The advent of probabilistic sampling methods has been commonly hailed as a revolution in the work carried by government statistical offices. To governments assuming newly defined functions triggered first by the 1930s Depression and then by an unprecedented war effort, the advent of a technique that could provide much-needed data rapidly and at cheaper cost was obviously a blessing. To national statistical systems driven by an ambitious but inconclusive attempt to ‘know everything about anything’ but whose energies were absorbed by cumbersome and costly censuses, sampling offered a practical way to meet the exponential growth in information requests. By replacing the idea of exhaustiveness with that of a reasonable and measurable precision, it became possible to multiply inquiries, and more timely statistical information could really be taken as a basis for decision. The long-standing dilemma between ‘many cases, few variables’ and ‘few cases, many variables’ could be elegantly resolved.

Giving an adequate account of how and why sample surveys came to play an important role in government information-gathering activities requires that we take into consideration a number of contexts. Clearly, there is a narrowly scientific aspect to the story. The idea of sampling, which ran counter to a century of efforts devoted to the improvement of statistical coverage by administrative control and cross-checking, could not be widely trusted before the community of statisticians had reached an agreement on strictly defined standards.