The offices that publish it

The infrastructure behind the figure

When a poverty estimate is published in a World Bank database or cited in a UN progress report on the Millennium Development Goals, the publication chain looks straightforward: a number appears, a source is cited, a year is given. What that citation obscures is the much longer chain that runs behind it. The World Bank did not, in most cases, send enumerators into the field. A national statistical office did. The international agency received a dataset, cleaned and harmonised it to a standard, ran a model or applied a poverty line, and produced the figure that ends up in the database. The office that actually made the measurement is several steps upstream, and its name rarely appears on the slide.

National statistical offices — NSOs in the standard shorthand — are government agencies with mandates set in national statistics law. They are responsible for census operations, household surveys, national accounts, and the civil registration systems that record births and deaths. In large economies with long statistical traditions, they are well-funded, technically staffed, and able to run a national household survey on a regular cycle. In smaller economies, particularly those that went through the HIPC debt-relief process in the early 2000s, the picture is different: hiring freezes tied to structural adjustment conditions, equipment that is a decade old, and survey cycles that depend on whether a donor project is active that year.

The OECD's Development Assistance Committee has tracked aid to statistics as a recognised category, and the figures reveal a persistent gap between the political importance attached to measurement — every MDG target required a number — and the actual investment in the offices expected to produce it. The Paris Declaration on Aid Effectiveness, agreed in 2005, committed donors to aligning aid with country systems, which in principle meant routing statistical support through NSOs rather than running parallel donor-funded data collection. In practice, alignment in the statistical domain has been partial. Parallel surveys, designed to meet a donor's specific indicator requirements, remain common.

What the offices actually do, and where the variation sits

A household survey — the primary instrument for measuring poverty, nutrition, education enrolment, and most of the indicators that fed MDG reporting — is a logistically complex operation. A sampling frame must be constructed, usually from a census. Enumerators must be trained, deployed, and supervised across geographically dispersed clusters. Questionnaires must be translated, pre-tested, and revised. Data must be entered, checked for internal consistency, and cleaned before analysis. The interval between fieldwork and a published report has historically been long: two to four years between data collection and a publicly available microdata file is not unusual in lower-income countries, which means the figures in any given year's report may describe conditions from several years earlier.

Capacity varies along several specific dimensions. Staffing is the most visible: an NSO in a country of thirty million people may have a professional statistical staff of a few dozen, against which an office in a European country of comparable population might field hundreds. Equipment matters for the georeferencing now standard in survey design and for the tablet-based data collection that has partly replaced paper forms. Legal and institutional independence matters too — an NSO that answers directly to a finance ministry faces different pressures when a politically sensitive figure is due to be published than one with a statutory guarantee of independence.

Financing is where the structural problem sits most clearly. Many NSOs in sub-Saharan Africa and parts of South and Southeast Asia have historically depended on external funding for a substantial share of their survey budgets. The World Bank's Living Standards Measurement Study programme has, since the 1980s, provided both technical assistance and funding for household surveys in low-income countries; UNICEF's Multiple Indicator Cluster Surveys, first launched in 1995, have filled gaps where DHS or LSMS coverage was absent. These contributions are genuine and the data they generate are used. But a survey funded by an external programme is designed partly around external priorities, and the office that implements it does not always retain full ownership of the microdata or the analytical capacity that comes from repeated independent operations.

The distance between production and publication

The international databases that researchers and policymakers actually use sit at the end of a chain that involves at least three distinct transformations of the original NSO output. The raw microdata is processed into a public-use file. That file is harmonised — variables renamed, categorical codes standardised, welfare aggregates reconstructed — to be comparable across countries and survey rounds. The harmonised data is then run through a poverty measurement methodology, which involves a poverty line expressed in local currency, adjusted for spatial price differences within the country, and then converted to an international comparison line using purchasing power parity exchange rates. Each transformation introduces choices, and the choices are not always visible to the reader of the final table.

Rebasing national accounts — which can shift a country's measured GDP by a large fraction in a single revision — creates a related problem for the poverty measurement chain. If the national accounts are revised, the consumption deflators used to construct welfare aggregates may need to be revised too. The figures in the database may then differ from figures previously published, without a clear explanation of which changed and why. An NSO that lacks the capacity to maintain consistent series over time, or that does not publish detailed methodological documentation, makes external verification difficult.

The statistical infrastructure required to produce reliable, timely, and consistent poverty data is not a background condition that can be assumed. It is a product of sustained institutional investment, legal frameworks that protect independence, and financing that is not contingent on a particular survey's alignment with donor priorities. The Paris monitoring surveys attempted to measure whether aid was becoming more aligned with country systems, but statistical capacity was not one of the twelve indicators against which progress was tracked — an omission that reflects how the effectiveness agenda prioritised the financial architecture of aid delivery over the technical infrastructure of measurement.

What the international databases publish is real, in the sense that it derives from real fieldwork and real analysis. But the quality of what is published is uneven in ways that the databases rarely make explicit. A figure with a confidence interval based on a well-powered survey run by a well-funded office with a long time series is a different kind of number from an estimate modelled from a single decade-old survey by an office with three professional statisticians. Both appear in the same table, formatted identically, with the same number of decimal places. The office that made the measurement is the place to look for which kind it is.