Data visualization

Why presentation decides the value

Data visualization puts numbers into a form that can be grasped quickly. A table with 4,000 rows holds the same information as a bar chart with eight values, but only one of the two gets used in a meeting. The purpose of visualization is therefore not beauty but reduction to the message.

The most important rule follows from that: question first, presentation second. Starting with the chart type produces graphics that are complete and still answer nothing.

Which chart fits which question

• Comparison across groups: Bars. Length is the easiest property to distinguish.
• Development over time: A line. The trajectory is the message, not the single value.
• Composition of a whole: Stacked bars. Pie charts only work with two or three shares.
• Relationship between two values: A scatter plot. Only this reveals whether a relationship exists or it is coincidence.
• Distribution: Histogram or box plot. Averages hide the actual information precisely here.
• One important value: A large number with a comparison to the previous month. A chart with a single point wastes space.

Where charts mislead

Three manipulations are common, sometimes unintended. First, the truncated axis: if a bar axis does not start at zero, a five percent difference looks like a jump. With lines a crop is acceptable, with bars it is not. Second, wrong proportions: when area or volume scale twice, a doubling looks like a fourfold increase. Third, a missing base: growth from 200 to 300 orders cannot be judged without the base, because it stays unclear whether sessions moved as well.

How dashboards stay useful

A dashboard rarely fails on technology, it fails on volume. Showing 40 metrics achieves that nobody looks at any of them. A hierarchy works better: three to five values at the top watched daily, detailed views below for the monthly review. Each metric also needs a decision attached. If it is unclear what would be done about a deviation, the value does not belong on the dashboard.

A fixed definition per metric, documented in one place, matters too. The most common reason for arguments about numbers is not data quality but two different calculations of the same metric in two reports. Referencing conversion rate once to sessions and once to users produces two truths and destroys trust in both.