How to Lie with a Chart SOC 1113 · Trend Lab · Midwestern State University
Every chart on this page plots real, published data, and the numbers never change. The controls change only how the numbers are drawn: where the axis starts, which way it points, which years you see, the shape of the frame, how much averaging is applied, whether you see totals or rates, and whether the curve is a running sum. Seven presentation choices, seven different stories, the same data every time. When you can name the trick, you can catch it.
How the buttons choose
Nothing hidden: the scary button searches every window at least seven years long for the largest percent rise between its endpoints. The reassuring button does the same for the largest percent fall. Both buttons find real windows in real data. That is the point. A motivated chartmaker can nearly always find a window that says what they need, without touching a single number.
Every other tab on this page slants a true story. An upside-down axis is different: the picture itself says the opposite of the data. The numbers on the axis are still printed correctly, which is the fig leaf. A real news graphic once showed gun deaths this way, and readers saw a decline where there was a rise.
Check the tick labels. If the numbers get smaller as they go up, the chart is upside down.
When smoothing is honest work
Averaging is a legitimate tool for seeing a long trend through noisy years, and analysts use it constantly. It becomes a trick in two situations: when the smoothing is not labeled, and when the window is chosen to erase the exact year the argument is about. The gray line stays on this chart for a reason. Ask any smoothed chart where its raw data went.
Each line gets its own y-axis, and each axis can be stretched and shifted independently. With two free axes, almost any two series that drift the same general direction can be made to lie on top of each other. The overlap is manufactured by the axes, not found in the data, and the picture whispers a causal story no one would say out loud.
Marriage is per 1,000 people. Homicide is per 100,000. They do not even share units.
The population roughly doubled, from 180.7 million in 1960 to 330.2 million in 2019. A flat rate spread over far more people produces far more total events. Totals answer how much. Rates answer how likely. A chart that shows totals when the question is about risk is answering a question nobody asked.
Totals here are computed from the official rate and World Bank population figures, so they track the true counts closely. Both source series are published and linked below.
A running total of anything that cannot go below zero can only rise. "All-time high" is true of a cumulative curve in every single year, by construction. It sounds like news and contains none. When a chart shows a total "since" some year, ask what the annual number is doing.
Annual wedding counts are computed from the official marriage rate and World Bank population figures.
Each round draws one of this page's real series with one presentation choice applied, or none at all. Nothing is fabricated. Read the axis, the years, and the units before you answer.
With one exception, every headline on this page is true
That is the uncomfortable part. Apart from the upside-down axis, whose picture is simply false, none of these tricks requires a false number. They change how big a change feels, which years count, which quantity answers the question, and what connection the eye infers. The defense is a habit, not a fact-check: read the axis, the window, and the units before you read the line.
When a zoomed axis is the honest choice
Sometimes small differences are the whole point. A hospital fever chart never starts at zero degrees, because a two-degree change in body temperature matters and every reader knows the scale. A zoomed axis is defensible when the meaningful range is well known and the chart says clearly what it is doing. It turns into a trick when the zoom is chosen to manufacture drama for readers who never look at the axis.
The checklist
- A y-axis that does not start at zero. The Axis tab. Small changes fill the whole frame.
- An upside-down axis. The Upside down tab. Tick labels that shrink going up mean the picture points the wrong way.
- A cherry-picked time window. The Window tab, buttons included. Ask what the full series shows.
- A frame shaped for drama. The Frame tab. Tall and narrow makes any slope a cliff.
- Unlabeled smoothing. The Average tab. Ask a smooth line where its raw data went.
- Two fitted axes. The Second axis tab. Free axes can make almost any two drifting lines coincide.
- Counts where rates belong. The Counts tab. Totals rise with population even when risk does not.
- A cumulative curve. The Cumulative tab. A running total can only go up.
- No source and no sample size. The one this page cannot demonstrate for you. A chart that will not say where its numbers came from has not earned your trust.
Questions to take with you
- Every headline on this page is factually accurate. What exactly made the alarming ones feel more alarming?
- Who benefits when a small change looks big? Think of a seller, a political campaign, and a news outlet, and name what each one gains.
- The axis trick can hide change as well as exaggerate it. How would you set up a chart to make a real rise look flat?
- The scary and reassuring buttons both search the same series. What does that say about arguments that begin with "just look at the trend since"?
- Totals or rates: which one answers "is flying getting safer," and which one answers "how many crashes were there"? Why do the two get swapped?
- If two free axes can make almost any two declining lines overlap, what should a two-line chart have to show before you believe the connection?
- Nearly every trick here has an honest use: zooming, smoothing, windowing, even dual axes. What separates the honest use from the trick, in one sentence?
- When is a zoomed axis the honest choice, and what does the chart owe its reader in that case?