1. The puzzle

The fundamental problem statistics exists to solve is that you cannot see everything. You want to know the average height of all 13-year-olds in a country, but you cannot measure all of them. You want to know whether a vaccine is safe and effective, but you cannot administer it to every person on Earth before approving it. You want to know whether smoking causes cancer, yet you cannot run a controlled experiment on human beings for ethical reasons. In every case; you must reason from a small piece of the picture to a claim about the whole. This is the **problem of induction** that David Hume identified in the 1700s, no number of observed white swans proves that all swans are white. Statistics does not solve the problem of induction. Nothing can. But it gives a rigorous framework for quantifying how confident you should be in your generalizations, how much data you need before a conclusion is reliable, and how likely it is that what you observed happened by chance.

2. The core idea

Statistics has two main branches, and the distinction between them is the first conceptual leap of the course. **Descriptive statistics** organizes and summarizes data you already have: if you measure the heights of every student in your class and report the average and the range, you are doing descriptive work, making no claim beyond the data in front of you. **Inferential statistics** uses a sample to draw conclusions about a larger population: if you measure 30 students and use that to estimate the average height of every student in the school, you need probability theory to do it honestly. The single most important distinction in the entire subject is between **correlation** and **causation**. Two variables are correlated if they tend to move together; they are causally related if one actually produces the other. Ice cream sales and drowning rates are positively correlated because both rise in summer, but ice cream does not cause drowning. Mistaking correlation for causation is the most common statistical error in media, in research, and in everyday reasoning.

3. Who discovered this

The history of statistics is a history of practical problems demanding new mathematical tools. The word itself comes from the Latin *statisticum*, meaning "of the state" — the first statisticians were government officials counting populations and tracking births and deaths. **John Graunt**, a London haberdasher with no formal mathematical training, published *Natural and Political Observations Made upon the Bills of Mortality* in 1662, the first systematic analysis of demographic data. He invented the life table, a tool insurance companies still use today. **Francis Galton** invented the concept of regression toward the mean in the 1880s while studying the heights of fathers and sons. **Karl Pearson** formalized the correlation coefficient. **Ronald Fisher** invented the randomized controlled experiment and the analysis of variance in the 1920s, the tools that made modern clinical research possible. **Florence Nightingale** used statistical graphics to convince the British government that more soldiers died of preventable disease than from battle wounds, revolutionizing military medicine. Statistics, more than almost any other field, grew from urgent human need.

4. How it works

A statistical investigation follows a recognizable structure: question, data collection, description, inference, and conclusion. The question must be precise. "Are students happier?" is too vague. "Do students at this school report higher average wellbeing scores after the new advisory program?" is testable. Data collection requires choosing a sample. **Random sampling**, where every member of the population has an equal chance of being selected, is the gold standard, because it eliminates systematic bias. Description means summarizing the data you collected: center (mean or median), spread (standard deviation or interquartile range), and shape (symmetric, skewed, or bimodal). Inference means using the sample to draw conclusions about the population, with an honest statement of uncertainty. A **confidence interval** says "based on this sample, the true population value is probably in this range." A **hypothesis test** says "if there were no real effect, how likely is it that we'd see data this extreme by chance?"

5. The notation

Here is how the abstract concepts of statistics translate into concrete numbers, using a sample of ten exam scores. These are the raw data points, before any analysis. The first thing we usually want to know is how many observations we have—this is our sample size, symbolized by a lower-case n. In this case, n equals ten. Next, we might want to find the center of this data. The most common measure is the arithmetic mean, often called the average, which we represent with x-bar—a lower-case x with a bar over it. To calculate the mean, we sum all the individual scores and then divide that sum by n, our sample size. For these scores, the sum is eight hundred and eight, divided by ten, giving us a mean of eighty point eight. Keep in mind, the mean is sensitive to outliers—extreme values can pull it significantly. Another measure of center is the median, which is the exact middle value. To find the median, we first arrange the data in ascending order. When we have an even number of observations, like our ten scores, the median is the average of the two middle numbers. Here, that is seventy-nine and eighty-three, averaged to eighty-one. The median is resistant to outliers; it is not pulled by extremely high or low values. Finally, we can describe the spread of the data. The simplest measure of spread is the range—the difference between the highest and lowest values. For our scores, that is ninety-five minus sixty-eight, which gives us a range of twenty-seven. Like the mean, the range is also highly affected by outliers. These fundamental measures give us our first glimpse into the story our data is telling.

6. Key properties

Three properties define any dataset's distribution. **Shape**: is it symmetric (roughly equal on both sides of the center), right-skewed (a long tail to the right, common in income and house prices), or left-skewed (a long tail to the left)? Shape tells you which summaries are appropriate and what process might have generated the data. **Center**: where is the middle? Mean and median both measure center but disagree when data is skewed. For symmetric data they agree closely; for skewed data the mean is pulled toward the tail while the median stays put. This is why **median household income** is the standard measure, a small number of extremely wealthy households would inflate the mean dramatically. **Spread**: how variable is the data? The range is sensitive to a single outlier; the standard deviation measures the average distance from the mean; the interquartile range covers the middle 50%. A summary without spread is incomplete: two classes can have the same average but very different distributions.

7. Worked example

Let us take a simple real-world example and apply some descriptive statistics. Here is a week of daily step counts from a high school student. The data shows seven observations, one for each day: four thousand two hundred steps, eight thousand seven hundred, eleven thousand two hundred, three thousand one hundred, nine thousand four hundred, seven thousand eight hundred, and six thousand five hundred. To find the mean, often represented as x bar, we add all these values together and divide by the number of observations, which is seven. The sum is fifty thousand nine hundred, and dividing that by seven gives us approximately seven thousand two hundred seventy-one steps. That is the average number of steps the student walked per day. Now, for the median, we first need to sort the data from smallest to largest. This gives us three thousand one hundred, four thousand two hundred, six thousand five hundred, seven thousand eight hundred, eight thousand seven hundred, nine thousand four hundred, and eleven thousand two hundred. Since there are seven observations, the median is the middle value—the fourth value in the sorted list—which is seven thousand eight hundred steps. Notice how this is slightly higher than our mean, mainly because the three thousand one hundred steps on one day pulled the mean down. Finally, we can calculate the range, which is the difference between the maximum and minimum values. The highest count was eleven thousand two hundred, and the lowest was three thousand one hundred. Subtracting these gives us a range of eight thousand one hundred steps. This tells us there was quite a lot of variability in this student's activity throughout the week.

8. Limitations

Statistical conclusions about samples apply only to the population the sample represents. Sampling only students in one city and drawing conclusions about students nationally invites error simply because the sample is not representative. Correlation, however strong, cannot establish causation without a controlled experiment, and many questions in medicine and social science cannot be studied experimentally for ethical reasons. **Statistical significance** is not the same as practical importance. With a large enough sample you can detect a difference of 0.001 points on a test that is statistically significant but utterly meaningless in practice. And every statistical model is a simplification, a linear regression line assumes a linear relationship, and if the truth is curved, the model will give systematically wrong predictions. These limitations do not make statistics useless; they make it honest.

9. Related ideas

Statistics is deeply connected to three neighboring fields. **Probability** is the mathematical theory of random events, the foundation on which statistics is built, but probability and statistics face opposite directions. Probability asks "given this model, what data should I expect?"; statistics asks "given this data, what model best explains it?" **Data science** is the modern applied descendant of statistics combined with computer programming. A data scientist uses the same core concepts, sampling, estimation, hypothesis testing, regression — with datasets that may have millions of rows and hundreds of variables. **Machine learning** is the part of data science that builds predictive models from data; every machine learning model is a statistical model.

10. Common mistakes

Four mistakes appear so often that identifying them is a core skill of this course. **Confusing correlation with causation**, always ask whether a confounding variable could explain both observations. **Survivorship bias**, drawing conclusions from the data you can see while ignoring the data that is invisible. During World War II, **Abraham Wald** pointed out that bullet holes on returning planes showed where planes could be hit and survive; armor was needed where there were no holes, because planes hit there did not return. **Using the mean when the median is more informative** for skewed data like income or house prices. **Ignoring sample size**, a result from 12 people is fundamentally different from the same result from 12,000.

11. Tools

The most accessible statistical tool is a spreadsheet. Google Sheets and Microsoft Excel can compute means, medians, standard deviations, and correlations with built-in functions like `AVERAGE()`, `MEDIAN()`, `STDEV()`, and `CORREL()`, and produce histograms and scatterplots in a click. Desmos has a statistics mode that fits regression lines and computes correlation coefficients. For more serious work, **R** is the professional standard for statistical computing, and **Python** with the pandas, scipy, and matplotlib libraries is equally powerful and more widely used in industry.

12. Try this

Find a news headline that makes a statistical claim, something like "people who drink coffee live longer" or "students who play video games get better grades." Ask four questions about it. How was the data collected — randomized experiment or observational study? Who was in the sample, and does that sample represent the population the headline claims? Is the headline claiming correlation or causation? How large was the effect? If coffee drinkers live 0.2 years longer on average, does "live longer" mean what you thought it did? Compare what the actual study says to what the headline claims. In almost every case; you will find that the headline overstates the finding or confuses correlation with causation. That gap is exactly what this course trains you to see.

13. Recap

- Statistics is the science of making decisions under uncertainty by reasoning from data. - It has two branches: descriptive statistics summarizes data you have; inferential statistics uses a sample to draw conclusions about a larger population. - Every dataset can be described by shape, center (mean or median), and spread (standard deviation or IQR). - The mean is sensitive to outliers; the median is resistant. Choose based on the shape of your data. - The most important distinction in the field: correlation is not causation. - The four questions to ask any statistical claim: how was the data collected, does the sample represent the population, correlation or causation, how big is the effect? Statistics has roots in government record-keeping, insurance mathematics, and agricultural experiments, but today it underpins every quantitative field from medicine to machine learning. The chapters that follow develop the tools, sampling, summarizing, modeling, testing, that make these four questions answerable.

Mastery quiz

1. According to the class, what is the single most important distinction in all of statistics?
   • Mean versus median
   • Correlation versus causation
   • Range versus standard deviation
   • Sample versus census
2. Which best describes the difference between descriptive and inferential statistics as defined in the class?
   • Descriptive summarizes data you already have; inferential uses a sample to draw conclusions about a larger population
   • Descriptive uses probability; inferential never does
   • Descriptive applies only to experiments; inferential only to surveys
   • They are two names for the same process
3. The class says the median is 'resistant' while the mean is 'sensitive.' Sensitive to what?
   • Sample size
   • Outliers (extreme values)
   • The order of the data
   • Rounding errors
4. In the survivorship-bias example, Abraham Wald argued the returning WWII planes should be armored where they had NO bullet holes. Why?
   • Those spots were cheapest to reinforce
   • Planes hit in those places did not survive to return, so that damage is invisible in the data
   • Bullet holes weaken the metal permanently
   • The visible holes were already self-sealing