Nidhi Kohli, associate professor in the Department of Educational Psychology’s quantitative methods in education program, and her colleagues recently published an article, “Detecting multiple random changepoints in Bayesian piecewise growth mixture models,” in Psychometrika. The article highlights a piecewise growth mixture model Kohli and her colleagues developed using a Bayesian inference approach that allows the estimation of multiple random changepoints (knots) within each latent class and develops a procedure to empirically detect the number of random changepoints within each class.
The study makes a significant advancement to Kohli’s existing research program in piecewise growth models. In all of the previous methods and applied substantive studies, researchers hypothesized and prefixed the number of unknown changepoint locations (i.e., the number of changepoints were specified in advance). There is no existing methodological study that empirically detects the number of changepoints (i.e., considers the number of changepoints as unknown and to be inferred from the data) within a unified framework for inference. This is limiting for many applications.
Piecewise studies of educational data typically assume one changepoint (Sullivan et al. 2017; Kohli et al. 2015b; Kieffer 2012). However, it is plausible that many learning trajectories will have at least two changepoints: one preceding a period of accelerated growth (an “a-ha” moment) and another preceding a period of decelerated growth (a “saturation point”) (Gallistel et al. 2004). Multiple changepoints are also plausible for many physical growth processes. For these and other applications a flexible inferential framework that allows for an arbitrary number of latent changepoints, as well as individual variation and population heterogeneity in the form of latent classes, is needed. This method fulfills that need.
The article includes a user friendly R package that makes easy for researchers and practitioners to apply this method to their data sets.