Increasing Understanding of Survey Re-Weighting with Visualization

Abstract

Surveys are a widely used tool for inferring information about a target population by collecting data from a smaller sample. However information is always lost, as the sample cannot in general fully represent the target population. Thus predictions made from surveys may contain biases. To mitigate these biases, sample subgroups can be re-weighted to match their known distributions in the target population. We introduce a web-based interactive tool to visualize the re-weighting process in surveys, with specific application to presidential election polls. A detailed description of the system’s user interface and re-weighting algorithm are provided. In addition, the results of a twenty-person user study evaluating the system are presented and discussed.

Publication
IEEE VIS Posters