Bar Graphs

Precise definition

A bar graph displays a categorical variable by assigning one separated bar to each category, with height or length proportional to frequency, relative frequency, percentage, or another explicitly labelled summary. Categories are distinct rather than intervals on a continuous number line.

Notation and mathematical language

Let $n_j$ be the count in category $j$, $n=\sum n_j$, and relative frequency $p_j=n_j/n$. Axes must name the category and measured quantity, including units. Bars may be reordered for emphasis unless the categories have a natural ordinal or temporal order.

Conceptual picture

Bar length encodes magnitude along a common baseline, which people compare more accurately than area or angle. Gaps emphasize that categories are separate. Grouped bars compare another categorical variable; stacked bars emphasize totals and composition but make interior segments harder to compare.

Conditions and key results

Categories should be mutually exclusive and collectively address the intended population, or overlaps and omissions must be disclosed. A truncated numerical axis can exaggerate differences, especially for bars whose length is the encoding; if used, the break needs prominent justification.

A reliable strategy

1. Define the observational unit, categorical variable, population or sample, and counting rule.
2. Create a frequency table and verify the total before plotting.
3. Choose count, proportion, or percentage; label axes, scale, units, sample size, and source.
4. Read comparisons from the displayed quantities and separate descriptive association from causal explanation.

Fully worked example

Interpretation and application

Bar graphs communicate election categories, product choices, classroom responses, and group summaries. Generalization depends on sampling design and nonresponse, not on how polished the graphic looks. Causal claims require a suitable experiment or defensible causal design.

Common mistakes

Verification and reasonableness

• Add category counts and compare with the stated sample size.
• Convert proportions back to counts when possible.
• Audit labels, zero baseline, missing categories, and whether conclusions stay within the sampled population.

Practice

1. What percentage is 18 of 60?
2. Why are bars usually separated?
3. Does a bar graph establish causation?

Further deduction

Uncertainty can accompany bars. If each height estimates a population proportion, confidence intervals may be displayed, but they describe sampling uncertainty only under the interval's conditions. They do not account automatically for coverage error, biased questions, or nonresponse. Error bars must be defined because standard deviations, standard errors, and confidence intervals answer different questions.

For paired group comparisons, use a common vertical scale and category order. Separate panels with automatically rescaled axes can make smaller values look equal to larger ones. Direct labels and denominators reduce lookup error. Three-dimensional bars distort length and area without adding information, so a flat display is usually more truthful and readable.

Because bar length encodes magnitude, a quantitative vertical axis should usually begin at zero. A truncated baseline exaggerates relative differences; if a specialized exception is justified, show the break conspicuously and state exact values so readers do not infer proportions from cropped lengths.

Related topics

• Frequency Tables
• Histograms
• Data and Variables