Abstract

The data revolution has broadened the sources available for public statistics. Beyond widely used administrative records, geospatial information, big data, and citizen-generated data can now complement, or even replace, traditional censuses and sample surveys. These sources can improve timeliness, reduce costs, and support official statistics production. However, using them effectively requires careful attention to coverage, representativeness, and available auxiliary information. For many policy questions, the information needed simply does not exist in any current source. It has to be measured.

The Fundamental Principles of Official Statistics call for statistics of demonstrated practical utility. They also ask producers to choose data sources by weighing quality, timeliness, cost and respondent burden. Within this framework, the first step in any survey is to establish the demand for information. This means identifying who will use the results, which population groups and domains matter, and how precise the results need to be. It also means asking a key question: can the essential data be obtained without conducting a survey? If the answer is no, survey methods offer distinct strengths. These include controlled data collection for specific objectives, a clear definition of the target population, a sound inferential framework, and established tools for assessing accuracy and handling nonresponse.

This presentation argues that well-designed quantitative surveys remain a cornerstone and a reliable way to produce fit-for-purpose statistics when data needs are specific, local, or new. Relevance is the first dimension of quality: it is the main reason for conducting a survey. The talk illustrates this with a recent Brazilian survey designed around clearly defined information gaps.

The survey measured the effects of a dam collapse, Brazil’s worst mining disaster, on labour, income, and subsistence across 45 affected municipalities. To support impact estimation, 3,381 census tracts in the affected area were each matched with a comparable tract in a comparison area. A retrospective labour-history module covering 2014, 2016 and 2019 captured pre- and post-disaster conditions that no existing source could provide. Fieldwork produced almost 10,000 complete interviews. The survey's objectives, design, and scope aligned with the relevant data needs.