Nonresponse in a Survey of Military Personnel: What Record Data Can Tell Us

• Activity Number: 313
• Type: Contributed
• Date/Time: Tuesday, August 3, 2010 : 8:30 AM to 10:20 AM
• Sponsor: Section on Survey Research Methods
• Author(s): Monica L. Wolford* and Lynn M. Milan+
• Companies: U.S. Army Research Institute and U.S. Army Research Institute
• Keywords: Unit nonresponse ; data mining ; decision trees ; nonrespondent characteristics ; military populations

Abstract

While even the random loss of respondents can create problems by eroding the power of analyses, many authors have noted that the most critical problem is the bias that the disproportionate loss of particular subgroups introduces into the analyses (Musick, Campbell, & Ellison, 2001). The size of the bias depends primarily on the relationship between the sources of nonresponse (e.g., demographic characteristics such as age) and their associated response probabilities, as well as the amount of nonresponse (Bethlehem, 2002). Reducing this bias by increasing cooperation in the underrepresented subgroups will result in more accurate estimates. This paper uses data mining techniques (specifically, decision tree analyses) to identify patterns of nonresponse in a Web survey of more than 60,000 U.S. Army Officers conducted by the U.S. Army Research Institute.