Shrinkage in a Bayesian random intercept model with time-varying coefficients
Sprache des Vortragstitels:
Englisch
Original Tagungtitel:
Österreichische Statistiktage 2024
Sprache des Tagungstitel:
Deutsch
Original Kurzfassung:
We consider regression models for panel data, where regression effects are allowed to vary over time. To account for within subject dependence and heterogeneity across subjects, we include a subject specific latent variable with weights that may also vary over time. We use a random walk prior on the regression effects as well as the weights to allow for smoothing over time and hierarchical shrinkage priors on the initial values and variances of these random walks to identify time-invariant and zero regression effects as well as time-invariant and zero weights. We investigate the performance of the shrinkage priors on the regression effects and weights in a simulation study and finally apply our model to the analysis of yearly earnings of Austrian mothers after their return to the labour market following a maternity leave.