The Art of Statistics

A headline announces: “Risk reduced by 50%.” That sounds enormous. Later, the text states that the risk fell from two cases to one per thousand people. The relative change is correct—yet without the baseline, it conveys a distorted picture.

David Spiegelhalter sees statistics as the art of learning from data. The focus is not calculation alone, but a transparent path from the question through the data to a cautious conclusion. Data literacy means neither blindly believing numbers nor dismissing them wholesale, but examining how they were produced and what they can tell us.

1. First the problem, then the data

It starts with a precise question. Who is to be described? Which difference or relationship interests us? Over what period? Only then can we judge what data are needed and whether the available data fit at all.

Spiegelhalter uses a statistical problem-solving process: problem, plan, data, analysis and conclusion. This sequence prevents us from searching a table for striking patterns and only afterwards inventing a suitable story. A sound plan also asks how observations were selected and characteristics measured.

2. Numbers need comparisons and reference measures

Percentages, averages and probabilities appear objective, but can mislead without context. “20% more” remains unclear without a baseline and comparison group. An average can conceal substantial differences. A high hit rate can be useless if an event is very rare and many false alarms occur.

Helpful questions therefore include: more than what? Out of how many people? How are the values distributed? Have absolute and relative risks been shown? And is the comparison group genuinely comparable? Often, a clear reference measure alone changes the interpretation.

3. Do not skip uncertainty and causality

Data are usually incomplete. Samples vary, measurements are prone to error and models rely on assumptions. Responsible statements make this uncertainty visible—for example, through intervals, sensitivity analyses or clear language about what is unknown.

Particular caution is needed with cause and effect. If two characteristics occur together, one may influence the other—but selection, chance or a third factor may also explain the pattern. A strong conclusion requires a research design that rules out alternative explanations as effectively as possible.

Your exercise: take a number apart

Take a current statistic from a presentation, news item or advertisement. Examine it for three minutes:

1. Minute 1 – Question and source. What exact question is the number meant to answer, who collected it and which group does it refer to?
2. Minute 2 – Comparison. What is the baseline or denominator? Express the figure as an absolute frequency as well, if possible.
3. Minute 3 – Limitation. Note one uncertainty and one alternative explanation. Which claim would be too strong for these data?

Then formulate a sentence containing both the insight and its limitation: “In this sample, we observed…; this does not yet establish…”

What we take from this

Statistical thinking strengthens data literacy and protects against premature certainty. In WEITER7 , it particularly supports the reality check: refine observations, choose suitable criteria and interpret results so that uncertainty remains part of learning.

Which number is influencing my decision right now—and what comparison or context do I need before trusting it?

Context: This insight condenses selected core ideas from Spiegelhalter’s introduction to statistical thinking. The exercise and WEITER7 connection are our own applications. This brief check does not replace proper analysis; for complex studies, design, data quality, model assumptions and reproducibility all matter. Cautious wording does not automatically make poor data reliable.

Sources and further reading: Official publisher’s page ↗ · Author’s University of Cambridge page ↗