Implications of Attrition in a Longitudinal Web-Based Survey: An Examination of College Students Participating in a Tobacco Use Study.

• Type of Material: Article
• Authors: Bennett McDonald, Emory University; Regine Haardoerfer, Emory University; Michael Windle, Emory University; Michael Goodman, Emory University; Carla Berg, Emory University
• Language: English
• Date: 2017-10-16
• Publisher: JMIR Publications
• Publication Version: Final Published Version
• Copyright Statement: ©Bennett McDonald, Regine Haardoerfer, Michael Windle, Michael Goodman, Carla Berg.
• License: Creative Commons Attribution 4.0 International
• Final Published Version (URL): http://dx.doi.org/10.2196/publichealth.7424
• Title of Journal or Parent Work: JMIR Public Health and Surveillance
• ISSN: 2369-2960
• Volume: 3
• Issue: 4
• Start Page: e73
• End Page: e73
• Grant/Funding Information: This research was supported by the National Cancer Institute (1R01CA179422-01; PI: Berg).
• Keywords: methods; tobacco use; risk factors; young adults; surveys and questionnaires
• Research Categories: Health Sciences, Public Health; Health Sciences, Epidemiology

Abstract

BACKGROUND: Web-based survey research has several benefits, including low cost and burden, as well as high use of the Internet, particularly among young adults. In the context of longitudinal studies, attrition raises concerns regarding the validity of data, given the potential associations with individual and institutional characteristics, or the focal area of study (eg, cigarette use). OBJECTIVES: The objective of this study was to compare baseline characteristics of nonresponders versus responders in a sample of young adult college students in a Web-based longitudinal study regarding tobacco use. METHODS: We conducted a secondary data analysis of 3189 college students from seven Georgia colleges and universities in a 2-year longitudinal study. We examined baseline tobacco use, as well as individual- and institutional-level factors, as predictors of attrition between wave 1 (October and November 2014) and wave 2 (February and March 2015) using multilevel modeling. Results: A total 13.14% (419/3189) participants were lost to follow-up at wave 2. Predictors of nonresponse were similar in the models examining individual-level factors and institutional-level factors only and included being black versus white (odds ratio [OR] 1.74, CI 1.23-2.46); being male versus female (OR 1.41, CI 1.10-1.79); seeking a bachelor's degree versus advanced degree (OR 1.41, CI 1.09-1.83); not residing on campus (OR 0.62, CI 0.46-0.84); past 30-day tobacco use (OR 1.41, CI 1.10-1.78); attending a nonprivate college (OR 0.48, CI 0.33-0.71); and attending a college with ≤10,000 students (OR 0.56, CI 0.43-0.73). CONCLUSIONS: Future longitudinal studies should assess predictors of attrition to examine how survey topic and other individual and institutional factors might influence the response to allow for correction of selection bias.