Comparison, trend, distribution, composition, and relation each map to different charts

What it is

Task-conditioned chart selection uses comparison, trend, distribution, composition, and relation as entry points because different questions call for different visual operations. Bars often support categorical magnitude comparison, lines change over an ordered domain, histograms or raw points distribution, stacks part-to-whole structure, and scatterplots relations between paired variables. These are candidate mappings, not prescriptions assigning exactly one chart to each task family.

Why it happens

Position, length, connection, binning, and spatial arrangement make some judgments direct while forcing conversion for others. Fit also depends on data semantics: trend needs a meaningful order, composition a well-defined total with a consistent denominator, and relation correctly paired observations. Identical field types do not imply identical tasks. Even “comparison” may concern difference, ratio, rank, or uncertainty, so a broad task label cannot replace the actual judgment.

Studying it

Saket, Endert, and Demiralp's task-by-chart experiments found that basic-chart effectiveness varied significantly by task. Their study covered three attribute-type combinations and two datasets, but that coverage should not be restated as proof of a general attribute-type main effect. Their ten low-level tasks came from Amar, Eagan, and Stasko; these five question families are chart-selection entry points, not a rewrite of Amar's ten-task taxonomy. Replications should compare candidates with the same data and information, separately measuring value retrieval, extrema, ordering, trend, correlation, and distribution through accuracy, time, and confidence. Vary data scale, effect strength, and reader experience, then check laboratory tasks against real decisions instead of transferring results from brief, small-data trials wholesale.

Where it stops holding

The five families are practical headings, not an exhaustive or mutually exclusive taxonomy; anomaly detection, clustering, spatial location, network paths, and causal judgment add other needs. A validated view may support a compound task, or coordinated views may divide it—one chart per label is not a rule. Tables, direct values, or prose may beat charts for exact lookup, a few items, or nonvisual access. A chart can support reading evidence but cannot turn association into causation.

Applying it

• Rewrite the question as what judgment on which objects at what precision, then compare two or three chart candidates rather than jumping from a task label to one type.
• Verify semantic prerequisites: genuine order, consistent whole, valid record pairing, and the role of missingness and uncertainty.
• Test candidates on representative data and compound tasks. Record primary and secondary tasks and failure modes instead of reviewing appearance alone.
• Provide task-matched structured data, summaries, and navigation for keyboard and screen-reader use; validate visual and nonvisual representations with the same user question.

Related

• Same group: U2.01.2 Define the question before choosing the chart · U2.01.3 The same data supports several tasks; draw a chart for each
• Adjacent: U2.17.1 A mismatched chart forces readers to compute the answer mentally · U1.08.5 The same field can take different types under different analytical tasks
• Search terms: task-based visualization · chart effectiveness · analytic task