Examples
These are worked patterns, not fixed prose. Adapt vocabulary, structure, and recommendation order to the user's data and audience.
Truncated bar chart
Pattern: start with the integrity issue, then offer a chart form that matches the comparison.
Direction: If bars make a modest difference look large because length is read from a hidden zero point, rebuild with a zero baseline for absolute comparison. If the task is deviation around a target, use a dot plot or difference chart with the target labeled explicitly.
Dual-axis time series
Pattern: separate the scaling problem from the substantive claim.
Direction: A dual-axis chart can manufacture or hide correlation because the two axes are independently scaled. Use indexed lines when relative change is the task, small multiples for separate histories, or a scatterplot if the relationship between variables is the question.
Map of raw counts
Pattern: ask whether geography or population explains the pattern.
Direction: If high values mostly identify high-population places, use a rate or exposure-normalized measure. Pair the map with a ranked dot plot when precise comparison matters.
Bar chart of means
Pattern: check whether the summary hides spread.
Direction: When group overlap, skew, sample size, or outliers matter, show raw points with intervals, a dot/box hybrid, or another distribution display. Keep the mean only if it answers the decision.
Dense dashboard
Pattern: redesign around decisions, not widgets.
Direction: Group panels by workflow, align time windows, show denominators, and put alerts next to trend/distribution evidence. Remove decorative gauges when a bullet, line, or table would show status plus context.
Anscombe-like data
Pattern: do not let summary statistics replace plots.
Direction: When multiple datasets share similar means, variances, or correlations, show scatterplots or residuals to reveal shape, outliers, nonlinearity, and leverage.