Distorting Data to Mislead

Data does not speak for itself. The moment someone turns numbers into a graph, a percentage or a headline, they make choices, and those choices can distort the message. To distort data means to present it in a way that pushes a particular viewpoint rather than showing the honest picture. The data can be completely accurate and the conclusion still be misleading, because the trick is in how the data is shown, not in the numbers themselves.

Imagine a company sold 96,000 phones last year and 98,000 this year, a rise of about 2%. Drawn honestly, the two bars look almost the same. But if you start the vertical axis at 95,000 instead of 0, the new bar suddenly looks three times taller. Same data, very different story. Learning to read past the picture to the real numbers is one of the most useful skills in data science.

Graphs are the most common place data gets distorted, because a picture is so persuasive. The classic trick is the truncated y-axis: by starting the scale above zero, small differences look enormous. A close cousin is an inconsistent or reversed scale, where the gaps between numbers are not even, or the axis runs backwards so a fall looks like a rise. Another is cutting off part of the data range, showing only the window that fits the story, sometimes called cherry-picking a window: a graph that begins on the coldest day of the year can make a warming trend vanish.

Watch out too for 3D and area tricks. When a bar is drawn as a 3D block, or one circle is made twice as wide as another to show "double", the eye reads the whole volume or area, which grows far faster than the real value, so the difference looks much bigger than it is. Finally, a graph with two different y-axes can place any two lines on top of each other and make it look like they move together, even when there is no real link.

Numbers can mislead even without a graph. Cherry-picking data means showing only the results that fit your conclusion and quietly dropping the rest: a supplement company might publish the one study where its pill worked and ignore the ten where it did nothing. Misleading averages are another favourite. The mean can be dragged far from what is typical by one extreme value, an outlier. If nine workers earn $50,000 and the boss earns $950,000, the mean wage is $140,000, but the median of $50,000 describes a typical worker far better. Choosing the average that flatters your case is a form of distortion.

Then there are percentages with no base. "Cases up 50%!" sounds alarming until you learn that means two cases became three. A small or biased sample distorts too: asking five friends and reporting "80% agree" hides how few people were asked and how unrepresentative they are. Finally, correlation presented as causation sneaks a cause into data that only shows two things moving together. Ice-cream sales and drownings both rise in summer, but ice-cream does not cause drownings; hot weather drives both. You will study cause and correlation properly in Lesson 14.

You cannot stop people distorting data, but you can refuse to be fooled by it. Build a quick checklist and run it on every chart, statistic or headline you meet. First, check the axes: does the scale start at zero, are the gaps even, and does the time range cover the whole story? Second, check the sample: how many people or items were measured, and were they chosen fairly or just the convenient few? Third, check the source: who made this, and do they gain if you believe it? An ad, a lobby group and an independent agency such as the ABS are not equally trustworthy.

Fourth, and easiest to forget, ask what is left out. What data was not shown? Which years, which results, which group of people are missing? Distortion often works by omission, the misleading part is the bit that was quietly removed. If you run these four checks, axes, sample, source and what is missing, most distortions fall apart in seconds.