Honest Charts and Data Visuals
Learn how charts mislead and how to prevent it: zero-based bars, direct labels and the lie factor, measured on real pictures of the same data drawn two ways.
In this lesson you will
- Explain why bars start at zero
- Measure the lie factor
- Choose the right chart type
- Label charts honestly
About 22 minutes
Two bars, one big claim
Priya is writing the library’s annual report. Visitors this year: 52 thousand. Last year: 50 thousand. It is a small rise, but she wants it to look like good news. In her chart she starts the vertical axis at 48 instead of zero, and suddenly this year’s bar is twice as tall as last year’s. The committee is impressed until Tom, who likes numbers, asks, ‘Did visitors double?’ They did not. They rose by 4 per cent. Priya’s chart told a story the numbers did not. This lesson is about charts that tell the truth, and about the one measure that tells you when they do not.
What a chart is for
A chart is a picture of numbers, and it works only if the picture’s proportions match the numbers’ proportions. When one bar is twice as tall as another, the eye reads ‘twice as much’. So the rule for bar charts is strict: the axis starts at zero, because the bars’ length is the data. For line charts there is more freedom, because the eye reads the slope and position rather than length, but even then a squashed or stretched axis can mislead.
Measuring the distortion
The statistician Edward Tufte proposed a simple test he called the lie factor. It compares the size of an effect as shown in the picture with its size in the data. If the picture exaggerates a 4 per cent rise by three times, the lie factor is three. A fair chart has a lie factor of about one.
The program below draws Priya’s two versions as real pictures, 52 against 50, once with the axis starting at zero and once at 48. Then it does something clever: it reads the finished picture back, counts the purple pixels in a column through each bar, and uses those heights to compute the ratios. That means the numbers describe what the viewer actually sees, not what we intended to draw.
from PIL import Image, ImageDraw
def draw_bars(values, axis_min, path):
W, H, base = 400, 300, 280
top = 20
img = Image.new("RGB", (W, H), "white")
d = ImageDraw.Draw(img)
scale = (base - top) / (max(values) - axis_min)
for i, v in enumerate(values):
x = 60 + i * 150
h = round((v - axis_min) * scale)
d.rectangle((x, base - h, x + 90, base), fill=(90, 40, 200))
img.save(path)
return img
def read_heights(img):
out = []
for x in (100, 250): # a column through the middle of each bar
col = [y for y in range(img.height) if img.getpixel((x, y)) == (90, 40, 200)]
out.append(len(col))
return out
values = [52, 50]
honest = draw_bars(values, 0, "honest.png")
cut = draw_bars(values, 48, "cut.png")
h1, h2 = read_heights(honest), read_heights(cut)
data_ratio = values[0] / values[1]
print(f"data: {values[0]} vs {values[1]}, ratio {data_ratio:.2f}")
print(f"zero-based bars: {h1[0]} vs {h1[1]} px, ratio {h1[0] / h1[1]:.2f}")
print(f"axis from 48: {h2[0]} vs {h2[1]} px, ratio {h2[0] / h2[1]:.2f}")
def lie_factor(shown, actual):
return ((shown - 1)) / ((actual - 1))
print(f"lie factor, zero axis: {lie_factor(h1[0] / h1[1], data_ratio):.1f}")
print(f"lie factor, cut axis: {lie_factor(h2[0] / h2[1], data_ratio):.1f}")
data: 52 vs 50, ratio 1.04
zero-based bars: 261 vs 251 px, ratio 1.04
axis from 48: 261 vs 131 px, ratio 1.99
lie factor, zero axis: 1.0
lie factor, cut axis: 24.8
With the axis at zero, the bars are 261 and 251 pixels tall, a ratio of 1.04, matching the data ratio of 1.04. The lie factor is 1.0, which is what honesty looks like. With the axis starting at 48, the bars are 261 and 131 pixels, and the ratio becomes 1.99. The data rose by 4 per cent, but the picture shows a rise of nearly 100 per cent. The lie factor is 24.8. Put another way, the chart exaggerated the change by about twenty-five times. Tom was right.
Rules for honest charts
- Bars start at zero. If the differences are too small to see, say so in words or show the numbers on the bars.
- One scale per chart. Do not give two lines their own axes to make them look as if they move together.
- Show the numbers. Label bars directly and cut the grid lines you do not need.
- Keep proportions fair. Do not stretch a chart so much that gentle changes look steep or steep ones look flat.
- Choose the right chart. Bars for comparing amounts, lines for change over time, a small table when only a few exact numbers matter.
- Use colour carefully. Follow lesson 22: label directly rather than leaning on colour, and test in greyscale.
Canva offers chart elements and spreadsheet-style tools in some designs, and the kinds of chart and how you enter data vary by version and plan. The rules above hold for any of them. If you must use a broken axis, mark it clearly with a visible break and a note, and consider whether another way of showing the data would be fairer.
Pie charts and decoration
Pie charts are fine for showing a few parts of a whole, such as three or four categories adding to 100 per cent, but they are poor for comparing similar slices or for many categories. Avoid three-dimensional effects, because tilting a chart changes how large each part looks. Skip pictures of books or people that scale in both width and height, because area grows with the square, and the eye misjudges it.
Choosing the axis to suit your argument. If you notice yourself adjusting a chart until it looks the way you want, stop and ask what a sceptical reader would see. Decide the honest scale first, and then work out how to make the real message clear.
Readers often judge a whole report by its charts. One misleading chart can cost you the trust of every number you show, so honesty is also good strategy.
Practice for this lesson
Check what you learned
Five quick questions. Pick an answer and you will see straight away whether it is right and why.
Where should a bar chart axis start?
- At the smallest value
- At zero
- At 48
- Anywhere
Bar length is the data, so a bar axis must start at zero.
What is the lie factor?
- The number of colours used
- A kind of font
- The size of the file
- Effect shown in the picture divided by effect in the data
A lie factor near 1 means the picture is as large as the data change; much larger means exaggeration.
What happened when the axis started at 48?
- The chart got smaller
- A 4 per cent rise looked like nearly 100 per cent
- The numbers changed
- The bars disappeared
The bars measured about 2 to 1, though the data ratio was 1.04.
Why read the heights from the finished picture?
- To save the file
- To change colours
- To add labels
- To measure what the viewer actually sees
Counting pixels in the image gives the real proportions viewers see.
Which is a good way to label a chart?
- Put the numbers on the bars
- Hide the numbers
- Use only colour
- Use a 3D tilt
Direct labels let readers see the real values without relying on the axis or colour.
Summary
- A bar chart’s axis must start at zero because length is the data.
- The lie factor compares the effect shown with the effect in the data.
- A cut axis made a 4 per cent rise look like a doubling.
- Label directly, keep one scale, and choose the chart type to suit the message.
Questions
Can a line chart start above zero?
Often yes, because the eye reads the slope and position, not the length. Still, label the axis clearly and check that the picture does not exaggerate the change.
Does Canva have chart tools?
Canva offers chart elements and ways to enter data, and what is available depends on your version and plan. The honesty rules apply to every tool.
What if the difference is too small to see?
Say so in words, show the numbers, or use a different kind of visual. Do not cut the axis to make a small change look big.
