Estimating prevalence of health and behavioral outcomes among survey non-respondents in an epidemiological study of child psychopathology in rural Maine
Bobashev, G. V., Zahner, G., Freeman, D., & Biemer, P. (2000). Estimating prevalence of health and behavioral outcomes among survey non-respondents in an epidemiological study of child psychopathology in rural Maine. In pp. 846–851. .
Background: An epidemiological survey of child psychopathology was conducted with children 4-11 years old residing in four rural counties in Maine, including children, parents and teachers as informants. It was difficult to obtain sensitive information (e.g., emotional problems) from this conservative, non-farm, rural population, which posed a challenge for estimating prevalence rates of the outcomes among survey non-respondents. Methods: In this study, we adjust for non-response in several ways using modelfree and model-based response propensity weights and also estimating prevalence of outcome variables among non-respondent population. Model-based adjustment was based on a brief non-respondent survey. First, time delay from first solicitation to actual response among survey respondents was used to select candidate survey items best associated with response delay. Then, these items were incorporated in a "brief" survey of nonrespondents and also were used to model the outcomes among respondents. Finally, we used the respondent model and the results of the "brief" survey of nonrespondents to predict the outcomes among the nonrespondents. Results: 2,133 eligible households were selected via a multi-level survey. Only 1,518 (71%) households have responded to the survey. About fifteen questionnaire items were selected for the nonrespondents survey. Response propensity was calculated. For outcomes with high prevalence (closer to 0.5) the model-based prediction of non-response bias is a useful tool, however, for small (