Collections > UNC Scholarly Publications > National Longitudinal Study of Adolescent to Adult Health (Add Health) > Article > Effects of Nonresponse on the Mean Squared Error of Estimates from a Longitudinal Study

Research in this study focuses on two related aspects of unit nonresponse (nonresponse by sampled members of study populations) in the rounds of the National Longitudinal Study of Adolescent Health (Add Health) (Chantala and Tabor, 1999): (i) round-specific nonresponse bias and its component contributions, and (ii) the statistical utility of alternative approaches to adjusting sample weights for nonresponse. This work is part of four research studies funded by CDC-NCHS, at the UNC Center for Health Statistics Research. Nonrespondents in surveys can be classified according to the reason for nonresponse (Lessler and Kalsbeek, 1992): 1) Not Solicited (NS): Sample members are not solicited as perhaps their address is unknown, or they are out of the country; 2) Solicited but Unable (SUA): Sample members are contacted but decline to participate based on inability. Reasons include physical or language limitations; 3) Solicited but Unwilling (SUW): Sample members are contacted but refuse to participate for reasons such as lack of time or, apathy; and 4) Other Nonrespondents (OTH): Sample nonrespondents give a reason that does not fit in any of the previous categories. Examples are lost schedules and partial respondents Response outcome information and data to obtain 13 different measures of health risk from Add Health are used to accomplish two main tasks in this study. First, we estimate the round-specific nonresponse bias and its component contributions corresponding to the four nonresponse categories described. The sign (negative or positive) of these components and the offsetting effects of some components on the overall bias is of particular interest. Second, we compare the statistical effects of alternate sample adjustments for nonresponse on the bias and variance of study estimates. It is important to note here that we are examining the effects of nonresponse in IH1 and IH2 separately, and not the cumulative effects of nonresponse through these rounds.