Public data workshop · Comparable numbers

Compare World Bank figures without mixing indicators or years

“Country A has overtaken Country B.” Before checking the arithmetic, recover the small print beneath the chart. Two numbers can both come from the World Bank and still make an unfair comparison: they may use different indicators, different years or different versions of the data.

Build the comparison from a shared indicator code and observation year. Then preserve the unit, country codes, source notes and download date. If those pieces do not align, describe the gap instead of forcing the figures into a ranking.

A familiar title can conceal a different measure

The World Bank publishes separate series for GDP in current US dollars, code NY.GDP.MKTP.CD, and GDP in constant 2015 US dollars, code NY.GDP.MKTP.KD. “GDP” alone is not an adequate label for a copied column.

Its current-versus-constant price explanation distinguishes prices of the period from prices of a reference year. Constant-price series help assess changes in output volume; current-dollar changes also reflect prices and currency conversion. Do not describe a rise in current US dollars as real growth without checking a suitable measure.

For each column, copy the complete indicator name and code. Preserve any per-person, percentage, currency or scale label. If the webpage displays millions but the downloaded file stores units, record the conversion you made. Multiplying one side by a million will not repair a mismatch between two different indicators.

Run the common-year test

On the indicator pages, the country table labels its columns “Most Recent Year” and “Most Recent Value.” Read both. A latest-available comparison is a different question from a comparison at the same point in time.

Invented teaching dataset: the figures below are fictional GDP levels in billions of constant 2015 US dollars. They are not World Bank observations or estimates about real countries.

Sorting the latest entries puts A above B. Comparing the common year 2023 puts B above A. Neither calculation supplies B’s missing 2024 value, so the table cannot establish that A overtook B in 2024.

A defensible caption would be: “In the fictional dataset, B is higher in 2023. A’s latest observation is from 2024, while B has no 2024 observation; a same-year comparison for 2024 is unavailable.” If latest-available values are useful for your question, you can still show them, with each year visible and the limitation in the caption.

Leave the missing cell missing. Replacing it with zero creates a new, false observation. Carrying forward 86 is an assumption; if your analysis requires that assumption, put it in a separate derived column and explain it. Do not silently overwrite the source.

Make one small download reproducible

1. Open the indicator’s original page and follow its data download. Keep the untouched file and its metadata together.
2. Select the intended economies by code and choose explicit years. Check whether a selected row is an economy or a regional/income aggregate before calling it a country.
3. Read the definition and notes for the selected series. Check coverage, estimation qualifications and any breaks relevant to your comparison.
4. Perform calculations on a working copy. Keep a note of filtering, unit conversion, rounding and missing-value handling.
5. Write the conclusion with the exact years and measure. Another reviewer should be able to rebuild it without guessing which “GDP” you meant.

When moving the file into a spreadsheet, keep country and indicator codes intact; the CSV import guide covers preservation of identifier columns. A file’s SHA-256 hash can help identify the copy you analysed, but cannot certify the figures themselves.

When an old number changes

A historical observation year and a download date are different dates. In its May 2012 article on GDP revisions, the World Bank describes how new source information and changes to estimation methods can revise earlier values. This is a historical explanation of revisions, not evidence that any particular value changed today.

If your saved file and today’s download disagree, keep both. First match the series, economy and year; then compare the metadata and revision notes. Without the older file or a documented earlier value, do not invent the size or cause of a revision. For a current comparison, use a consistent data version across the selected rows.

Attach this compact record to the evidence section of your report:

Question and proposed comparison:
Indicator name / code / unit and scale:
Economy codes / common observation year:
Download URL / retrieval date / saved filename:
Source notes and missing observations:
Transformations and rounding:
Result supported by these rows:
Unanswered comparison:

The goal is a comparison someone can reproduce, including the point at which the available numbers stop answering the question.