Prof. Herbert Hoijtink & Dr Rens van de Schoot
The use of expert data in Bayesian latent growth curve models
My research focuses on the use of informative priors in latent growth curve models, with a small sample size. In social sciences, researchers often experience difficulties collecting enough data and obtaining statistical power, due to small or hard to access target groups or prohibitive costs, resulting in a limited data set. The use of Bayesian statistics with informative priors increases the amount of statistical power. Background knowledge from previous publications can then be translated into the statistical prior distributions. However, for limited data sets, this might not be available, as such data remains usually unpublished. We can overcome these obstacles by using background knowledge from experts, such as researchers, clinicians, and experts-by-experience. The background knowledge from these experts can then be used to construct the statistical prior distribution which can then be updated with the data.
The overall objective of my PhD-project is to study the use of expert data in latent growth curve models, and to develop guidelines for researchers who suffer from limited data. I am using simulation studies to investigate how much expert data is needed to compensate for the small sample size, which parameters in the model need informative priors, and how informative these priors should be. I will also investigate how expert data can be translated into statistical distributions, and what the effect is of disagreeing experts. Furthermore, a systematic review will be conducted to give an overview of papers in which simulation studies are used to investigate the performance of Bayesian estimation, when the sample size is small.
NWO – Vidi grant Dr. Rens van de Schoot
1 January 2016 – 31 December 2019