Abstract:
|
Non-probability (or design-free) surveys are becoming more prevalent because they offer both increased speed in obtaining data on emerging issues (e.g., online panel surveys) and decreased costs compared to many probability-based surveys. Evaluation studies to date have shown that many population estimates from non-probability surveys miss the mark (i.e., lack external validity) and suffer from errors associated with coverage, selection, or model misspecification. Few of these studies investigate if these estimates better serve subdomains within the population instead the population as a whole.
In this brief discussion, we will review various data collection methods employed in non-probability surveys. Next, we will review analytic approaches used to maximize the utility of the data from these surveys; where appropriate we will note the effectiveness of these methods in yielding comparable estimates to the chosen (sub)population "gold standards." We will conclude with several interesting aspects for further research, thus continuing the conversation on the fit for purpose of non-probability surveys.
|