Using Ranking to Adjust Initial Sampling Weights (303250)

Keywords: survey research, convenience sample, bias, raking, survey weights, poststratification, SAS macros

Survey research can result in a sample distribution that differs from the population distribution. This could happen due to intentional sample design decisions; differential nonresponse patterns; or, in less rigorous settings, as the result of employing a convenience sample as opposed to a probability sample. In any of these cases, questions may arise regarding potential bias in population estimates. We will describe how to use an iterative algorithm to adjust survey estimates from under- or over-represented strata. Based on a raking procedure, the algorithm uses a SAS macro to iteratively adjust sample weights based on ratios of population and sample proportions for each demographic dimension. Two examples, one from a driver behavior study and one from a business establishment survey, will be detailed to show how this poststratification technique can be broadly applicable across many industries and purposes. We will highlight how the adjusted sample estimates differ from the raw sample estimates, and how the adjusted values better represent the target population. While knowledge of SAS Macro Programming Language is not essential, basic macro tools and concepts will be applied.