← Economics Explained · Learning path
How do you start an economic data analysis?
Economic data analysis starts with a question precise enough to test. Instead of asking whether an economy is strong, ask whether household purchasing power improved over the past two years. Choose the income measure, inflation adjustment, population and comparison period that answer that question. Write down the relationship you expect to find before examining the results.
Build a reproducible dataset
Keep the original download unchanged. In a separate working table, record series codes, URLs, retrieval dates, units and transformations. Flag revisions, breaks and missing values. Align frequencies carefully: averaging a monthly rate is not the same as summing a monthly flow. Do not mix nominal local currency, constant price series and current US dollars in one unexplained comparison.
Use calculations that answer the question
Percentage change is (new/old minus 1)×100, when the denominator is meaningful and nonzero. A percentage point change subtracts two percentages. Compound annual growth is (end/start)^(1/years) minus 1 for positive, comparable levels. Per capita values divide by population. Rebase a positive index as value/base×100 to compare paths, while remembering that rebasing hides absolute size. Ratios with very small or negative denominators need special care.
Correlation is a clue, not identification
Two series may move together because of a third factor, reverse causality, common trends or chance. Regression controls only for included variables under its assumptions. A convincing causal claim needs a defensible design: a suitable comparison group, a natural experiment or another identification strategy, plus checks of assumptions. Statistical significance does not establish a large practical effect or a reliable forecast.
Scenarios and falsifiable conclusions
Separate observed facts, your interpretation and assumptions. Write a central scenario with an upside and downside, naming the mechanisms and triggers. Probabilities should not be invented simply to make a table look precise. If forecasting, compare performance with a simple benchmark and assess it on data not used to fit the model. Use the data vintage that was actually available at the forecast date.
An example argument
Suppose retail sales rose while consumer confidence fell. Check whether sales are nominal, whether prices rose, whether durable purchases were brought forward and whether the survey covers the same population. A defensible conclusion may be “nominal spending rose, but evidence of stronger real demand is incomplete.” State the next observation that would strengthen or overturn it.
Sources and further reading
Examples use hypothetical figures unless stated otherwise. Updated 8 October 2026.
