Review and Simulation of Different Sampling Methods

Obtaining information about hard-to-reach populations is a major challenge in the market research field. In our case of analyzing German minorities in Denmark, only a small fraction of the total population belongs to the target population. Therefore, selecting minorities by generating telephone numbers at random would result in very high costs. Alternative sampling methods have to be used, but there are no practices to identify the best approach. This article tries to fill this gap and creates a comparison of snowball sampling (SS), random digit dialing (RDD), gravity sampling (GS) and facility-based sampling (FBS). Sample data has been extracted by a previous survey (Hoops, Schnapp and Schaefer-Rolffs 2013) and a further model extended by randomly generating and simulating all four sampling methods using bootstrapping procedures. This enabled us to estimate the cardinality of the sample space, the bias and the variance of the inclusion probabilities in the sample for each method. Only GS and RDD create samples which are asymptotically unbiased. The combination of gravity and complete as well as non-overlapping citizens registers produces the highest cardinality of the sample space. But in contrast to RDD, citizens registry-office methods allow no household samples. So gravity analyses help to identify regions with a high prevalence of the target population to create samples with roughly varying inclusion probabilities. Our simulations indicate that gravity sampling methods using official databases produce very high quality samples. For cost reasons this method should be tested in practice to conduct surveys with hard-to-reach populations.