Begin with units
If a model predicts sales = 100 + 3 × advertising spend, the slope is uninterpretable until the units are stated. Is spend in pounds, hundreds of pounds or thousands? Is the outcome daily orders or monthly revenue? A coefficient’s size alone cannot answer whether the relationship is useful.
Prediction and explanation are not the same
A fitted relationship can summarise the sample or support prediction under stable conditions. A causal interpretation requires more. Omitted factors correlated with the included variable can bias a coefficient, and reverse causation may explain some patterns.
Read uncertainty honestly
A confidence interval reflects a specified estimation procedure and assumptions. It does not cover every kind of measurement error or model failure. Narrow intervals can accompany a misleading design. Extrapolation beyond the observed range is a separate risk.
A practical reading checklist
Write down the outcome, units, population, period, coefficient, uncertainty and identifying assumptions. Then ask whether the conclusion is about association, prediction or causation. Those questions make economic data more informative without pretending every reader needs to become a statistician.