A CONNECTED LEARNING PATH / 18 GUIDES
Analytical reasoning.
Follow a question, explore a worked example and choose one useful action. Read in order or start with the challenge in front of you.
Explore every topicSeparate correlation from an explanation of cause
Separate observed association from causal explanation, compare credible alternatives and choose evidence that could distinguish them.
Segment a result before explaining the average
Explain an aggregate result by examining subgroup rates, subgroup weights and stable definitions before attributing a change to performance.
Inspect seasonality before interpreting a trend
Check recurring calendar patterns, exposure and comparable trading periods before interpreting adjacent observations as a sustained trend.
Describe what a small sample can and cannot show
Use small samples to investigate experiences and refine questions while keeping population claims, uncertainty and recruitment limits explicit.
Recognise survivor bias in a success report
Recognise when a success report excludes people who left, failed or disappeared, and rebuild the relevant starting population.
Check how a sample was selected
Review recruitment, coverage, response and eligibility to understand whose experience a sample can describe.
Look for a confounding change
Check simultaneous audience, operating and measurement changes before attributing an observed result to one intervention.
Understand regression toward the mean in a follow-up
Interpret follow-up results after an extreme observation by checking selection, ordinary variability and a credible comparison.
Use sensitivity analysis to expose a fragile conclusion
Test whether a recommendation changes when uncertain inputs move within justified ranges, and identify the assumptions worth resolving.
Bring the base rate into an alert review
Use event prevalence and a complete alert confusion table to distinguish detection rate from the chance that an alert is genuine.
Report absolute and relative changes together
Show counts, rates, percentage points and relative movement with a consistent baseline and denominator.
Explain uncertainty instead of adding false precision
Communicate ranges, assumptions and unresolved evidence without presenting unsupported decimal places or invented statistical intervals.
Backtest a forecast against information available at the time
Evaluate forecasts using frozen information sets, consistent horizons and transparent error measures instead of later corrected knowledge.
Keep missing-data uncertainty visible in an analysis
Keep absent groups and unresolved values visible, test defensible bounds and explain when missing data could reverse the decision.
Avoid searching many comparisons for one attractive result
Distinguish exploratory segment searching from a planned test, disclose the search scope and seek independent confirmation before strong claims.
Build a causal hypothesis as a testable chain
Turn a proposed cause into an observable sequence with timing, competing explanations and evidence that could contradict the mechanism.
Combine qualitative context with quantitative observations
Use interviews and operational records together while preserving each source’s scope, timing and limits on causal interpretation.
Set an action threshold using consequences
Choose alert and action thresholds by comparing consequences, review capacity and uncertainty rather than selecting a convenient score.