Abstract:
|
High-quality surveys rely on a sampling frame, a pre-specified sampling process and weighting that maps results to an intended population. In practice this gold standard is tarnished by non-participation and item non-response; virtually all surveys depart from the ideal. In this context, low front-end cost and rapid accrual make web-based and other non-probably sampling methods attractive. Their use increases, and research is needed to develop designs and analyses that maintain sufficient quality relative to the real world performance of the tarnished gold standard. This call to action mirrors our proposal that epidemiological, clinical and other "science" communities must increase attention to survey goals. We counter the extreme view proposed by some that requiring "representativeness" in scientific studies is unnecessary, possibly counter productive. Their narrow definition is equivalent to self-weighting, but representation depends only on availability and use of appropriate weights. This form of representation is the primary goal for surveys and should be a priority for epidemiological and other studies in order to produce generalizable findings.
|
ASA Meetings Department
732 North Washington Street, Alexandria, VA 22314
(703) 684-1221 • meetings@amstat.org
Copyright © American Statistical Association.