Why Definitions Decide Debates in Public Statistics

When two commentators cite wildly different numbers for the same phenomenon, the explanation is usually not that one dataset is fake. It is that the two numbers measure differently defined things. Three familiar examples show the pattern.

#Unemployment: jobless is not the same as unemployed

Under the International Labour Organization definition used for headline rates, a person is unemployed only if they are without work, actively seeking it, and available to start. People who have stopped searching, or who work a few hours while wanting more, fall outside the headline count — which is why statistical agencies publish broader measures of labour underutilisation alongside it. Neither number is “the truth”; they answer different questions.

#Poverty: absolute or relative?

The World Bank’s international extreme-poverty line measures consumption against a fixed threshold adjusted for purchasing power — revised in 2025 to $3.00 a day at 2021 prices — and is designed to track absolute deprivation. The European Union’s headline indicator instead counts people below 60% of national median income: a relative measure that tracks inequality’s lower tail. Absolute poverty can fall while relative poverty rises, and vice versa; debates that ignore the distinction talk past each other by construction.

#Migration: foreign-born or foreign citizens?

Migrant statistics may count the foreign-born (a fixed fact of biography) or foreign citizens (which changes with naturalisation), and flows may or may not include returning nationals, students, or short stays. Countries differ in which they emphasise, and the resulting figures can differ by large factors while all being individually correct.

#The practical rule

Definitions are not deceptions — they are unavoidable choices, each fit for some purposes and unfit for others. The integrity test is whether the definition is stated, stable over time, and appropriate to the claim being made. A statistic quoted without its definition is unfinished; our checklist for such cases is in Five Questions to Ask Before Trusting a Statistic.