Analysis and Interpretation to Support Decision-Making
From description to decision
Analysis moves through levels: descriptive (what happened - totals, averages, trends), diagnostic (why - comparisons and correlations), predictive (what is likely - forecasting and modelling) and prescriptive (what should be done). Candidates should know measures of central tendency (mean, median, mode) and spread (range, standard deviation, interquartile range). The mean is sensitive to outliers; the median is robust. A dataset with a mean income far above the median signals a skewed distribution.
Correlation and causation
A central trap: correlation does not imply causation. Ice-cream sales and drowning incidents rise together, but heat drives both. Questions often present a strong correlation and ask for a valid conclusion - the correct answer notes a possible confounding variable or recommends further investigation. Another trap is extrapolation: projecting a trend well beyond the data range is unreliable.
Interpreting for decisions
Findings must connect to a recommendation. State the finding, the evidence, the limitation and the recommended action. Consider costs and benefits, and note that a statistically detectable difference may be too small to matter practically.
Worked example
A retailer finds customers who join the loyalty program spend 40% more on average. Concluding the program causes extra spending is unsafe - loyal heavy spenders may simply be more likely to join. A valid interpretation notes the correlation, identifies self-selection as a possible confounder, and suggests a controlled trial before rolling the program out nationally.