A chart is an argument, so check what it compares

The idea

A table lets the reader look up a number. A chart does something stronger: it makes a comparison for the reader, instantly, before any number is read. The eye compares positions, lengths and slopes, and it trusts what it sees more than the labels beside it.

So choosing a chart is choosing the comparison you are asking the reader to make. A chart whose visual comparison matches the real comparison is honest. One whose picture says something the data does not is misleading, even if every number on it is correct.

Why it exists

A dashboard or a slide is usually read in seconds by someone who will not open the data. Their decision comes from the picture. That makes the chart, not the table under it, the thing that actually communicates. If a team's marketing deck says "desktop converts several times better than mobile" and the supporting chart agrees, the budget moves, even if the true difference is 60%.

The same logic runs the other way. A true, important gap deserves a chart that shows it at its real size. A 60% difference in conversion is already a reason to look into mobile, and an honest chart makes that case without needing to shout.

An analyst who only checks the numbers is checking the wrong thing. The question to ask of any chart is: what will a reader conclude by looking, and is that what the data supports?

How it works under the hood

Bars must start at zero

A bar encodes a value as a length. Length is only proportional to the value if the bar begins at zero. Cut the axis and you remove a fixed amount from every bar, which changes the ratios between them.

Harbor & Pine's desktop conversion is 3.87% and mobile is 2.41%. Draw the axis from 2%, and the bars are 1.87 and 0.41 points tall.

The same two numbers, drawn from zero, look like what they are.

Lines are different. A line encodes position and slope, not length, so a line chart does not need a zero baseline, and starting at zero can flatten a real trend into nothing. Use zero for bars, whatever fits the data for lines, and label the axis either way.

Many series: small multiples

Six lines on one chart force the reader to trace each one through a tangle. When the question is "how does each channel behave?", six small charts with the same axes beat one crowded chart, because the eye compares shapes side by side. This is the idea of small multiples. Fix the scales the same across panels, or the comparison breaks again.

Noise: smoothing with a rolling average

Daily revenue jumps with weekdays, weekends and promotions, and the wiggle can hide the trend. A 7-day rolling average, daily.rolling(7).mean() in pandas, replaces each day by the mean of that day and the six before it, so every point covers one of each weekday and the weekly cycle cancels out. Show the raw data faintly behind it, and say a rolling average was applied. Smoothing never adds information. It trades detail for readability, and it lags the real series by a few days.

Parts of a whole

A pie chart asks the reader to compare angles and areas, which people do badly. When the slices are close, the difference is invisible. A sorted bar chart of the shares, or a table of percentages, makes the ranking obvious. Keep a pie for two or three parts where the point is a clear majority.

Review any chart with four questions

Before a chart leaves your hands, ask: what comparison does the eye make first? Does the axis start where the encoding needs it to? Could a reader come away with a conclusion the data does not support? Would the answer survive if the chart were shown to the people it is about? A chart that fails any one of them gets fixed, even when the numbers behind it were right. Fixing the chart often starts with fixing what it plots: the metric, the period or the baseline.

A dashboard answers a decision

A dashboard with every metric the data allows answers nothing. Start from the decision the reader makes, such as whether to move budget between channels, and show exactly the comparisons that decision needs, in the order they need them.

Common misconceptions

"The chart uses the real numbers, so it cannot be misleading"

Accurate numbers can still give a false picture. The cut-off axis above uses exact conversion rates and makes a 60% gap look like 360%. Honesty is about the comparison the picture invites, not only the data behind it. Read the axis first, then the bars.

"A line chart with the y axis starting at zero is always the honest choice"

For bars, yes. For lines, a zero baseline can flatten a meaningful change: a conversion rate that moves from 2.8% to 3.4% looks like nothing on a 0 to 100 axis. Fit the axis to the data, label it, and keep the same range across charts that will be compared.

"Smoothing a chart is cheating"

Smoothing is fine when it is disclosed and the raw data is still available. A 7-day average of daily revenue is a standard tool because it removes the weekly rhythm and leaves the trend. It is misleading only when presented as the raw data, or when the window is so wide it erases a real event.