Oftentimes, researchers are faced with the challenging task of targeting rare domains in a population while maintaining the probability-based nature of the employed sample. For instance, in a national RDD sample it might be necessary to oversample households with small children or those with even less prevalent attributes. While an epsem sampling design, whereby all numbers have the same chance of selection, will provide the most efficient sample with respect to the precision of survey estimates, from a cost perspective such a design can be completely prohibitive due to the required level of screening for reaching eligible households. This is where a cleverly designed stratified sampling alternative that employs disproportional allocation can prove highly valuable.
In practice, an optimal sample allocation scheme takes into account the unit cost per interview in each sampling stratum. As such, a stratum with a high incidence of reaching members of the target population will receive a higher allocation as compared to other strata. This disproportionate sample allocation should be exercised while providing a non-zero chance of selection for all telephone numbers to ensure a probability-based sample.
The objective of this stratification is to provide a means for over sampling the target populations by segregating higher incidence households into distinct sampling strata. This is done by matching all numbers against commercial databases, which contain household and individual level demographic data, and identifying the numbers that meet the specified target.
With access to all the top commercial databases Marketing Systems Group can provide cost-effective solutions for sample surveys that aim to target rare domains. By placing such telephone numbers in the “top” or high incidence stratum and the remaining telephone numbers covering the geography of interest in another, you can create a complete sampling frame. Subsequently, using an optimization procedure a higher sampling fraction will be determined for the top stratum cognizant of the design effect that will result from a disproportional sample allocation and will need to be adjusted for when weighting.