The National Children's Study (NCS): establishment and protection of the inferential base.

The National Children's Study (NCS) is a unique study of environment and health that will follow a cohort of 100 000 women from prior to or early in pregnancy and then their children until 21 years of age. The NCS cohort will be a national multi-stage probability sample, using a U.S. Census Bureau geographic sampling frame unrelated to factors that might influence selection into the sample (e.g. access to health care). I present the case for the use of a national probability sample as the design base for the NCS, arguing that selection of the original cohort should be as free from selection bias as possible. The dangers of using a selected or nonprobability sample approach are demonstrated by an example of its use in outlining the clinical management of children with febrile seizures, an infrequent disorder, which was so wrong for decades. In addition, I stress the importance of and the NCS approach to avoiding selection bias that might occur after the initial selection of the cohort. The selection of and maintenance of an unselected cohort is an important element for the validity of inferences in this major undertaking.