How would you visually represent statistical uncertainty in a chart?

WHAT THIS TESTS: This question tests whether you recognize that a plotted point is not the parameter itself but an estimate, and that humans have a strong perceptual bias toward treating whatever is drawn as exact truth. The interviewer wants to know if you can design a visualization that resists false precision and helps stakeholders make decisions under uncertainty.

A GOOD ANSWER COVERS: A strong answer hits four things in order. First, it acknowledges Wilke's observation that viewers instinctively read a data point at a specific location as the precise true value, so an isolated bar chart of conversion rates implies certainty that does not exist. Second, it states that because the dataset contains uncertainty, you must explicitly visualize the uncertainty of the point estimates rather than hiding it; this means augmenting or replacing the simple bar with an interval that shows the range of plausible values. Third, it notes that the choice of representation changes perception, so you should pick a method that lets the audience see overlap and judge magnitude simultaneously. Fourth, it brings in advanced techniques such as hypothetical outcome plots to make the abstract notion of uncertainty concrete by showing possible values rather than a single static interval.

COMMON WRONG ANSWERS: The biggest red flag is suggesting you can communicate uncertainty with a p-value table or asterisks above bars, because that forces the audience into binary significant or not thinking and hides the actual range. Another mistake is claiming that standard error bars alone fully solve the problem without discussing how they are often misread as hard boundaries or without addressing the visual clutter they introduce when comparing many groups.

LIKELY FOLLOW-UPS: An interviewer might push you to defend your chosen interval type versus another, ask how you would handle the visual complexity if you have twenty variants instead of two, or ask whether animation in hypothetical outcome plots introduces its own cognitive biases. They may also ask how you would explain the chart to an executive who just wants the winner.

ONE CONCRETE EXAMPLE: Imagine an A/B test where variant A converts at 12 percent and variant B at 14 percent. Instead of two bars ending at those heights, you draw a point estimate dot for each variant anchored to its bar, then extend a 95 percent confidence interval vertically so the viewer can immediately see whether the intervals overlap. If you want to go further, you render five frames of a hypothetical outcome plot, each frame sampling from the posterior distribution, so the stakeholder sees the 14 percent bar sometimes dropping below 12 percent and understands the result is not guaranteed.

Source: Claus Wilke, Fundamentals of Data Visualization