Start with what the number represents
A chart can make an estimate look more exact than it is. A bar labeled 48% may appear to describe an entire audience precisely, even when it comes from a sample. Before choosing a visual treatment, identify the quantity being estimated: a proportion, an average, a difference between groups or a change over time. Then identify what uncertainty information is available for that quantity.
A margin of error is commonly used to describe sampling uncertainty around a survey estimate. For example, an estimate of 48% with a margin of error of 4 percentage points can be displayed as an interval from 44% to 52%. That interval is not a promise that the true value lies within those bounds. Its interpretation depends on the method used to calculate it, including the confidence level and assumptions about how the sample was obtained.
Match the display to the question
For a small set of estimates, dots with horizontal intervals often work better than bars. The dot marks the estimate, while the line shows its uncertainty; readers can compare both without mistaking the length of a bar for certainty. If bars are important to the story, error bars can be added, but they should be visually distinct and explained in a nearby note. State whether they represent margins of error, confidence intervals, standard errors or something else. These are not interchangeable labels.
For a trend, a line surrounded by a lightly shaded interval can show how uncertainty changes across measurements. Avoid connecting points in ways that suggest continuous measurement when data were collected only at separate times. When displaying a range of possible outcomes rather than an estimate’s sampling uncertainty, use a different label, such as “scenario range.” A forecast range and a survey confidence interval answer different questions, even if both appear as shaded bands.
Make comparisons without overstating them
Readers often want to know which group is ahead or whether a result has changed. Two estimates can differ on a chart without providing strong evidence of a difference in the wider audience. Conversely, overlapping intervals do not, by themselves, prove that there is no meaningful difference. If the comparison matters, calculate and display the uncertainty around the difference where the research design allows it. This is especially important when the estimates come from the same respondents or other related observations.
Scale also shapes interpretation. Use an axis that lets readers see the interval boundaries, and avoid cropping an interval without a clear break or explanation. For percentages, keep the possible 0% to 100% range in mind; a narrow axis can make modest differences look dramatic. Round estimates and interval endpoints consistently so the labels do not imply more precision than the research supports.
Explain which uncertainty the chart does not capture
A conventional survey margin of error generally addresses sampling variability under particular statistical assumptions. It does not automatically account for people who were not reached, declined to answer, misunderstood a question or gave an answer influenced by its wording. Weighting and complex sampling can also affect interval calculations. If estimates are weighted, check that the displayed intervals use a method appropriate to the design rather than assuming the raw respondent count tells the whole story.
Be especially careful with subgroup charts. A survey may have many respondents overall but relatively few in one audience segment, making that segment’s estimate less precise. Show the subgroup’s base size alongside its interval where practical, and define what the base includes. For opt-in or other nonprobability samples, do not attach a conventional probability-sampling margin of error unless the method and its assumptions support one. A candid methods note is more useful than a number that suggests unwarranted certainty.
Build a chart readers can use
A useful uncertainty chart should pass a quick reading test: Can someone identify the estimate, interpret the interval and understand what was measured without searching for a separate report? Give the chart a specific title, such as “Share who prefer each option,” and label the population, question or measure, and units. In a concise note, name the interval type and confidence level when applicable, plus any important base-size or sampling limitations.
Do not rely on color alone to distinguish estimates from their intervals. Use visible endpoints, sufficient contrast and direct labels where space permits. If the chart becomes crowded, split it into small panels or highlight the comparisons that matter to the decision. Showing uncertainty is not about making every result seem inconclusive. It helps readers distinguish a clear pattern from a close call—and decide what is worth investigating further.
