Common mistake 1 with Bar chart

Truncating the value axis. This is the single most common way a bar chart misleads, and it is invisible unless the reader checks the axis.

Common mistake 2 with Bar chart

Bars for a time series with many periods when a line would show the shape better. Bars suit a handful of periods; lines suit hundreds.

Common mistake 3 with Bar chart

Averaging where you meant to total, or the reverse. Mean wage by region and total wages by region tell different stories and the chart looks identical.

Common mistake 4 with Bar chart

Plotting a mean per category without showing how many observations each rests on. A category with three observations gets the same visual weight as one with three thousand.

EconStatsCraft › Bar chart

Bar chart

The chart for comparing quantities across categories. Repeated rows are combined for you, because a bar sits on a category rather than at a point on a line - and a bar chart of raw rows is one bar per row.

What it needs

Categories
What to measure

The intuition

A bar chart is the most reliable chart there is, because it asks the eye to compare lengths from a common baseline - the one perceptual judgement people make accurately. That reliability is entirely conditional on the baseline being zero. Cut the axis and the bar lengths no longer stand in proportion to the values, and the chart becomes precisely as misleading as the amount you cut.

When to use it

Comparing a quantity across categories: regions, sectors, countries, years. Horizontal bars when the labels are long, which they usually are for place names.

Where the estimator comes from

Each category k is drawn as a rectangle of length proportional to y_k from a baseline b. The reader compares LENGTHS, so the perceived ratio between two bars is (y_1-b)/(y_2-b) rather than y_1/y_2. Those agree only when b=0. With y_1=102, y_2=100 and a baseline at 99, the true ratio is 1.02 and the perceived one is 3.0 - the chart is not exaggerating by a little, it is showing a different number.

What it assumes

One bar means one category, so rows are aggregated when there are several per category. The chart says which function it used; if the aggregation is the point, read Summary by group instead.
The baseline is zero. Bars encode length, so a truncated axis multiplies every visual difference by an arbitrary factor.
The categories are exhaustive, or the omission is stated. A bar chart of the top ten implies the rest are small, which may not be true.
The bars are independent quantities rather than parts of a whole. If they sum to something meaningful, the reader will try to read shares off a chart not designed for it.

Statistical properties

Bars must start at zero. This is not a stylistic preference; the encoding is length, and length from a non-zero base is not proportional to the value.
Cleveland and McGill ranked visual tasks by accuracy: position along a common scale is the most accurate, then length, then angle, then area, then colour saturation. A bar chart uses the first two, which is why it beats a pie chart at the same job.
Sort by value unless the categories have a natural order. Alphabetical ordering is almost never the reader's question, and an unsorted bar chart makes rank comparisons needlessly hard.
Horizontal bars are better for long category labels, and no worse for anything else. Rotated axis text is a reliable sign the chart should have been horizontal.
Do not use bars for a mean without showing spread. A bar of height 50 from data ranging 0 to 100 and one from data all near 50 look identical - the "bar-bar plot" problem widely criticised in the sciences.
Too many categories turns a bar chart into a comb. Above roughly twenty-five, aggregate the tail into "other" or use a different display.

How to read the output

Sort by value unless the categories have their own order. An alphabetical bar chart makes the reader do the ranking themselves.
Grouped bars compare within a category well and across categories badly. If the comparison is across, make that the x axis instead.
A lollipop is the same information with less ink, which helps when there are many categories and the bars become a solid block.
The order is an argument. Sorting by value makes the ranking the message; keeping a natural order makes the shape the message, and the two support different claims.

A worked reading

Two categories at 102 and 100. Drawn from zero the bars are practically the same height, which is the truth: a 2% difference. Drawn from a baseline of 99 one bar is three times the other, and every reader will take away "three times as much". The underlying numbers never changed and neither did the chart type. That is why the zero baseline is treated as a rule for bars while remaining a judgement call for lines - the line encodes position, the bar encodes length, and only one of those breaks.

Symptom → diagnosis → what to do

If you see this	It means	Do this
Differences looking dramatic	Check the baseline	A non-zero base breaks the length encoding
Rotated 45-degree labels	Categories too long for vertical bars	Make it horizontal
Bars in alphabetical order	Rarely the reader's question	Sort by value
Bars showing group means	Spread is hidden	Add error bars, or use a box plot
Forty categories	Beyond comfortable comparison	Take the top n and group the rest
A negative value	Bars cross the baseline	Fine, but keep zero visible and labelled

Common mistakes

Truncating the value axis. This is the single most common way a bar chart misleads, and it is invisible unless the reader checks the axis.
Bars for a time series with many periods when a line would show the shape better. Bars suit a handful of periods; lines suit hundreds.
Averaging where you meant to total, or the reverse. Mean wage by region and total wages by region tell different stories and the chart looks identical.
Plotting a mean per category without showing how many observations each rests on. A category with three observations gets the same visual weight as one with three thousand.

Where to read this properly

Cleveland and McGill (1984), "Graphical Perception: Theory, Experimentation, and Application to the Development of Graphical Methods", JASA 79(387).
Weissgerber et al. (2015), "Beyond Bar and Line Graphs", PLOS Biology 13(4) - on hiding distributions.
Cairo, How Charts Lie (2019), ch. 3.

All analyses

Data & variables

Each example is simulated from a known model, so you can check whether the estimator recovers the truth - and nothing here is anyone's private data.

Filter variables

Drag a variable into a slot on the right - or just click it.

Choose an example dataset above to begin.

Choose a transformation - it opens in the data editor, with a live preview.

Apply to

Multiply by

Columns ctrl-click for several

Group by ctrl-click for several

Within groups

Statistic

Bands

How many

Cut points

Keep rows where

Value

and

Fill with

Value

Also mark which rows were filled

Rules

Base period = 100

Which part

Percent at each tail

Threshold

Survey weight optional

Direction

Names column

Values column

Largest first

Delete them, not just count them

Read this column as

Tidy the text

New name optional

EconStatsCraft rigorous, plain-spoken econometrics

Econometrics that explains itself.

Regression, panel data, time series and the whole battery of tests. Every result is followed by a verdict you can defend.

Load a dataset to see it here.

Bar chart

What it needs

The intuition

When to use it

Where the estimator comes from

What it assumes

Statistical properties

How to read the output

A worked reading

Symptom → diagnosis → what to do

Common mistakes

Where to read this properly

See also

Related Charts

All analyses

Charts

Maps

Explore

Regression

Time series - one series

Time series - several series

Panel data

Multivariate statistics

Marketing research

Causal

Finance & risk

Actuarial

Tests & breaks

Machine learning

Econometrics that explains itself.