The number is made in somebody’s front room· 1 of 4
The number is made in somebody’s front room
Every headline poverty rate begins with a single enumerator at a single door, and the properties of the final figure — its accuracy, its comparability, its revision history — are determined before any statistician sees the data.
The questionnaire at the door
A household survey is the primary instrument. In most low- and middle-income countries, the canonical form is either a Living Standards Measurement Study survey, developed by the World Bank, or a national variant — a household income and expenditure survey, a demographic survey, sometimes a labour force survey pressed into service as a proxy. Whichever design is chosen, its central act is sending an enumerator — a trained interviewer, often working for a national statistical office on a fixed-term contract — into a sampled household to administer a structured questionnaire, usually over one or two visits lasting an hour or more.
The questionnaire asks what a household consumed over a reference period: food purchased, food produced and eaten, food received as gifts, non-food items, services. The reference period matters enormously. A seven-day recall for food and a twelve-month recall for durables, in the same instrument, introduce systematically different recall error. The household's report of what it ate last Tuesday is shaped by the day of the week, whether there was a feast or a funeral, whether the respondent is the person who does the cooking. None of these variations appear in the final poverty rate. They are absorbed into the number.
The enumerator records what the respondent says. If the enumerator is in a hurry — and in a large national survey with ambitious cluster targets, enumerators often are — responses are anchored, leading questions slip in, and implausible entries are smoothed. These are not failures of character; they are the predictable outputs of fieldwork under time pressure, and they are well-documented in the survey methodology literature. The World Bank's LSMS team has published detailed assessments of enumerator effects, and the finding is consistent: interviewer identity is a statistically detectable source of variation in consumption data.
From doorstep to database
The raw questionnaire moves through several transformation stages before it becomes the consumption aggregate that anchors the poverty calculation. Data entry — still manual in some national surveys, increasingly done by tablet in others — introduces its own error rate. Editing routines flag implausible values and impute replacements. Prices collected at local markets are used to deflate consumption to a common spatial and temporal base, and the quality of those price indices varies by region and by how often they are updated.
The poverty line itself is a further layer of decision. An absolute line based on the cost of a caloric minimum plus a non-food allowance depends on the reference population chosen to estimate the non-food share, the specific price data used, and whether the line is updated between survey rounds. The international line — the dollar-a-day figure that has migrated from $1.00 to $1.08 to $1.25 to $1.90 to $2.15 as purchasing power parity revisions accumulate — is set by the World Bank using ICP price data and is not the line most national statistical offices use for their official headcount. The gap between the international rate and the national rate for the same country in the same year can be substantial, and the two figures answer different questions.
Weighting and expansion complete the journey from sample to national estimate. The sampling frame — usually drawn from a recent population census — is used to construct expansion weights, and the accuracy of the poverty rate is bounded above by the accuracy of that frame. In countries where the last census is more than a decade old, or where the census itself missed mobile populations, informal settlements or nomadic communities, the frame is stale, and the estimate inherits its gaps. The final figure is published as a point estimate with, if the publishing office is careful, a confidence interval. In practice, confidence intervals are routinely omitted from headline reporting, and national statistical offices are not always resourced to compute them.
What the uncertainty looks like from outside
When the figure enters an international database — the World Bank's PovcalNet, now succeeded by the Poverty and Inequality Platform, or UNICEF's MICS compilation — it carries metadata: the survey name, the reference year, the welfare aggregate type. What the database entry cannot carry is the full chain of fieldwork decisions that shaped the number, because those decisions are embedded in survey documentation that is often incomplete, sometimes not publicly archived, and occasionally written after the fact.
Comparability across countries and time depends on whether those fieldwork decisions were similar enough. A switch from a per-capita welfare aggregate to an adult-equivalent scale, a change in the recall period, a revision to the price deflator: any of these can move a poverty rate by several percentage points without any change in the material circumstances of the population being measured. The 2015 assessment by researchers at the World Bank and elsewhere of measurement non-comparability across African household surveys documented precisely this: a large share of apparent poverty trends could not be distinguished from changes in survey design.
This is not an argument against using the figures. A poverty rate with known, documented uncertainty is more useful than no figure, and the methodological literature exists precisely because statisticians have been willing to examine the instrument. But the uncertainty has a shape and a location. It does not accumulate symmetrically or randomly. Enumerator effects, recall length, frame quality and price deflation each introduce bias in particular directions, and those biases interact with the characteristics of the poorest households — who are more likely to be in remote clusters, more likely to be missed by a stale census frame, more likely to have consumption patterns that resist the standard questionnaire categories.
The number that appears in a ministerial report, an MDG progress table or a donor's country strategy was produced by a person at a door, asking questions in a language that may not have been the respondent's first, recording answers into a form designed somewhere else, carrying a sample weight calculated from a census that may predate the survey by a decade. Every subsequent operation — editing, deflation, aggregation, international comparison, trend analysis — is applied to that starting point. The precision with which the final figure is stated is a property of the database, not of the measurement.
Read next