Predictors of Nonresponse in a Longitudinal Survey of Adolescents

Introduction Longitudinal design in surveys provides a unique opportunity to study the causes and effects of nonresponse, particularly when participants in a given round are recruited from a cohort of respondents from the prior rounds. The availability of survey data for respondents and nonrespondents in a given round allows one to directly estimate nonresponse bias and its components, as well as to gauge the relative effectiveness of weight adjustment strategies (Ka lsbeek et al. 2001). In addition, multiple contacts with the same sample over several rounds of data collection can be utilized to investigate the role that the survey design and process play in determining the final recruitment outcome for individual followup rounds. Recruiting the same sample on multiple occasions expands the list of potentially viable predictors of sample recruitment outcomes in longitudinal cohort studies. The more times one works with a sample, the more opportunities one has to observe those things that may influence recruitment in later rounds. Despite this, relatively few published studies on the predictors of nonresponse in longitudinal studies have been done. De Maio (1980), for example, examined the role of past survey experience, while Aneshensel et al. (1989) studied the role of characteristics of the baseline interview (e.g., length). More recently, Lengacher, et al. (1995) have studied the effect of incentives, and Campanelli and O’Muirchartaigh (1999) have reported on the role of interviewer continuity across consecutive rounds. Most of the remaining research on this topic has focused on nonresponse in cross-sectional in-person and telephone surveys. This paper examines factors affecting several outcomes of subject recruitment in followup rounds of multi-round cohort samples. In particular, our goal is to identify those design and process features which affect any of four 0/1 recruitment outcome variables, defined later for the National Longitudinal Study of Adolescent Health (the Add Health Study), a national school-based health survey of teenagers with several rounds of in-home followup after an initial in-school administration. These variables presume that recruitment can lead to study participation, or any of the following four types of nonresponse, defined according to where in the recruitment process the end result occurs (Lessler and Kalsbeek, 1992): 1) Not Solicited (NS) : Sample members are not solicited for participation by the interviewer because: they have moved and their new address is unknown, they are out of the country, or interviewers were not able to talk to them about survey participation after having established contact with their place of residence. 2) Solicited but Unable (SUA) : Sample members are asked to participation in the study, but they decline because of their inability to do so. Possible reasons include: physically/mentally incapable, language barriers, scheduling problems, and so on. 3) Solicited but Unwilling (SUW) : Sample members are asked to participate but they refuse. Reasons for declining in this way include: confidentiality concerns, mistrust of government, just too busy, topic too personal, don’t do surveys, and so on. 4) Other nonrespondents (OTH) : Sample members fail to become participants for a reason that does not fit in any of the three previous categories. Some examples are lost schedules, partial respondents, and other noninterviewable respondents. Following a conceptual framework for round-specific recruitment outcomes in certain longitudinal interview surveys, we fit separate multivariate logistic regression models for various recruitment outcomes of the in-home Wave II (IH2) round of the Add Health Study. Questionnaire and process data from the prior in-home Wave I round (IH1) were used in our search for predictors of four round-specific recruitment outcomes.