Data Visualization: Choosing Clear and Honest Charts

Decide what the reader needs to see

Data visualization makes a pattern inspectable; it should not substitute for the reasoning needed to interpret it. Begin with a question, the audience and the data’s unit. Are you showing change over time, comparing groups, revealing a distribution or examining a relationship? A chart that serves one purpose can obscure another. Choose the display after identifying the claim that a reader should be able to check.

Inspect the data before drawing. Define each variable, source, time period, population and missing value. A line for monthly service use may count visits rather than distinct people. A map of rates needs a denominator. A chart of percentages across groups may hide that the groups contain very different numbers of observations. State these facts near the visual so a reader can understand what a mark represents.

Ask whether a table would work better. A few exact values are often easier to read in a compact table; a chart helps when shape, ranking or variation matters. Avoid filling a page with decorative charts that repeat the same finding. Give each visual a clear role in the argument and refer to it in the text with a conclusion supported by the plotted data.

Match the visual form to the comparison

Use a line chart for observations across ordered time points when the continuity is meaningful. Do not connect categories as if the space between them has numerical meaning. Use a bar chart to compare categories, with a visible baseline when bar length conveys magnitude. A dot plot may make small differences easier to compare without emphasizing area. A scatterplot helps inspect the relationship between two measured variables, including clusters and unusual cases.

For a distribution, show more than the average when variation matters. A histogram can reveal a long tail; a box plot can summarize spread but may hide multiple modes. Choose bins, scales and summaries openly. A single average mark cannot show whether most observations are near it. In a class assessment example, two groups can share a mean while one has a wide range of scores and the other a narrow range.

Maps require particular care. Large areas draw attention even when few people live there. If the point is risk to residents, plot an appropriate rate and explain unstable values in small populations. If the point is total demand for services, counts may be relevant instead. The chosen unit and denominator change what readers perceive as the problem.

Make comparisons honest

Set scales that allow the intended comparison without visual distortion. A truncated vertical axis can make a modest change look dramatic, especially when using bars. A line chart may sometimes use a narrowed range to reveal variation, but label the range clearly and explain why it helps. Avoid changing scales across adjacent panels unless the distinction is obvious. Provide exact values or annotations when a visual effect could mislead.

Color should encode meaning, not decorate. Use a limited palette with sufficient contrast and do not rely on color alone to distinguish groups. Direct labels often reduce the effort required to match lines to a legend. A color gradient implies order; use it for ordered data rather than unrelated categories. If values are missing, show them as missing instead of coloring them as zero.

Keep the denominator and time frame consistent. A year-to-year rise in recorded incidents may reflect a change in reporting. Show where a definition or collection system changed. When categories overlap, say so; stacked segments should not imply a neat partition that the data do not support. Avoid plotting percentages from tiny groups without their counts.

Show uncertainty and limits

When estimates come from samples, indicate uncertainty where it is important to the claim. Intervals can reveal that an apparent ranking is unstable. Explain what an interval represents and what assumptions produced it; do not treat it as a decorative whisker. If an estimate is particularly uncertain because of missing responses, mention that source of uncertainty rather than showing an unjustifiably precise number.

Outliers may be data errors or meaningful cases. Check the source before excluding them. A logarithmic scale can make a wide range legible, but label it and explain how equal distances on it differ from equal arithmetic differences. Never use a transformation to make an unwanted result disappear. Readers should be able to understand how the visual relates to the original units.

Correlation plots should not imply causation. A trend line summarizes an association under a chosen model; it does not show that one variable changed the other. Consider confounding, selection and time order. Describe what the plotted points are and whether repeated observations from the same unit are independent. A compelling diagonal arrangement still requires a study design to support a causal claim.

Build an accessible explanation

Give the chart a title that describes the content without pre-deciding the interpretation. Label axes with units, define abbreviations and note the source and coverage. If a visual is shared online, provide an accompanying text description of its main pattern and any important exceptions. Ensure labels remain readable at the intended display size and the content can be understood without distinguishing colors alone.

Write the surrounding prose to interpret, not duplicate, the picture. “Visits rose after March, with a sharp temporary drop in June” is useful if the chart shows those features. Add the relevant context: a policy change, revised data collection or seasonal pattern where supported. Do not claim an event caused the shift simply because it occurred near the same date.

Test the design with a reader unfamiliar with the dataset. Ask what they think is measured, where the largest change occurs and what uncertainty they notice. If they infer something false from the chart, revise the design or caption. A visualization succeeds when its meaning is clear without relying on an elaborate oral explanation from the creator.

Audit the final figure

Compare every plotted value with the source, check units and rounding, and confirm that filters did not silently remove observations. Review labels, legends and captions against the final data version. If the chart was revised after feedback, make sure the accompanying prose reflects the new figure. Store enough information about the data and chart choices for a colleague to reproduce the result.

Conclude with the inference the display permits and the question it leaves open. A well-designed visual makes a comparison easier to inspect while revealing its scale and uncertainty. It respects the data’s context and helps a reader challenge, rather than merely admire, the conclusion.