We investigate the problem of determining changepoints in segmented regression. Segmented regression, composed of several continuous linear phases, is often used to describe changes in trends and to model non-linear associations. While several methods exist for implementing segmented regression, their comparative performance remains largely unexplored, leaving gaps in understanding their relative strengths and applications.
This thesis evaluates methods for estimating single or multiple changepoints in regression models, proposing novel approach that are assessed through simulation studies and illustrated with comparisons using simulated datasets.
This study extends changepoint detection methods in segmented regression to beta and quantile regression models. Furthermore, the study used insights gained from simulation analasyes to real-world data from the SLOfit database, examining trends in the physical fitness of Slovenian children from 1989 to 2019. The use of quantile regression reveals changes across fitness percentiles, while beta regression acts as a sensitivity check. Segmented regression identifies structural changes in physical fitness trend with potental to offering data-driven insights that could support policymaking in public health.
|
