How to lie with data?
In a world this uncertain, data is supposed to be the point of reference: something you can rely on to tell you what is actually true. Yet two people can take exactly the same data, change not a single digit, and reach opposite conclusions. I brainstormed every way that can happen.
The setup
Meridian is a made-up city of 12,000 people with a free job-training program called PathUp. I generated all of its data myself, so I know the truth: PathUp raises the yearly earnings of everyone who takes it by exactly $1,400.
Two important things about the data before we move forward:
- People choose whether to join, and the program recruits where need is greatest, so the people who sign up start out poorer than the people who don’t, before the program has done anything.
- Each person has a hidden “stability” score, standing in for things like steady housing, childcare and health. It helps people earn more and helps them finish the course, and it appears in no dataset.
We have two groups of people looking at the same data:
- The Program Office, which wants to show that PathUp changes lives.
- The Opposition, which wants to show that PathUp wastes $9.6 million a year.
Now let’s see how both of them can get what they want without changing the data.
1. Zooming in
These charts show Meridian’s unemployment rate, the share of people who want a job but don’t have one, for each month of 2025. All three plot the same twelve numbers. The first uses a scale from 4.9% to 5.6%, the second a scale from 0% to 10%, and the third adds every month back to 2018.
Each side picks the chart that tells its story.
The Program Office, with the first chart
“Unemployment is in freefall. PathUp is transforming Meridian’s job market.”
The Opposition, with the second chart
“Unemployment hasn’t budged all year. PathUp isn’t doing anything for jobs.”
2. Picking the start date
These charts all show Meridian’s monthly unemployment rate. The first covers September 2018 to April 2020, the second April 2020 to August 2023, and the third the whole line from 2018 to 2025, with the first two stretches shaded.
Each side picks the stretch that tells its story.
The Program Office, with the second chart
“Unemployment has fallen 76% since PathUp expanded.”
The Opposition, with the first chart
“Unemployment rose 121% on this administration’s watch.”
3. Changing what you divide by
This chart shows how much PathUp’s spending changed between 2018 and 2025, measured four ways: in total dollars, in dollars per resident, in dollars after inflation, and as a share of the city’s whole budget.
Each side picks the measure that tells its story.
The Opposition, with the top bar
“PathUp spending has exploded, up 66% since 2018.”
The Program Office, with the bottom bar
“PathUp’s share of the city budget has actually shrunk.”
4. Picking an average
This chart shows five ways to answer one question: what does a typical PathUp participant earn in a year? All five are calculated from the same list of participants’ earnings after the program.
Each side picks the average that tells its story.
The Program Office, with the most common amount
“The typical participant now earns $40,000 a year.”
The Opposition, with the mean
“The average participant earns just $34,298.”
5. Comparing the wrong groups
These charts show average yearly earnings after the program. The first compares everyone who joined PathUp with everyone who didn’t. The second splits the same 12,000 people into four groups by education, with the share of each group that joined written underneath.
Each side picks the chart that tells its story.
The Opposition, with the first chart
“PathUp participants earn $5,647 a year less than everyone else. The program is hurting the people it serves.”
The Program Office, with the second chart
“Whatever your education, joining PathUp means earning more.”
6. Cleaning the data until it agrees
Every dataset gets cleaned before it is analysed: bad records removed, extreme values trimmed, like compared with like. This chart runs the same comparison, people who joined against everyone else, under nine different cleaning rules, each with the reason someone would give for it. The dashed line is the true effect, $1,400.
Each side picks the rule that tells its story.
The Program Office, with the last rule
“Among low earners who finished, PathUp adds $2,087 a year.”
The Opposition, with the second rule
“Once bad records are removed, PathUp participants earn $6,355 less.”
7. Looking until you find something
This chart measures PathUp’s effect separately in 19 groups of people: by education, neighbourhood, age, children, transport and past earnings. Each dot is one group’s result. The dashed line is the true effect, which I built to be exactly $1,400 for everyone.
Each side picks the group that tells its story.
The Program Office, with the top dot
“PathUp works 2.6 times as well for degree holders: $3,616 a year.”
The Opposition, with the bottom dot
“In Fairmount, PathUp delivers just $1,081 a year.”
8. Two lines that rise together
The first chart shows two numbers from 2018 to 2025: PathUp’s monthly spending, on the left scale, and the number of streaming subscribers in Meridian, on the right scale. The second takes each month’s change in both numbers and plots one against the other, one dot per month.
This time both sides use the same chart.
The Program Office, with the first chart
“PathUp spending tracks streaming subscriptions almost perfectly. The city’s investment is changing how households spend.”
The Opposition, with the first chart
“As PathUp spending rose, so did streaming. The program is paying people to stay home and watch TV.”
9. Only counting the people who finished
The first chart shows average yearly earnings for three groups: people who never joined PathUp, people who joined and dropped out, and people who finished. The second shows the share of each group with a stable life (steady housing, childcare, transport and health), the hidden score no dataset records.
Both sides use the first chart.
The Program Office, with the first chart
“PathUp graduates earn almost $8,000 a year more than dropouts. Finishing is what counts, so fund completion support.”
The Opposition, with the first chart
“People who joined and dropped out earn almost $10,000 less than people who never joined. PathUp leaves people worse off.”
10. Moving where “good” starts
PathUp asked participants to rate it on a five-point scale. The first chart shows how they answered. The second shows five headlines written from those same answers.
Each side picks the headline that tells its story.
The Program Office, with the first headline
“69% of participants rated PathUp neutral or better.”
The Opposition, with the third headline
“63% of participants were not satisfied.”
Over to you
Every chart above uses true numbers. What changed each time was a choice: where the scale starts, which months to show, what to divide by, which people to count. None of those choices looks like lying, and that is what makes them dangerous. Once you know the answer you want, each choice quietly leans toward it, until the truth is hard to see at all. As the economist Ronald Coase put it, if you torture the data long enough, it will confess.
So the next time you read a chart, or make one, think about the choices behind it. What would it show if they had gone the other way? And what questions would you ask to make sure you are seeing what the data says, rather than the story someone wanted it to tell?