Title: Random effects dynamic panel models for unequally-spaced repeated measures
Authors: Fiona Steele - London School of Economics (United Kingdom) [presenting]
Abstract: Dynamic models, also known as autoregressive or lagged response models, are widely used for the analysis of longitudinal data in social science and health. However, standard discrete-time models assume that measurements of the response and time-varying covariates are taken at the same equally-spaced occasions. Unequal spacing is a common feature of longitudinal studies, which may arise by design or because of nonresponse. A general random effects dynamic model is proposed to handle unequally-spaced responses that are measured less frequently than time-varying covariates. The approach is suitable for continuous, binary or ordinal multivariate responses. The methodology is assessed in a simulation study, and applied to bivariate binary data on bidirectional exchanges of support between adult children and their non-coresident parents from the British Household Panel Survey and UK Household Longitudinal Study. Of particular interest are the effects of changes in children's circumstances on help received from and given to their parent(s). Using annual data on partnership and employment status and children, we estimate the effects of partnership and employment transitions between year $t-1$ and $t$ and of the presence and age of children at $t$ on exchanges in each direction at $t$. A bivariate model is used to estimate the reciprocity of exchanges.