmetic mean is this add-and-divide number, and average is a looser word for a centre; outside a statistics class, people saying average almost always mean the arithmetic mean.
OpenStax Introductory Statistics 2e, §1.1 Definitions of Statistics, Probability, and Key Terms §1.1, p. 7 — the mean and the Note on mean and average
Worked example
The book reports this mean as 84.3 to one decimal place. Do it in full, so that every move is visible, and then check it properly.
Add the scores
Why: a mean needs a total first.
\[ 86 + 75 + 92 = 253 \]
Count the scores
Why: the number you will divide by.
\[ n = 3 \]
Divide the total by the count
Why: share the 253 points equally across three exams.
\[ \bar{x} = \dfrac{253}{3} = 84.333\ldots \]
Round to one decimal place
Why: the second decimal is 3, below 5, so round down.
\[ 84.333\ldots \approx 84.3 \text{ points} \]
Why one decimal place? The scores are whole points, and a mean reported to one more decimal than the data is the usual convention: precise enough to separate 84.3 from 84.7, without pretending to a precision that three exams cannot support.
Keep the unrounded value for any further calculation and round only at the end. Rounding early is a common source of answers that are off in the last digit.
Check by multiplying back
Why: three exams at the mean should return the total; with the rounded mean you land within rounding of it.
\[ 3 \times 84.3 = 252.9 \approx 253 \]
So your mean exam score is 84.3 points. The tiny gap of 0.1 point is the rounding, not an error, and multiplying back is how you tell the difference.
OpenStax Introductory Statistics 2e, §1.1 Definitions of Statistics, Probability, and Key Terms §1.1, p. 7 — the three exam scores
Estimation
A quick estimate catches most arithmetic slips. Use the balance idea on the three Knoll Academy uniform amounts from Try It 1.1: 65, 75 and 95 dollars.
Predict first
Without dividing, where must the mean of 65, 75 and 95 lie?
Correct: A little below 80, pulled up from 75 by the 95
Why: The 95 is 20 above the middle value of 75 while the 65 is only 10 below it, so the balance point must sit above 75. The exact mean is 235 divided by 3, which is 78.33 dollars, a little below 80.
\[ \dfrac{65 + 75 + 95}{3} = \dfrac{235}{3} \approx 78.33 \]
Two quick rules come from the balance picture. A mean always lies between the smallest and the largest value, so any answer outside 65 to 95 is wrong. And a value far from the rest pulls the mean towards it, so here the mean must be above 75.
The option of exactly 80 is a tempting shortcut: the midpoint of the smallest and largest values. It ignores the middle value entirely, and it would only be the mean if the middle value sat exactly halfway.
Concept
The second summary is for categorical data. In a math class of 40 students, 22 are men and 18 are women. The proportion of men is the count of men over the whole class.
Figure (svg): A grid of 40 squares in four rows of ten. The first 22 squares are orange, labelled men; the remaining 18 are blue, labelled women.
\[ \dfrac{22}{40} = 0.55, \qquad \dfrac{18}{40} = 0.45 \]
A proportion is always between 0 and 1: none of the group, or all of it. Multiply by 100 and it becomes a percentage, so 0.55 and 55 percent say the same thing.
\[ 0.55 + 0.45 = 1 \]
That line is a free check. Every student is in exactly one of the two categories, so the two shares must use up the whole class. If your proportions for categories that cover everyone do not add to 1, a count or a denominator is wrong.
Notice what the denominator is: the whole group, 40, not the other category. Dividing 22 by 18 gives 1.22, a ratio of men to women, which is a different number answering a different question.
OpenStax Introductory Statistics 2e, §1.1 Definitions of Statistics, Probability, and Key Terms §1.1, p. 7 — the class of 40
Fill the middle
Do the women's proportion yourself, with the whole class as the denominator. Then check it with the rule that the two shares cover everyone.
Fill in the blanks
\dfrac400.45} = ___
Why: The denominator is the whole class of 40, not the 22 men, so the proportion of women is 18 divided by 40, which is 0.45. It adds to the men's 0.55 to make 1, which confirms it.
If you wrote 22 in the denominator, you computed 18 over 22, about 0.82, the number of women per man. It is a ratio, not a share of the class, and the failed check would catch it: 0.55 plus 0.82 is far more than 1.
The same structure covers every proportion in this lesson: the dummies with head injuries over all 75 cars, the sued doctors over all 500 sampled, the clean licences over all 1,000 drivers. The denominator is always the group the proportion describes.
Picture it
Mean and proportion look like two separate tools. They are one tool, and the picture shows why. Code each man in the class as 1 and each woman as 0, then find the mean of the forty codes.
Figure (svg): A dot plot with only two positions, 0 and 1, on a number line. Eighteen dots stack above 0 (women, coded 0) and twenty-two above 1 (men, coded 1), drawn as two columns of squares. A yellow balance point sits at 0.55.
\[ \bar{x} = \dfrac{22(1) + 18(0)}{40} = \dfrac{22}{40} = 0.55 \]
The mean of the codes is exactly the proportion of men. In the picture, the balance point of 22 squares at 1 and 18 at 0 sits at 0.55, a little past halfway, because there are slightly more squares at 1.
This is why the yes or no code was safe for the crash-test dummies when the party code was not. With exactly two categories coded 1 and 0, the mean has a meaning, the share coded 1, and it does not change if you swap which category gets the 1; it simply becomes the other share.
Later chapters lean on this fact heavily: much of what you learn about sample means carries over to sample proportions, because a proportion is a mean of zeros and ones.
Worked example
Means are easy to misuse when you combine groups. One section of 10 students averaged 80 on a test and another section of 30 averaged 90. What is the mean for all 40 students? These sections are made up.
The trouble is that a mean has thrown away the group size. To combine means, you have to put the size back, and the way to do that is to rebuild each group's total, because totals can be added and means cannot.
Rebuild the first section's total
Why: a mean times its count gives back the total.
\[ 10 \times 80 = 800 \]
Rebuild the second section's total
Why: same move.
\[ 30 \times 90 = 2700 \]
Add the totals and the counts
Why: a combined mean needs the combined total over the combined count.
\[ 800 + 2700 = 3500, \qquad 10 + 30 = 40 \]
Divide
Why: share the combined total among all 40 students.
\[ \dfrac{3500}{40} = 87.5 \]
Figure (svg): A number line from 75 to 95. Above 80 stands a short stack labelled section of 10, mean 80; above 90 stands a stack three times as tall labelled section of 30, mean 90. A yellow balance triangle sits at 87.5, the combined mean. A red dashed marker at 85 is labelled average of the averages, wrong.
Before trusting the answer, test it against the balance idea: a combined mean must lie between the two section means, 80 and 90. It does, and it sits nearer to 90, where most of the students are.
The answer is not 85, the midpoint of 80 and 90. Three quarters of the students are in the section that averaged 90, so the combined balance point sits three quarters of the way from 80 to 90.
Check with the weights directly
Why: weight each section's mean by its share of the students; the result must agree with the total method.
\[ \tfrac{10}{40}(80) + \tfrac{30}{40}(90) = 20 + 67.5 = 87.5 \]
So the 40 students averaged 87.5 points. Averaging the two averages would give 85 and understate the class by 2.5 points.
Tweak it
The slider generalises the last example. The first section is fixed at 10 students averaging 80. The horizontal axis is the size of the second section; the slider sets the second section's mean.
Parameter explorer
With the second section's mean at 90, read the combined mean when the second section has 10 students, then 30, then 100. What value does the curve approach?
\[ \text{combined mean} = \dfrac{10(80) + n({m})}{10 + n} \]
Only one point on the curve gives the simple average of the two means: where both sections have 10 students. Everywhere else the bigger section pulls the combined mean towards its own mean, and as it grows the curve flattens towards that mean.
Set the second mean below 80 and the curve falls instead, towards the new value. The rule is symmetric: the combined mean always lies between the two section means, nearer to the bigger section.
This is the same balance idea as before, with whole sections as the weights. You will meet it again as the weighted mean in Chapter 2.
Trap
Section one, 10 students, averaged 80. Section two, 30 students, averaged 90. A student reports:
\[ \dfrac{80 + 90}{2} = 85 \]
The reasoning: the mean of two numbers is their sum over two, and here are two numbers.
The two numbers stand for different numbers of people. Go back to totals.
\[ \dfrac{800 + 2700}{40} = 87.5 \]
Averaging the averages works only when the groups are the same size.
The same slip with proportions: averaging the share of men in a class of 10 and a class of 40 gives the wrong share for all 50.
Ranking
Whatever the data, a reliable summary follows the same sequence of moves. Put them in order; the order is what protects you from the traps on the last few slides.
Put in order
Put the steps for producing and reporting a summary in order.
Why: The type of variable decides the summary, so it comes first. Then you need the total or the count and the group size before you can divide. Checking comes before rounding so the check uses exact numbers, and the report comes last, naming the units and exactly which group the number describes.
The first move is the one people skip. Jump straight to dividing, and you can average jersey numbers or party codes without noticing. Asking about the variable first is what keeps the arithmetic meaningful.
The last move is where descriptive and inferential get mixed. A summary of the sample is reported as a statistic about the sample; a claim about the population is a separate step, and it should be labelled as an estimate.
Section
Part 7
Pattern
Everything in this section reduces to six short questions. Put them, in this order, to any statistic you meet, in a textbook, a news report or your own work.
The first two questions are where most misleading numbers go wrong, long before any arithmetic. The bookstore line and the directory of doctors were both failures there, and no calculation afterwards could have repaired them.
The last question is the bridge to the rest of the course. Descriptive statistics answer it with certainty; inferential statistics answer it with probability, and say how sure they are.
Check
One study, one number. Decide which group it describes before reading the choices.
Check your understanding
A city surveys 400 randomly chosen households and finds that 31 percent have a home solar panel. Which statement is correct?
Answer: A
Why: The 0.31 was computed from the 400 households surveyed, so it is a statistic. The city's true share is the parameter, and the survey's figure is an estimate of it. Random selection makes the estimate trustworthy; it does not turn the estimate into the parameter.
The tempting distractor is the one that leans on randomness. Random sampling is what makes a statistic worth trusting; it never changes which group the number describes.
If you chose the option calling 0.31 the variable, return to the three-pair diagram. Variables are questions asked of one member; summaries of many members live in the parameter and statistic pair.
Check
The type of variable decides the summary. Classify first, then choose.
Check your understanding
A survey records each respondent's favourite streaming service. Which summary is appropriate for this variable?
Answer: A
Why: Favourite service is a categorical variable, so the right summary is the proportion of respondents in each category. Proportions do not depend on any coding, so they describe the respondents rather than a labelling.
The coded-mean distractor is common in real spreadsheets, where every column holds numbers whether or not they are amounts. The check is the one from Part 4: would recoding the categories change the answer? If so, the summary is not describing the data.
The last option overcorrects. Categorical data cannot be averaged, but they can be counted, and counting leads directly to proportions.
Check
A combining question. Rebuild the totals before choosing anything.
Check your understanding
A store's 20 weekday shifts averaged 120 customers each and its 10 weekend shifts averaged 210. What is the mean number of customers per shift over all 30 shifts?
Answer: A
Why: Rebuild the totals: 20 shifts at 120 is 2,400 customers and 10 shifts at 210 is 2,100. Together that is 4,500 customers over 30 shifts, which is 150 per shift. The weekday shifts outnumber the weekend ones two to one, so the combined mean sits nearer to 120.
\[ \dfrac{20(120) + 10(210)}{30} = \dfrac{2400 + 2100}{30} = \dfrac{4500}{30} = 150 \]
The wrong options each break one rule from the slides. One ignores the group sizes, one weights the wrong group, one forgets that a mean must lie between the values it summarises. Check any combined mean against all three.
Explain it to yourself
The idea under this whole section is the gap between the number you know and the number you want. Putting it in your own words is what makes it usable on a study you have never seen.
Discussion prompt
In three or four sentences, explain to someone who has never studied statistics why a careful poll that finds 52 percent support cannot simply announce that 52 percent of the country supports the policy. Use the words sample, statistic and parameter.
If your explanation used the Pine College hat draws, it probably went well: four fair samples gave four different statistics while the parameter sat still at 150.
If it said the poll was wrong, look again. The statistic is exactly right about the sample. What it cannot be is exactly right about the population, and the course's later chapters are about measuring that gap.
Exit ticket
One study, every idea in the lesson. Answer in full sentences, naming the evidence for each answer.
A library wants the mean number of books borrowed last year by its 6,000 cardholders. It draws 150 cardholders at random from its membership list. Three of them borrowed 4, 0 and 11 books. Of the 150, 45 borrowed no books at all.
Discussion prompt
Name the population, sample, parameter, statistic, variable and data. Say whether the variable is numerical or categorical. Then compute the proportion of the sample who borrowed nothing, and say whether it is a statistic or a parameter.
The most common slip is the variable: writing the mean number of books instead of the number one cardholder borrowed. The mean is the parameter or the statistic; the variable is the question asked of each person.
The second most common is calling 0.3 the parameter because it is a clean number. Clean or not, it was computed from the 150, so it is a statistic.
Recap
This section built the vocabulary the rest of the course is written in. Each row of the table is one pair of words, with the mistake it protects you from.
| the pair | what it separates | the trap |
|---|---|---|
| descriptive / inferential | summarising what you measured / claiming beyond it | a sample's average quoted as everyone's |
| population / sample | who the question is about / who was measured | the bookstore line; the unlisted doctors |
| parameter / statistic | the fixed number you want / the varying number you have | reporting 0.55 for one class as true of all |
| variable / data | the question for one member / its answers | defining the variable as an average |
| numerical / categorical | equal units / categories | averaging jersey numbers or party codes |
| mean / proportion | total over count / part over whole | averaging two averages of unequal groups |
\[ \text{statistic} \;\xrightarrow{\text{estimates}}\; \text{parameter} \]
That arrow is the course in one line. Probability, sampling distributions, confidence intervals and hypothesis tests are all ways of saying how far a statistic is likely to be from the parameter it estimates.
Section 1.2 picks up straight from here: the kinds of data in more detail, how to choose a sample so that it is representative, and why bigger samples fix some problems but not others.
OpenStax Introductory Business Statistics 2e, §1.1 Definitions of Statistics, Probability, and Key Terms §1.1 — the same key terms in business settings
Want this taught 1-on-1? Alexander tutors Statistics — $55/session, free consultation.