Start every analysis with the question, not the data

Spiegelhalter's PPDAC cycle — Problem, Plan, Data, Analysis, Conclusion — insists that statistical work begins with clearly defining the real-world question before touching any numbers. Analysts who skip straight to available data risk answering a convenient question instead of the one that actually matters, a mismatch he calls one of the most common sources of misleading statistics in practice.

He illustrates this with examples where a dataset technically answers a question but not the one headlines claim it answers — such as conflating correlation found in an observational dataset with a causal claim the data was never designed to support. The "Plan" stage forces analysts to decide in advance how data will be collected and analyzed, preventing the temptation to fish through results afterward for whatever pattern looks most compelling.

Only once the problem and plan are locked in does the actual data collection and number-crunching happen, followed by a conclusion appropriately hedged by the plan's original scope.

Takeaway: before trusting any statistic, ask what specific question it was designed to answer.