A data visualization and data mining approach to response and non-response analysis in survey research

Survey data based on self-selected samples are inherently subject to the threat of selection bias. In this study, both data visualization and data mining techniques were employed to examine whether nonresponse bias had affected a survey regarding 1:1 computing conducted at Arizona State University. Unlike conventional hypothesis testing, data visualization/EDA attends to pattern recognition instead of probabilistic inferences. In addition, unlike logistic regression, classification trees in data mining are capable of ranking independent variables in terms of their predictive power. In contrast to the findings of other studies, this study reveals that academic level, gender, and race were not identified as crucial factors in determining the response rate. Rather, the nature of the subject matter might be more important for science/engineering and law students seemed more interested in this technology-related survey.