Misleading Charts Examples and How to Spot Them

What makes a chart misleading if every number on it is correct? Usually it is the drawing, not the data. A misleading chart takes true figures and encodes them so your eye reads a different size, order or relationship than the numbers support. The bar is too tall, the dates are shuffled, or two unrelated lines are scaled until they seem to move together.

This page is for anyone who builds or reads dashboards, slide decks and news graphics: analysts, marketers, product managers, researchers. It walks through the specific tricks, each tied to a real example you can look up, and gives you a check for each one.

The thing most people get wrong is the rule itself. "The axis must start at zero" is true for bar charts and often false for line charts. Treating it as universal makes you miss some deceptive charts and wrongly accuse honest ones.

Every chart turns a number into something visual: a length, a position, an angle, an order. The chart misleads when that visual property stops being proportional to the number, or when it implies a relationship the data does not have.

The clearest case is the bar chart. You judge a bar by its length measured from the baseline, the line it grows out of. As Nathan Yau of FlowingData put it, truncating the axis is something "you don't do with bar charts, because the visual cue is the length of a bar relative to the origin." Cut off the bottom of the bars and the remaining stubs no longer compare in the ratio the numbers do.

A line chart works differently. You read it by the position of points and the slope between them, so what matters is change over time rather than the total height above zero. That is why a bar chart perception paper by Correll, Bertini and Franconeri argues that bar charts need a zero baseline while line charts showing a trend can sometimes justify a non-zero one, depending on context.

1. The truncated bar axis: Fox News, 2012

In 2012 Fox News aired a bar chart titled "Top Tax Rate." It compared the Bush-era top income tax rate of 35% with the 39.6% rate that would return if the Bush tax cuts expired, according to FlowingData. The y-axis started at 34%, not zero.

The real difference was 4.6 percentage points, a ratio of about 1.13 to 1. On screen the second bar looked about six times taller than the first, by the arXiv paper's count; Tableau's account says "more than five times." The arithmetic below shows why both descriptions are close. Media Matters catalogued this chart, and others like it, in a piece the paper cites called "A History of Dishonest Fox Charts."

2. Out-of-order categories: Georgia's COVID-19 chart, May 2020

On Sunday, May 10, 2020, the Georgia Department of Public Health posted a bar chart of new confirmed COVID-19 cases in the state's five hardest-hit counties, Route Fifty reported. The dates were not in calendar order. Bars were sorted by case count, so April 28 appeared before April 27. The counties also swapped left-to-right positions each day, with the highest always on the left.

Sorted that way, any set of numbers slopes downward. The chart showed a steady decline that the real, more volatile data did not. State Rep. Scott Holcomb raised it publicly and wrote to the governor's office. The governor's spokeswoman, Candace Broce, first said the sorting was meant "to show descending values to more easily demonstrate peak values and counties on those dates." A DPH spokeswoman told the Atlanta Journal-Constitution it was a sorting error. By May 18, 2020, the chart had been redrawn in chronological order with each county in a fixed position, and Broce said: "Our mission failed. We apologize. It is fixed."

What makes this case instructive is that sorting bars by value is normally good practice. Andrew Gelman's Columbia statistics blog used it on May 18, 2020 to make exactly that point: ranking by size helps when categories have no natural order, and destroys the trend when they do, as dates always do.

3. Recent data drawn as if it were final

The Georgia chart had a second problem. Case counts there could take about 14 days to be finalized because of testing and backdating lags, so the most recent days always looked low. Researchers at Georgia Tech argued in a May 15, 2020 Slate article that officials were confusing reporting lags with real declines. Part of the fix was shading recent data points as "preliminary." Any dashboard fed by data that fills in late, from sales bookings to sign-ups attributed after a trial, has the same trap at its right-hand edge.

4. Two axes that manufacture a relationship

A dual-axis chart puts two series on one plot, each with its own vertical scale. Because each scale can be stretched independently, the designer controls where the lines sit relative to each other. "The scales of dual axis charts are arbitrary and can therefore (deliberately) mislead readers," writes Lisa Charlotte Muth of Datawrapper.

Her example: plotted on separate non-zero axes, German and global GDP from 2004 appear to rise at similar rates. Put on a shared zero baseline, global GDP rose about 80% through 2014 and German GDP about 40%. A study she cites, Isenberg et al. (2011), found superimposed charts of this kind "performed poorly both in terms of accuracy and time"; it had 15 participants, so treat it as suggestive rather than settled.

The Federal Reserve Bank of St. Louis posted a military spending chart, using data from the Stockholm International Peace Research Institute, that gave the United States its own y-axis. Economist Matthew Yglesias called it, in a January 22, 2023 post on X, "One of the most cursed dual-axis charts you will see in your lifetime," as Flourish documents. Flourish's redraw on a single $0 to $1,000 billion scale showed a very different relative gap. Noah Smith flags a Financial Times dual-axis chart that made the Chinese renminbi's share of global reserves, about 3%, look close to the dollar's, about 60%.

The joke version is Tyler Vigen's chart of pool drownings against Nicolas Cage films, which Smith cites: two unrelated series, two truncated axes, and lines that track each other perfectly.

5. A baseline that hides a real change

Truncation exaggerates. Its opposite minimizes. In 2015 National Review tweeted a chart from the Power Line blog captioned "The only #climatechange chart you need to see," with a temperature axis starting at 0°F. The Correll paper notes that this compressed the warming trend until it was illegible. Zero was there, and the chart was still misleading, because zero degrees Fahrenheit is not a meaningful floor for a change of a few degrees.

6. Bars not drawn to scale

Sometimes there is no axis trick at all; the bars are simply the wrong heights. Kaiser Fung of Junk Charts checked a chart of the "best acquisitions of all time." Facebook-Instagram, at $175.5 billion, should have a bar about 10 times Disney-Marvel's $17.5 billion. It was drawn at under 9 times. His test is the one to copy: divide the two values, then measure the two bars, and compare.

7. A window or metric chosen for the story

The time range you show and the thing you measure can carry the whole claim. In October 2026 a chart titled "Daily Activity over 3 Months" spread after Ukrainian drone strikes damaged Russian Yandex data centers between October 7 and 11. It claimed AfD activity on X fell by up to 56%, implying more than 99% of pro-party Russian bots had vanished. Austrian fact-checkers Mimikama found the underlying drop, Cloudflare traffic falling to 0.4% of its two-week peak, measured web requests, not political accounts. A separate tally of 611 Polish-language posts on October 11 and 12 had no pre-attack baseline, so no decline could be calculated from it at all.

The Christy and Spencer climate chart, debunked in coverage dated January 20, 2026, set 73 warming models against satellite observations that looked flat. Critics pointed to a 2017 Journal of Climate study in which 5 of 6 corrected satellite datasets showed warming. The chart's form was ordinary; the dataset choice did the work.

What about 3D charts?

Tilting a pie or bar chart into 3D makes the slices or bars nearest the viewer look larger than those further back, and makes heights hard to read against the axis. None of the sources on this page documents a specific dated 3D example, so this page does not offer one. The check is the same as for any chart: if you cannot read a value off it, it is decoration.

Run these steps on a chart you are about to publish or one someone has put in front of you. Each targets one of the tricks above. Most take seconds once you know what to look for.

Calling every non-zero axis dishonest. This is the overcorrection. Kaiser Fung reviewed a forest-area chart covering 1990 to 2023 and did not fault its non-zero y-axis, because the values are far from zero and change slowly; starting at zero would have drawn a flat line. Noah Smith's advice cuts both ways: "Don't just assume it starts at zero!" Recognise it by asking whether the chart is a bar or a line. Fix it by applying the zero rule to bars only.

Assuming a zero baseline proves honesty. The 2015 National Review climate chart started at 0°F and still hid the change it was supposedly showing. Recognise it when a line looks flat but the subject is one where small moves matter. Fix it by choosing a range that fits the size of change that is meaningful for that measure.

Applying a good habit to the wrong data. Georgia's spokeswoman defended sorting by value as a way to highlight peaks, and in most bar charts that instinct is right. Recognise it when the categories are time periods. Fix it by locking time-ordered categories into sequence before any sort is applied, and by fixing the position of each series.

Wasting the frame. The same forest-area tracker started its time axis at 1960 although the data begins in 1990, leaving about half the plot empty, with gridlines at awkward values such as 33.01. Nothing is falsified, but the data is squeezed into half the space and becomes harder to read. Fix it by fitting the axis to the data you have.

Mistaking a deliberate cut for cherry-picking. Not every trimmed window is a trick. In the same Junk Charts post, a reader's alternative chart covered 2000 to 2023, dropping the first ten years, and Fung judged it far more informative than the original. "Choices frequently involve trade-offs," he wrote. The test is whether the choice is visible and whether the conclusion would flip with the full range.

Small changes with large consequences. Tableau's June 1, 2020 post on truncation uses human body temperature: a rise from 37°C to 38°C is medically significant but nearly invisible on an axis starting at zero. A line chart of temperature, conversion rate or interest rates can legitimately zoom in, provided the axis is labeled clearly.

Values far from zero that drift slowly. Global forest area, population, or a company's total headcount fall here. A zero baseline turns any trend into a flat line, which is its own distortion.

Comparing amounts. Whenever the reader is meant to compare totals, use bars from zero, even if a line chart of the same data could start elsewhere. The chart type follows the question, and the baseline follows the chart type.

Dual axes. None of the sources here describes a case where a dual-axis chart was the best option; Datawrapper's advice is to index the series or split them into side-by-side panels. If you keep two axes, state both scales prominently and do not let the crossing or tracking of lines carry a claim.

Charts of probability. Some bars and gauges show the chance of an outcome rather than an amount, and they get misread as amounts. A forecast giving one party an 85 percent chance is a different thing from an 85 percent vote share, whatever the baseline.

What these figures cannot tell you is how common each trick is. Tableau's post notes it does not have a verified figure for how often truncated axes appear in bad visualizations, and none of the other sources counts frequency either. What would settle it is a systematic audit of published charts by type, which these sources do not report.

The cost of a misleading chart is also not just reputational. Georgia's chart sat in public view for about a week during a period when people were judging whether case counts were falling, and the Atlanta Journal-Constitution reportedly called it the third data "error" from state officials in as many weeks.