ContentsThe library

Chance, Spread and What to Expect

A Table Instead of a List

Last timeHow Far From Expected

Two quantities at once need a table of weights rather than a list, and independence is the strong claim that the table is nothing more than its own two margins multiplied.

Everything so far has concerned one quantity. In practice there are always

several, and the interesting questions are about how they relate. Does a slow

request make the next one slow? Does a user who clicked once click again? Does a

model that is confident tend to be right?

The object needed is a small extension of the first lesson. With one quantity the

outcomes formed a list. With two, an outcome is a pair, so the weights form a

table.

FIG 1A joint table for two quantities
next was fastnext was slownext failedrow totals
request was fast0.450.100.050.60
request was slow0.150.120.030.30
request failed0.020.030.050.10
column totals0.620.250.131.00
Nine pairs, each with a weight, summing to one. The final row and column are not part of the table proper; they are the totals. The marked cell is the striking one: a failure is followed by another failure in five cases in a hundred, against the one and a third in a hundred that would be expected if the two were unrelated, which is nearly four times as often.

The two rules are unchanged. No entry is negative, and the whole table sums to

one. Everything there is to know about the relationship between the two

quantities is already sitting in that table; the rest of this lesson is about

reading it.

Margins

Often only one of the two quantities is wanted. What is the chance this request

is slow, regardless of what the next one does? That is the sum of a whole row,

and the row and column totals are called the margins, from the practice of

writing them in the margin of the table.

The margins are what most reporting shows. A dashboard that gives the failure

rate for this hour and the failure rate for the next hour has shown two margins

and thrown the table away, and the discarded part is exactly the part that says

whether failures arrive in clusters. Two services with identical margins can

behave completely differently: one fails at random moments and the other fails in

bursts that take out a whole minute. The hourly numbers are the same. The

experience of using them is not.

The lesson stops here

4 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 Listyou are here
  5. 05What Is Left After the Newsopening only
  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