Common mistake 1 with Stacked bar

Using it to compare a middle segment across bars. That is what the reader will try to do, and it is the one thing the chart does badly.

Common mistake 2 with Stacked bar

Too many segments, or segments ordered differently from bar to bar.

Common mistake 3 with Stacked bar

Negative values, which produce overlapping segments and a total that is not the sum of the parts.

Common mistake 4 with Stacked bar

Using it when the question is about the middle segments. If the argument is about a component that is neither first nor the total, small multiples answer it and this chart does not.

EconStatsCraft › Stacked bar

Stacked bar

Each bar is a whole and each segment a part, so it answers "what is this made of" for several categories at once. Only the bottom segment sits on a flat baseline, so it compares totals well and middle segments badly.

What it needs

Categories
Components - Two or more that stack into a total.

The intuition

A stacked bar shows composition: each bar totals to something, and the segments are its parts. It does one job well - comparing TOTALS across bars - and one job badly. Only the bottom segment starts from a common baseline, so every segment above it floats on a moving base and cannot be compared across bars by eye. That single geometric fact explains most of what goes wrong with this chart.

When to use it

When a reader needs both "how big is each group" and "what is each group made of". If only the composition matters, normalise the bars to 100%.

Where the estimator comes from

Segment j of bar k occupies the interval \left[\sum_{i<j}y_{ik},\ \sum_{i\le j}y_{ik}\right]. Its LENGTH encodes y_{jk}, but its POSITION depends on every segment below it. Comparing lengths that begin at different places is the "position along non-aligned scales" task, which Cleveland and McGill ranked well below position on a common scale - the bottom segment gets the accurate judgement and the rest do not. A 100% stacked bar normalises each column to \sum_i y_{ik}=1, which makes shares comparable and discards the totals entirely.

What it assumes

The segments are parts of the bar's whole, and all non-negative. Mixing signs breaks the arithmetic the chart implies.
There are few enough segments to tell apart - beyond about five, colours stop being distinguishable.
The segment order is the same in every bar, or the chart is unreadable. Reordering segments per bar to put the largest at the bottom destroys the only comparison the format supports.
The whole is meaningful for every bar. Stacking counts across categories with very different totals invites a comparison of segment heights that the differing totals make invalid.

Statistical properties

Only the bottom segment and the overall total are read accurately. Everything in between is judged by comparing floating lengths, which people do poorly and inconsistently.
Put the series you most want compared at the bottom, and the most stable one there if no series is privileged - a volatile bottom segment makes every layer above it jump for reasons that have nothing to do with those layers.
A 100% stacked bar answers a different question. It shows composition and hides size, so a category with 3 observations and one with 3,000 look equally substantial.
Beyond about five segments the colours become unreadable and the middle bands too thin to see. A grouped bar or small multiples handle more.
Negative values break the stack: segments overlap or reverse direction and the total no longer corresponds to the bar length.
If the reader's question is "how did this one component change", the stacked bar is the wrong chart entirely. Plot that component on its own.

How to read the output

Totals compare well, and so does the bottom segment. Everything in between is measured from a moving baseline.
Put the segment the argument is about at the bottom, where it can be read against a flat line.
For shares rather than levels, compute the percentages first: a stacked bar of levels answers a different question from a stacked bar of shares.
Only the total and the bottom segment have a fixed baseline. Every other segment floats on the ones below it, so its apparent change mixes its own movement with theirs.

A worked reading

Four quarters of revenue split into five product lines. The chart answers "which quarter was biggest" instantly and reliably, because that is a total and totals are the full bar length from zero. Ask it "did product C grow" and it cannot help: product C sits third from the bottom, so its segment starts at a different height in every quarter, and the eye is being asked to compare four floating lengths. The honest response to that question is a second chart with product C alone against a zero baseline.

Symptom → diagnosis → what to do

If you see this	It means	Do this
Middle segments hard to compare	They float on a moving baseline	Inherent. Move the key series to the bottom or facet
A component seeming to jump	The segment below it moved	Plot that component alone to check
Bars all the same height	It is a 100% stack	Totals have been normalised away; show them separately
Eight or more segments	Beyond colour capacity	Group the tail, or use small multiples
Negative values in the stack	The geometry breaks	Use a grouped bar or a diverging chart
Reader comparing the top segment across bars	The least reliable comparison on the chart	Give it its own panel

Common mistakes

Using it to compare a middle segment across bars. That is what the reader will try to do, and it is the one thing the chart does badly.
Too many segments, or segments ordered differently from bar to bar.
Negative values, which produce overlapping segments and a total that is not the sum of the parts.
Using it when the question is about the middle segments. If the argument is about a component that is neither first nor the total, small multiples answer it and this chart does not.

Where to read this properly

Cleveland and McGill (1984), "Graphical Perception", JASA 79(387) - the accuracy ranking.
Few, Show Me the Numbers (2012), ch. 6.
Heer and Bostock (2010), "Crowdsourcing Graphical Perception", CHI - a replication of Cleveland and McGill.

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.

Stacked bar

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.