Find the Base Rate
Introduction 🎉
Welcome to Evaluate Rare Events and Test Results! In the previous course, you built a strong foundation for spotting randomness, resisting pattern illusions, and questioning evidence from small samples. Those ideas share a common thread: our intuitions often fail when randomness, uncertainty, and limited information are involved. In this course, you will apply that same caution to test results, alerts, and screening outcomes, especially when the event in question is uncommon.
This first lesson introduces the single most important ingredient in that process: the base rate. By the end of this lesson, you will have learned to:
- Define a base rate as the background frequency of an outcome before any test, filter, or extra information is applied.
- Identify the correct outcome and reference group so you know exactly what the base rate describes and avoid applying it to the wrong population.
- Convert percentages into natural frequencies such as “20 out of 1,000” to make rare events easier to picture and reason about.
Imagine a doctor tells a patient, "This test is 95% accurate and your result is positive." Most people hear that and assume there is a 95% chance they have the condition. But whether that conclusion is reasonable depends almost entirely on how common the condition was before the test was run. If only 1 in 1,000 people in the relevant population actually has it, the math changes dramatically.
The background frequency acts like an anchor. Without it, even an impressive-sounding accuracy number can lead you to very wrong conclusions. Before you can evaluate any test or alert, you first need to find that anchor.
What Is a Base Rate? 📊
A base rate is how often something occurs in a given group before any test, filter, or additional information is applied. Think of it as the "background frequency" of an event.
Here are a few everyday examples:
- Fraud detection: Out of every 10,000 credit card transactions, about 15 are fraudulent. The base rate of fraud is .
- Email spam: In a particular inbox, roughly 200 out of every 1,000 messages are spam. The base rate is .
- Medical screening: A certain genetic condition affects 1 in 500 newborns. The base rate is .
Notice that a base rate always comes with two pieces of data: a frequency (how many) and a group (out of whom). Missing either piece makes the number meaningless.
One of the most common mistakes when working with base rates is attaching the frequency to the wrong group. Consider this claim: "5% of people who visit the emergency room have a heart attack." That 5% applies specifically to emergency room visitors, not to the general population. If you accidentally applied it to everyone in a city, your estimates would be far too high.
When identifying a base rate, always ask two questions:
- What is the outcome? (the event we care about)
- What is the reference group? (the specific population the frequency describes)
Getting the group wrong is like measuring temperature with the wrong thermometer. The number might look precise, but it will point you in the wrong direction.
From Percentages to Natural Frequencies 🔄
Percentages are compact, but the human brain handles whole counts much more naturally. A technique called natural frequencies converts rates into simple "X out of Y" statements using a concrete population size. This small translation step makes the math dramatically easier to follow.
Example: Suppose a disease affects 2% of a population. Let's convert that into natural frequencies using a group of 1,000 people.
Now instead of thinking in abstract percentages, picture a room of 1,000 people where 20 people have the disease and 980 people do not. This sets the stage for every calculation that follows.
A simple population sketch makes this split easier to see:
| Group | Count |
|---|---|
| Have the disease | 20 |
| Do not have it | 980 |
| Total | 1,000 |
Choosing a round total like 100, 1,000, or 10,000 keeps the arithmetic clean. Pick whichever size turns the percentages into whole numbers.
Tracing a Full Example 🧭
Let's walk through a complete mini-scenario from start to finish. A bank reports that 0.3% of its daily transactions are fraudulent. Yesterday the bank processed 10,000 transactions.
Step 1 — Identify the base rate. The background frequency of fraud is 0.3%, and the reference group is daily transactions at this bank.
Step 2 — Convert to natural frequencies.
So out of 10,000 transactions, you expect roughly 30 to be fraudulent and 9,970 to be legitimate. These concrete numbers become the foundation for later questions such as, "If the fraud detector flags a transaction, how likely is it actually fraudulent?"
Conclusion and Next Steps
In this lesson you learned that a base rate is the background frequency of an outcome within a clearly defined group, and that every base rate requires two parts: the frequency and the reference group it belongs to. You also practiced converting abstract percentages into natural frequencies — concrete counts that make it much easier to reason about how outcomes spread across a population. These ideas may seem straightforward on their own, but they are the foundation for everything else in this course.
Up next, you will put these concepts into practice with a set of hands-on exercises. You will extract base rates from real-world scenarios, translate percentages into counts, and even explain to someone why ignoring the base rate leads to faulty conclusions — so let's dive in!
