Bar Chart

A bar chart represents categorical data using rectangular bars whose length or height is proportional to the value each category represents — revenue by product line, headcount by department, survey responses by answer option. It's one of the most universally used and most reliably interpreted chart types specifically because it encodes quantity through position and length along a common axis, the visual encoding that research by statisticians William Cleveland and Robert McGill in the 1980s found humans judge more accurately than nearly any other common visual encoding, including angle, area, and color.

Bar charts come in several standard structural variants, each suited to a different comparison task: a simple bar chart compares a single value across categories; a grouped (or clustered) bar chart places bars for multiple series side by side within each category, enabling direct comparison across both the category and the series dimension simultaneously; and a stacked bar chart stacks multiple series' values on top of each other within each category bar, showing both each category's total and the composition of that total at once. The tradeoff between grouped and stacked bars is a real, consequential design decision rather than a stylistic preference: grouped bars make it easy to compare one specific series across categories (because each series' bars sit on a shared baseline, which Cleveland and McGill's research identified as the most accurately judged comparison), while stacked bars make the category totals easy to read but make comparing any individual series across categories considerably harder, since only the bottom-most segment in a stack shares a common baseline — every segment above it starts from a different, varying point.

A truncated y-axis — one that doesn't start at zero — is the single most common and most consequential way a bar chart misleads, because bar length is the entire visual signal a bar chart uses to represent quantity, and starting the axis above zero visually exaggerates the relative difference between bars far beyond what the underlying numbers actually show. A bar chart comparing values of 98 and 100 looks like a near-total difference if the y-axis starts at 95 rather than 0, even though the real difference is only about 2%; this specific manipulation (or honest mistake) is widely enough recognized that "starts the axis at zero" has become one of the most frequently cited baseline rules in data visualization literacy, distinguishing bar charts in this respect from line charts, where a non-zero baseline is often more defensible since a line chart's primary signal is the trend's shape and direction, not an absolute bar length a viewer is implicitly comparing to zero.

A horizontal bar chart — rotating the standard vertical bar chart ninety degrees — is commonly used specifically when category labels are long (full product names, lengthy survey response text) and would overlap or require awkward rotation if forced along a horizontal axis, a practical readability fix rather than a different underlying chart type, and ordering the bars by value (rather than alphabetically or by an arbitrary default order) is a simple, frequently underused change that makes the largest and smallest categories immediately visible without requiring a viewer to scan the entire chart.

Read a bar chart as a chart type whose reliability comes directly from its reliance on position and length, the two visual encodings humans judge most accurately according to Cleveland and McGill's research — the choice between grouped and stacked variants is a real tradeoff between comparing individual series and reading category totals, not an interchangeable stylistic option, and the truncated-axis distortion remains the single most important and most commonly cited pitfall to check for, precisely because bar length is the entire signal a bar chart relies on to convey quantity in the first place.