During the 1990–92 recruitment campaign, the ALSPAC target sample was a dynamic population for whom no convenient sampling frame to support systematic invitation of all eligible individuals was available. The eligible study sample has been defined retrospectively, based on ALSPAC recruitment records and maternity, birth and child health records.5 Comparing these health records with ALSPAC study records, we are able to describe study recruitment (Figure 3). Recruitment was not complete, which is unsurprising given its opportunistic nature. It is impossible to fully ascertain retrospectively where non-recruitment represents failure to invite as opposed to active non-response. Records of the returned expression of interest cards and the proportion of these women declining participation are known (Figure 3); however, the number of invitation cards distributed is unknown. The methods used in retrospectively defining the cohort allowed the use of more accurate geocoding resources, unavailable to recruiters in the early 1990s. These resources identified 229 pregnancies (233 children), living on the geographical periphery of the study area, that were incorrectly assumed to be eligible in 1990–92. As these individuals enrolled and contributed data, we
