Guide section 15 of 31

Chart Selection Guide

User questionData relationshipRecommended visualNotes / alternatives
Trend over timeOrdered timeLine chartColumn for discrete periods; area only when magnitude emphasis matters.
Compare categoriesCategorical comparisonClustered bar/columnBar is strong for long labels/ranking.
Part-to-wholeComposition100% stacked bar/column; stacked bar/columnPie/donut only for very few simple shares.
DistributionSpread/outliersScatter; binned column where appropriateCustom box plot only if distribution statistics are essential and native options are insufficient.
CorrelationRelationship between measuresScatter/bubbleAvoid implying causation.
RankingOrdered category comparisonSorted barRibbon only when rank changes over time are the question.
VarianceActual vs expected/differenceBar/column + variance; Waterfall for contributionKeep baseline/reference explicit.
Target performanceActual vs targetCard + linear bar/reference; KPIGauge only when the radial metaphor adds real value.
Contribution to changePositive/negative driversWaterfallUse ordered contributors and a clear start/end context.
Root causeDrivers by hierarchy/dimensionDecomposition tree + waterfall/barFixed hierarchy when path is known; decomposition tree for genuine flexible exploration.
Geographic distributionSpatial relationshipAzure MapsAvoid legacy Bing map visuals for new standards.
Process stagesStage volume/conversionFunnel or ordered barsFunnel only when the stages truly represent a process.
Hierarchy/precise valuesExact values across levelsMatrixPrefer table/matrix over chart when lookup and accuracy dominate.

Charts to use cautiously

  • Pie / donut: difficult precise angle/area comparison; keep to a few obvious categories.
  • Gauge: space-inefficient for dense KPI monitoring.
  • Treemap: useful for hierarchical area overview, weaker for precise ranking.
  • Funnel: misleading if stages are not truly sequential.
  • Stacked area: hard to compare non-baseline series.
  • Dual-axis combo charts: only when measures are genuinely related and scales cannot mislead.
  • Radar/spider, violin, Sankey, Gantt, box plot and similar custom visuals: use only for a real analytical need, governance approval and performance/accessibility validation.