Overall considerations Showing uncertainty in charts

Often there is uncertainty associated with our data. This page outlines when and how to show uncertainty in charts. 

Whether to Show Uncertainty

In some cases, showing uncertainty is important to give an accurate impression of the data. However, it also adds complexity that makes charts harder to read. Some chart types, such as stacked bar charts, cannot effectively show uncertainty. 

Types of Uncertainty

This guidance refers to uncertainty ranges: two values indicating the likely range of an estimate, for example, confidence or prediction intervals. 

Sometimes uncertainty is represented by a single number, such as a standard error, variance or coefficient of variation. These should not be shown in charts, as they are hard to interpret and compare. However, they should be included in datasets (if appropriate). 

When Not to Show Uncertainty

You should not show uncertainty on charts if it is not necessary for interpreting the data. For example: 

• if the uncertainty is consistently very small and would not change the interpretation
• if there is little or no overlap between uncertainty ranges for values being compared, for example, if the values are clearly different
• if there is some overlap between the uncertainty of adjacent points, but the overall trends are still very clear 

As set out in the page Principles (opens in a new tab) , the aim of charts is not to display the entire dataset. Instead, include full data and uncertainty estimates in downloadable datasets. 

A chart with unnecessary uncertainty ranges

In this chart, the uncertainty ranges are narrow and do not overlap. They add visual noise without improving understanding. Removing them would not lead to a different interpretation of the chart. 

When to Show Uncertainty

You should show uncertainty when it is important for understanding key trends in the data and when it would fundamentally change the interpretation. For example: 

• if there are large overlaps between uncertainty ranges for different categories, suggesting comparisons may not be statistically valid 
• if uncertainty is notably higher or lower in specific categories or time periods  

Example of a chart with necessary uncertainty ranges

Without the uncertainty ranges shown, one might wrongly conclude that men always have higher values than women. The intervals show this is only true in 2016 and 2023.

Data with Very High Uncertainty

If uncertainty is so high that no meaningful trends or comparisons can be made, do not show the data in a chart. This could be confusing and risk misinforming users. There are exceptions  when: 

• the aim of the chart is to show no difference or no trend 
• this is clearly explained in the chart title 
• the data is based on a robust, large sample size 

How to Show Uncertainty

To show uncertainty we use a shaded band around a central estimate. This is the clearest method and avoids adding unnecessary complexity. 

Chart Types That Can Show Uncertainty:

• bar charts, clustered bar charts, and column charts (using dot plots with ranges)
• line charts with shaded uncertainty bands 

Styles for these charts are set out in our Notion chart style guide (opens in a new tab) .

We do not use bar or column charts with error bars, as they are harder to interpret accurately. 

Example of a chart with shaded uncertainty bands

With the uncertainty ranges shown, one can conclude that the two younger age groups are similar, but that there is a difference between the two older age groups. 

The visual styling for these charts is set out in our Notion chart style guide (opens in a new tab) .

Explaining Uncertainty in Charts

Use legends or annotations to clearly explain what is being shown. Use plain language, such as, “we are 95% sure the value falls in this range”. 

Avoid technical terms like “confidence interval” or “statistical significance,” or, if they are needed, provide simple explanations. 

Legend styles depend on chart type and are set out in our Notion chart style guide (opens in a new tab) .