Turn relative changes into absolute ones and ask for the base rate.
Pick up any news story today that reports a percentage change and look for two numbers: where the figure started and where it ended. If the article doesn't give both, write that down. A percentage on its own hides the size of what changed, and there are four common ways it misleads. You can check all four with arithmetic you already know.
Say a company reports that customer complaints are up 50 percent (an invented example). That could mean complaints went from 4 to 6. It could also mean they went from 4,000 to 6,000. Both are 50 percent rises. The first is two extra unhappy customers, and the second is a serious problem.
A percentage change says nothing about size on its own. The same rise can be trivial or a crisis depending on the starting number, so whenever you see one, ask for the starting number and the ending number. When a small base produces a large percentage, the story is usually smaller than it sounds.
If a report or article leaves the base out, note it. The writer had the base and chose not to show it.
Priya's manager forwards a news headline to the team chat with the single word "Worrying." The headline says a side effect of a common medicine has risen by 50 percent. Two colleagues reply that they're stopping the medicine. Priya opens the article, scrolls to the eighth paragraph and finds the actual numbers.
Priya replies in the chat with three lines: the old risk was 2 in every 10,000 people, the new estimate is 3 in every 10,000, and that's one extra case per 10,000 people. One colleague answers that it's nothing like what they thought. Nobody needs to stop anything without asking a doctor first.
Going from 2 to 3 is a 50 percent rise, so the headline was accurate. It still gave a misleading impression, because it reported relative risk: how much more likely something is in one group compared with another. What Priya found in the eighth paragraph was the absolute risk: how likely something is in the first place. Headlines prefer relative risk because it produces a bigger number. Advertisements prefer it too, but for reductions. A product that "cuts your risk by half" may be cutting it from 2 in 1,000 to 1 in 1,000.
Whenever you see relative risk, convert it by asking three things: out of how many people, how many before, and how many after. For someone taking a medicine for a good reason, a small absolute rise changes very little, and the decision about whether to continue belongs with their doctor.
When a rate is already a percentage, a change in it can be described in two ways, and people mix them up constantly. Suppose unemployment in one sector goes from 2 percent to 3 percent. That is a rise of one percentage point, which is the difference between two percentages: 3 minus 2 equals 1. It is also a 50 percent rise, because the rate went from 2 to 3. Both descriptions are true at once, and they sound very different.
A loan rate that moves from 4 percent to 5 percent (an invented example) can be reported as "rose 1 point" or "jumped 25 percent". The same applies to unemployment rates, interest rates, survey results and conversion rates. When you read a change in a rate, check which description is being used. When you write one, say "percentage points" for the difference between two percentages.
Base rate neglect is the trap that catches even careful people. It happens when a vivid detail, such as how accurate a test is, makes a result feel decisive, while the base rate, meaning how common the thing is in general, is what decides the answer.
Take a bank's fraud system. Fraud happens in 1 in every 1,000 transactions. The system catches 99 percent of real fraud and wrongly flags 2 percent of honest transactions. The question is how likely a flagged transaction is to be real fraud. Most people asked this guess close to 99 percent. The correct answer is under 5 percent.
You can solve it by working through a large round number of cases:
Start with 100,000 transactions. At 1 in 1,000, 100 of them are fraudulent. The system catches 99 percent of those, which is 99. That leaves 99,900 honest transactions. 2 percent of 99,900 is 1,998 false alarms. Add 99 and 1,998 to get 2,097 flagged transactions in total. Divide 99 by 2,097 and you get under 5 percent.
So more than 95 percent of the flags are false alarms. When the thing being looked for is rare, even a good test produces mostly false alarms, because the small error rate applies to a very large number of honest cases.
The same method works for any problem of this kind. Pick a large round number of cases, apply the base rate to get the number of real cases, apply the detection rate to get the true positives, apply the false-alarm rate to the remaining cases to get the false positives, add the two together for the total flagged, and divide the true positives by that total.
Base-rate logic applies well beyond fraud. A job applicant with a polished CV, a startup that "looks like" a past success, and a customer whose behaviour "fits the profile" of someone about to leave all invite the same mistake. Before judging a case, ask how common the outcome is in general. Lesson 7.4, "Update your beliefs with new evidence", shows how to start from the base rate and adjust it.
The habit to build is three conversions: relative change to absolute numbers, percent to percentage points, and a vivid case to how common it is. The activity below asks you to do the conversion on three real headlines, and any news site will give you more practice material this week.
Take three headline statistics from recent news and rewrite them as absolute numbers. Note where the article did not give you enough to do it.
Junxiong-WFG Organisation is an authorised representative of AIA Financial Advisers Private Limited (Reg. No. 201715016G).