Methodological Extensions of Innovative Outlier Detection

Google map

This presentation focuses on the diagnostic procedure for detecting outliers in state-space models developed by You et al. (2020). Specifically, we focus on the detection of innovative outliers in the state equation, implemented through the dynr.taste function in the R package dynr. The main contribution lies in testing this technique in continuous-time models using developmental data, which differ substantially from the data used in the original study. Developmental datasets typically display trends and contain considerably fewer measurement occasions. We introduce novel modeling approaches to enhance outlier detection under these challenging conditions. Furthermore, we discuss the statistical measures employed for detection and propose new perspectives on these statistics aimed at maximizing true outlier detection while minimizing false positive detection.