ContentsThe library

Chance, Spread and What to Expect

What Is Left After the News

Last timeTwo Things at Once

Conditioning restricts the table to the row you now know you are in, and reversing the direction of a conditional requires the base rate, which never quietly cancels.

Something has been learned. A request failed, a test came back positive, a user

clicked. The question is what is now true of everything else. This is the single

most useful operation in the subject, and it is also where trained people make

confident errors, so it is worth doing slowly.

Start from the table of the last lesson. Learning a piece of news means that

some rows or columns are no longer possible. Delete them. What remains no longer

sums to one, because weight has been thrown away, so divide by what survived.

FIG 1Conditioning is deletion followed by a division
the weight that is consistent with both, the part of A that survives the news
the total weight that survived the news, the normalising divisor
the weight on A within the restricted world, read as the chance of A given B
the division is the whole trick, and it is why conditioning on something unlikely can change a number enormously
The definition requires that B have non zero weight, which is to say that you cannot condition on news that was impossible. Everything else in this lesson is a rearrangement of this one line.

Multiply both sides by the divisor and you have a statement worth naming in its

own right: a joint weight is a conditional times a margin.

FIG 2Building a joint distribution in the direction people think in
the joint weight, which is what the table actually holds
how often the second follows when the first has happened, the mechanism
how often the first thing happens at all, the base rate
This is how joint distributions get written down in practice. Nobody knows a table of nine numbers directly. People know a base rate and a mechanism: how common the condition is, and how the test behaves on someone who has it. The joint is then assembled rather than measured.

Turning a conditional around

Here is the step that matters. The mechanism is almost always known in one

direction and wanted in the other. A test is designed and validated by giving it

to people whose condition is known, so what is measured is the chance of a

positive given illness. What a patient wants is the chance of illness given a

positive. These are different numbers.

Write the joint weight both ways, once with each quantity in the conditioning

position. Both expressions equal the same joint weight, so they equal each other,

and solving for the one that is wanted gives the reversal rule. The base rate

appears in it because it has to: the two directions genuinely differ, and the

base rate is exactly the information that separates them.

It is worth being clear about what that derivation is and is not. It is two

lines of algebra applied to a definition, with no new assumption anywhere, which

is why the result holds universally and why there is nothing to argue with. The

controversy that has attached itself to the name belongs to a different question,

namely where the base rate comes from when nobody has counted it. The rule

itself is bookkeeping, and the three lessons that precede this one already

contain everything it uses.

The lesson stops here

3 more paragraphs to go

You have read the opening. The rest of the argument, the problems that check whether it landed, and the lines worth keeping at the end all come with a plan.

The first lesson of every course in the library reads the whole way through, free, so you can see exactly what the rest of them are.

See the planThe contents

This is the reading half

Starting the course gives you your own copy of it. Every idea on every page has problems standing under it, marked with a reason rather than a tick, and any sentence you do not believe can be opened and argued with. None of that can happen on a page nobody owns.

The contents

The rest of this course

  1. 01A List, and a Number on Each Item
  2. 02One Number Standing in for a Whole Listopening only
  3. 03The Second Number Every Average Needsopening only
  4. 04A Table Instead of a Listopening only
  5. 05What Is Left After the Newsyou are here
  6. 06Four Times the Data for Half the Erroropening only
  7. 07Height Is Not Weightopening only
  8. 08The Number and the Bar Beside Itopening only

Read alongside