Misleading Graphs: 10 Examples and How to Spot Each One

A misleading graph shows accurate numbers in a way that makes them look bigger, smaller, steadier, or more connected than they are. Most rely on a small set of recurring tricks, and each one leaves a visible clue in the axes, the scale, the time range, or the labels. Below are the ten you'll see most, how to spot each one, and the one-line fix.

What makes a graph misleading?

A graph is misleading when the impression it gives at a glance differs from what the numbers support. The data itself is usually accurate. The distortion comes from construction choices: where an axis starts, which way it runs, how marks are sized, and what time period and denominator are shown. Edward Tufte's first principle of graphical integrity, from The Visual Display of Quantitative Information (1983), is that the size of the marks on a graphic should be directly proportional to the quantities they represent. His lie factor measures how far a chart departs from that rule. Truncated bars, scaled icons, and 3D effects break it directly. The other tricks can pass the proportionality test and still mislead through axis direction or spacing, the time window, what is being counted, or the precision they imply.

Misleading doesn't require intent. Auto-scaled axes, default 3D styles, and "total to date" dashboard tiles produce the same distortions as deliberate spin, so apply the same checks to every chart, including your own.

What are the 10 most common misleading graph examples?

These ten tricks turn up again and again in news stories, marketing decks, and internal dashboards. Each entry gives the mechanism, the red flag to look for, and the fix.

1. Truncated baseline on a bar chart

Bar length encodes the value, so bars have to start at zero. Start the axis at 90 and a bar for 100 is drawn twice as tall as a bar for 95, even though the difference is about 5%. It is one of the most common tricks, and dashboard tools that auto-scale the axis can produce it without anyone deciding to.

Spot it: the lowest value on a bar or column chart's axis isn't 0, or the axis is hidden and the bar heights don't match the ratio of the printed values.

Fix: start bars at zero, or switch to a dot plot or line chart with a clearly labeled axis if the meaningful variation is narrow. Line charts don't carry the same zero-baseline obligation; see axis truncation ethics for when a non-zero axis is defensible.

2. Uneven axis intervals

Readers assume equal distances on an axis mean equal steps in value. Tick labels of 0, 10, 20, 50, 100, 500 at equal spacing, or a time axis that jumps from 2000 to 2010 and then moves one year per tick, bend the slope wherever the interval changes. A steady trend appears to speed up or slow down purely because of the spacing. Spreadsheet line charts that treat dates as text labels do this automatically when periods are missing.

Spot it: check that the gap between each pair of tick labels is the same size, both in values and in dates.

Fix: use a continuous, evenly spaced axis. If the data spans orders of magnitude, use a log scale, which is uneven by design and should be labeled as such.

3. Cherry-picked time window

Whoever picks the start and end dates picks the story. If a metric went from 100 to 140 and then to 120 over three years, a chart that starts at the peak shows a 14% decline, while one that starts at the beginning shows 20% growth. Every number on both charts is true.

Spot it: ask why the chart starts where it does. Be suspicious of a start date that lands on an unusually high or low point, or a window much shorter than the data that exists.

Fix: show the full available history, or at least one full cycle before the period in question, and highlight the window the claim is about instead of cropping to it.

4. Dual axes that fake a correlation

With two y-axes, each with its own minimum and maximum, the designer can make almost any two trending series overlap, diverge, or cross wherever they like. The apparent relationship comes from the axis ranges, not the data, and readers who see two lines moving together tend to infer that one drives the other.

Spot it: two y-axes with different scales, especially when both are in the same unit, or two lines that track each other suspiciously closely.

Fix: index both series to 100 at a common start date on one axis, or stack them as two panels that share the x-axis. Dual axis charts covers the few cases where two axes are justified.

5. Icons, circles, and areas scaled the wrong way

If a value doubles and the designer doubles both the height and width of an icon, or the radius of a circle, the area quadruples. Readers compare these marks by their area, so the graphic shows a 4× change for a value that only doubled, a lie factor of 3.0. Area is also a weak encoding to begin with: Cleveland and McGill's 1984 ranking of elementary perceptual tasks places area judgments below position and length.

Spot it: pictograms, bubble charts, or map circles where a value that is twice as large looks far more than twice as big.

Fix: scale area, not height or radius, in proportion to the value (radius proportional to the square root of the value), or replace the icons with bars. The lie factor page shows the math.

6. 3D perspective

3D bars and pies add a dimension the data doesn't have. Perspective draws near elements larger than far ones, so on a tilted pie the front slices look bigger than equal slices at the back, and on 3D columns it is hard to tell where a bar top meets the axis. Tufte's best-known lie-factor example, a 1978 New York Times chart of fuel-economy standards drawn as a road receding into the distance, exaggerated the real increase 14.8 times.

Spot it: depth, tilt, shadows, or perspective on a chart of one-dimensional values.

Fix: flatten it. A 2D bar chart shows the same numbers accurately; the full case against decoration is in avoiding chartjunk.

7. Inverted axis

By convention, larger values sit higher on a vertical axis. Flip the axis and rising data looks like it is falling. The standard example is a 2014 Reuters graphic of gun deaths in Florida. Its y-axis put 0 at the top and the area was shaded red from the top down, so the rise in firearm murders after the state's 2005 "Stand Your Ground" law moved down the page and read at a glance as a decline. The numbers were correct; only the orientation was unconventional.

Spot it: read the tick labels and confirm the values increase upward. Filled area charts hanging from the top edge are a common tell.

Fix: put the low values at the bottom. Where an inverted axis is the convention, such as rankings with No. 1 at the top or depth below the surface, label the direction on the axis.

8. Cumulative totals that can only rise

A running total of anything that can't be negative, such as sales, sign-ups, or downloads, can only go up or stay flat. A cumulative chart therefore always looks like growth, even when every period is worse than the one before. A slowdown shows up only as a gentle bend in the curve that few readers notice.

Spot it: "total," "cumulative," "all-time," or "to date" in the title or axis label, or a line that never dips.

Fix: plot the per-period values (daily, weekly, or monthly) instead of, or next to, the running total.

9. Raw counts instead of rates

Bigger populations produce bigger counts, so a chart or map of raw counts by city or state mostly shows where people live. City A with 1,200 incidents and 2,000,000 residents has a rate of 60 per 100,000. City B with 300 incidents and 250,000 residents has a rate of 120 per 100,000. City A has four times the incidents, but City B has twice the rate. The same logic applies to money over time: nominal dollars rise with inflation, so long time series should be adjusted before they are compared.

Spot it: comparisons between groups of very different sizes with no "per," "%," or "rate" in the axis label, or dollar amounts across decades labeled without "inflation-adjusted" or "real."

Fix: divide by the denominator that matches the question (per resident, per user, per mile driven) and name it in the axis label.

10. Missing uncertainty

Survey results, polls, forecasts, and A/B test lifts are estimates, and drawing them as crisp bars or a single line implies precision they don't have. Two survey results of 48% and 46%, each with a margin of error of ±3 points, look like a clear winner on a bar chart, but the gap is small enough that the true order could be reversed.

Spot it: sample-based numbers or forecasts with no error bars, no shaded band, and no stated margin of error.

Fix: add error bars or a confidence band and say what the interval is, such as a 95% confidence interval. Showing uncertainty in charts covers which form fits which kind of estimate.

What does a misleading graph look like in numbers?

Putting numbers on a distortion makes it hard to argue with. Here are two of the tricks above applied to small datasets, measured with Tufte's lie factor (effect shown in the graphic divided by effect in the data) or recomputed per period:

Truncated bar chart (y-axis starts at 90)
  Product A score = 95, Product B score = 100
  Effect in data    = (100 - 95) / 95 = 0.0526   -> +5.3%
  Bars as drawn     = 95 - 90 = 5 units, 100 - 90 = 10 units
  Effect in graphic = (10 - 5) / 5    = 1.00     -> +100%
  Lie factor        = 1.00 / 0.0526   = 19.0

Cumulative line (running total of monthly sign-ups)
  Monthly sign-ups  = 500, 400, 300, 200     -> down 60%, Jan to Apr
  Running total     = 500, 900, 1,200, 1,400 -> up 180%, Jan to Apr
  Headline the chart invites: "Sign-ups up 180% since January"

The truncated chart makes a 5.3% difference look like a doubling, 19 times the real effect. With the axis at zero, the bars are 95 and 100 units tall and the lie factor is 1.0. The cumulative chart rises every month while monthly sign-ups fell from 500 to 200. Neither chart contains a false number, which is why checking the data alone won't catch either one.

How do you spot a misleading graph quickly?

Read the title, axis labels, and notes before you look at the shape of the data, then run through the red flags below. Each red flag is visible from the axes, labels, and title alone.

Trick	Red flag	One-line fix
Truncated bar baseline	Bar axis starts above 0	Start bars at zero or switch to dots or a line
Uneven axis intervals	Equal gaps between unequal steps	Use an evenly spaced or labeled log axis
Cherry-picked window	Starts at a peak or trough	Show the full history and highlight the claim period
Dual axes	Two y-axes, lines that track too neatly	Index to 100 on one axis or use two panels
Area and icon scaling	2× the value drawn at 4× the area	Scale area, not radius or height
3D perspective	Depth, tilt, or shadows	Use a flat 2D chart
Inverted axis	Values increase downward	Put low values at the bottom, or label the direction
Cumulative total	"To date" or "all-time," a line that never dips	Show per-period values
Counts instead of rates	Unequal group sizes, no "per" or "%"	Divide by the right denominator
Missing uncertainty	Estimates with no error bars or bands	Add intervals and say what they are

What are the common mistakes when judging misleading graphs?

The opposite error is real too: flagging honest charts as misleading, or checking the wrong thing and missing the actual problem.

• Calling every non-zero axis a lie. The zero-baseline rule applies to bars and filled areas, where length or area is the encoding. A line chart of interest rates or body temperature can start well above zero if the axis is clearly labeled.
• Treating a log scale as a trick. A labeled log axis is uneven on purpose and is the right choice for growth rates or data that spans orders of magnitude. The problem is an unlabeled one.
• Assuming misleading means fake. Every trick on this list works with accurate data, so verifying the numbers alone won't catch any of them.
• Trusting the data labels. On a truncated bar chart, the printed values are correct. The bar lengths are what distort, and the shape is what readers take away.
• Picking the wrong denominator. Converting counts to rates only helps when the denominator matches the question. Traffic deaths per resident and per mile driven answer different questions and can rank places differently.
• Reading causation into lines that move together. Even on a single, honest axis, two trends rising at the same time don't show that one causes the other.