Introduction to Statistics

MIT 18.650 Statistics for Applications, lecture 1, explained in plain English

Want this lecture to stick? In LessonBound you get checked practice questions for every part, can ask about anything you don't get, and can listen with your screen off. Free, no ads.

Learn this lecture free

Watch or read the original lecture: MIT OpenCourseWare.

Part 1 · 0:00–10:00

This part explains what the MIT 18.650 statistics course covers, how it is graded, and why it focuses on math and theory.

Key ideas

The course is mainly about math. It aims to give theoretical guarantees and to show how sure you can be of a result, not just to work with real data. 1:28
Error bars show how certain a number is. The course teaches you to build them, because a single number alone does not tell you how much to trust it. 1:52
A key skill is turning a real-life question into a math problem. You first decide what kind of answer you need, such as a yes or no, a number, or error bars. 2:36
No statistical method is perfect. You should understand its limits, since conclusions drawn from data can be wrong. 3:44
Grades: homework is 30% (the best 10 of 11 sets count), the best of two midterms is 30%, and the final exam is 40%. 9:34

Words to know

error bars: A range placed around a number to show how sure we are of it. A wider range means less certainty.
theoretical guarantees: Promises about how a method behaves, proven with math rather than only tested with examples.
estimator: A method for guessing an unknown number from data. The lesson says the teacher must choose between two of them.
cheat sheet: A single page of notes you may bring into a midterm. In this course it is two-sided and letter-sized, and you can write whatever you want on it.

Example

A weather app says there is a 70% chance of rain. Another app says 70%, plus or minus 10%. The second one shows how sure it is, just as error bars show how sure a statistical result is.

3 checked practice questions for this part are in the app.

Part 2 · 10:00–20:00

This part explains what you need to know before the course, why you should question news about studies, and why searching through lots of data can produce false results.

Key ideas

You need some basic probability, calculus, and linear algebra. The instructor says this is not a strict requirement, but the course will use these ideas. 10:23
There is no required textbook. The instructor recommends All of Statistics by Wasserman, but says the course covers only part of it. 11:29
A news headline like 'study finds' does not automatically prove something. Studies use samples and models, so their results often describe a general pattern, not what happens to you personally. 15:30
If you look at many studies, some will be wrong by chance. The instructor says that in a list of 20 studies, there would probably be at least one wrong one. 15:52
Searching through lots of data until something looks interesting, which is called p-hacking, can make results look real when they are not. The statistics in this course is meant for clean use of data, where the plan is set before looking at the results. 18:49

Words to know

voxel: A tiny cube-shaped piece of a brain scan image. A scan records activity in thousands of these pieces.
parameter: A number that describes a model of the world, such as the typical size of an effect. It describes a pattern, not one person.
p-hacking: Trying many different analyses or hypotheses on the same data until one looks like a success, then reporting only that one.
eigenvectors: Special directions in a grid of numbers (a matrix) that keep their direction when the matrix is applied. You will need this idea later for principal component analysis.

Example

Imagine you test 100 random ideas, like whether a coin is lucky, and check each one. Even if none of them is real, a few will look special just by chance. That is why one exciting result from a big search is not proof.

3 checked practice questions for this part are in the app.

Part 3 · 20:00–30:00

This part explains why statistics is needed: real life is full of randomness, and statistics helps us measure it and make decisions from data.

Key ideas

Dike builders in the Netherlands used past flood data to pick a safe height, instead of guessing. Statistics means using data to make decisions. 20:19
Statistical modeling lets us guess what we have not seen yet, such as a flood bigger than any recorded one. The lesson calls this extrapolating beyond the data. 21:00
Insurance prices are based on expectation, which is the average cost you are likely to cause the company. Randomness, such as a chance of breaking your phone, is what makes this math necessary. 21:00
Statistics turns big, vague questions like 'Is a drug effective?' into simple number questions like 'Is p larger than 0.5?' 22:48
Probability is the study of randomness. Statistics is about finding the basic numbers that probability needs, because in real data those numbers are not given to you. 28:23

Words to know

expectation: The average result you would get if you repeated something many times, with each outcome weighted by how likely it is.
extrapolate: To make a guess about things outside the range of what you have already observed.
control group: A group of people who do not get the treatment, used to compare with the group that does.
significance: Whether a result is strong enough that it probably did not happen just by chance.

Example

Think of a dike built from only ten past floods. If the highest flood so far was 3 meters, a builder using statistics asks how likely a 4-meter flood is, rather than simply saying 'let's build it 500 meters high to be safe.' The data helps make a sensible guess about the future.

3 checked practice questions for this part are in the app.

Part 4 · 30:00–40:00

This part explains how probability starts from known rules, how statistics works backward from data to estimate those rules, and why every model is a simplification with some error.

Key ideas

Probability starts with a simple process whose basic outcomes are known. A fair die has a 1/6 chance for each face, and you can build up to harder questions, like the chance that two dice sum to 7. 31:26
In the dice game, 7 is the best pick because it is the most likely sum of two dice. You can compute the chance of every sum from 2 to 12. 30:42
A statistician works the other way. Without knowing the true rules, they roll the die many times, count the results, and estimate the probabilities from those counts. 32:08
Models are simplifications, so they will have errors. If the die is not really fair, assuming it is creates a model error, but that is the price for getting anything useful from data. 32:50
A good model is simple enough to be useful but complete enough to include the things that matter. Anything left out becomes part of the random noise. 37:11

Words to know

probability: How likely something is to happen, written as a number from 0 to 1.
expectation: The average result you would get if you repeated a game many times.
parameter: A number that controls how a model behaves, such as the chance of heads on a coin.
model: A simplified description of how something works, used to make predictions or estimates.

Example

Imagine you find a coin and don't know if it is fair. You flip it 100 times and count the heads, then use that count to estimate the chance of heads. This is statistics: starting from data to guess the rule, the reverse of probability, which starts from a known rule to predict the data.

3 checked practice questions for this part are in the app.

Part 5 · 40:00–50:00

This part explains how statistics works backward from data to the hidden truth behind it, and why that makes it different from probability.

Key ideas

Probability starts with the truth. If you know how a random process works, you can predict what data it will produce. 40:00
Statistics goes the other way. You start with data you already have and try to figure out what truth could have produced it. 40:17
Statistics is about averaging and summarizing. A short list of numbers, like a mean and a standard deviation, can describe a large set of data better than the full list. 40:58
Because you only have one data set, statistical answers come with error bars. These are a range around your estimate that will hopefully contain the true value. 43:08
The course focuses on the math behind methods and assumes the modeling assumptions are correct. One key assumption is IID, which means the data points are alike and one does not affect another. 43:51

Words to know

standard deviation: A number that shows how spread out a set of values is around the average. A small number means the values are close together.
IID: Short for independent and identically distributed. It means every data point comes from the same kind of process, and one point does not change the others.
error bars: A range placed around a number to show how unsure you are about it. The true value is likely to fall somewhere inside that range.
estimator: A rule or formula that turns your data into a best guess about a hidden number, such as an average.

Example

Imagine a teacher who has the test scores of every student in a class. Instead of remembering all of them, she tells a friend the average score and how spread out the scores were. Those two numbers give a good picture of the whole class, and the friend can use them to guess how a new student might do.

3 checked practice questions for this part are in the app.

Part 6 · 50:00–1:00:00

This part shows how to turn a simple yes-or-no observation into a number, and why the number only means something if the data is collected fairly and in large enough amounts.

Key ideas

The fancy-sounding study is really just counting which way couples lean their heads when they kiss. Statistics starts by turning a question like this into numbers. 50:00
Where your data comes from matters. Photos from Google Images are not a fair sample, because the people who post them are not typical, and many are actors posing on purpose. 51:28
Before you trust a result, ask which group it describes. Couples seen at airports may not represent all couples in the world. 53:34
To measure a result, code each outcome as a number, such as 1 for right and 0 for left. Then estimate the unknown proportion p by dividing the count by the total: 80 out of 124 is 64.5%. 55:24
A result from a small group can change a lot from one group to the next. To decide if 64.5% is really bigger than 50%, we need to model the random process behind the data and ask how large the sample should be. 56:07

Words to know

parameter: A fixed but unknown number that describes a whole group, like the true share of all couples who lean right. We estimate it from data.
proportion: A share of the total, written as a fraction or percent. For example, 80 out of 124 is a proportion of about 64.5%.
representative: Being a fair picture of the whole group you care about. A sample that is not representative can mislead you even if it is large.
sample size: The number of items or people you actually collect data from. In statistics it is usually written as n.

Example

Imagine you want to know how many people at a city bus stop prefer tea to coffee. If you ask only three people at the airport gate, your answer may be far off, even if all three say tea. Asking 124 people from many places gives a much fairer picture of the whole city.

3 checked practice questions for this part are in the app.

Part 7 · 1:00:00–1:10:00

This part explains how to tell whether a number calculated from data is a good guess for the real value, and why we model the couples as random coin-like outcomes.

Key ideas

An estimator is a recipe for guessing a value, with the data still left open. An estimate is the number you get when you plug in the data you actually collected. 1:02:13
Because the data could have come out differently, an estimator is a random variable. A good estimator should not jump around too much from study to study (low variance), and its typical value should be close to the true value (low bias). 1:03:16
Each couple's answer is 0 or 1, so it follows a Bernoulli distribution, which is the simplest model for a yes-or-no outcome. 1:05:47
We assume every couple has the same chance p of turning right, and that the couples are independent of each other. This lets us pool all the data into one estimate. 1:06:10
Treating the population as one group with one p is a simplifying assumption. We lose detail about individuals, but we gain the ability to learn from everyone together. 1:07:55

Words to know

Estimator: A rule or formula for guessing a value from data. Before you collect the data, its result is unknown, so it counts as a random variable.
Indicator: A simple switch variable that equals 1 if something is true and 0 if it is not.
Bernoulli: The name for a random outcome that can only be one of two values, usually 0 or 1, like heads or tails.
Independent: Each outcome does not affect the others. One couple's choice tells you nothing about another couple's choice.

Example

Imagine guessing the share of red marbles in a large bag by pulling out handfuls. A handful of two marbles gives a very jumpy guess, while a handful of 124 gives a steadier one. A guess of 72% that never changes is steady, but it may be far from the truth.

3 checked practice questions for this part are in the app.

Part 8 · 1:10:00–1:17:51

This part shows how to make a simple guess about a class's grades from a small random sample, using a Gaussian (normal) model and averages.

Key ideas

The teacher assumes each student's grade follows a Gaussian (normal) distribution with some mean mu and variance sigma squared. This is a modeling choice, not a law of nature. 1:10:22
The Gaussian choice is mainly for convenience. It even gives some chance of negative scores, but the teacher says that chance is tiny, so it is good enough. 1:10:43
A good estimate for mu is the average of the scores. For the 15 sample grades, the average is 67.5. 1:12:12
The main rule of statistics is to replace each expectation (the long-run average) with an average from the data. Using this rule, the estimate for sigma comes out to about 18. 1:13:47
Before trusting these numbers, you should ask whether the grades are IID (independent, with the same distribution). The point of the exercise is to think about these assumptions. 1:15:42

Words to know

estimator: A rule or formula that uses the data to guess an unknown number, such as the average grade.
Gaussian distribution: A bell-shaped pattern of values, also called a normal distribution, where most values sit near the middle.
expectation: The average value you would expect over many repeats of something random.
IID: Independent and identically distributed. It means each item is separate from the others and follows the same pattern.

Example

Think of guessing the average height of people in a town by measuring 15 random people and taking their average. You cannot measure everyone, so you use the sample average as your guess, and you ask whether those 15 people were chosen fairly and independently.

3 checked practice questions for this part are in the app.

Notes written by AI (Claude) from the lecture's official transcript, so check anything important against the lecture. MIT OpenCourseWare · CC BY-NC-SA 4.0 · not affiliated with or endorsed by MIT. These notes are shared under the same license.

LessonBound · Free lecture notes · Privacy · Terms · New York