A1284
Title: A circular linear interval-valued regression model
Authors: Najmeh Nakhaeirad - University of Pretoria (South Africa) [presenting]
Thasmika Mohan - University of Pretoria (South Africa)
Sphiwe Skhosana - University of Pretoria (South Africa)
S Yaser Samadi - Southern Illinois University Carbondale (United States)
Abstract: Many real world data are naturally recorded as intervals, yet most standard statistical methods are developed for single valued observations. In practice, interval-valued data are often reduced to their midpoint or mean, which can lead to loss of important information. This issue is particularly critical for circular data, where interval-valued observations are more appropriately represented as arcs on the circle and require methods that respect the underlying circular structure. A circular linear regression framework for interval-valued circular data is proposed that jointly addresses periodicity and measurement uncertainty. The model allows for a circular interval-valued response while incorporating both circular and linear covariates that may themselves also be interval-valued. A maximum likelihood estimation procedure is developed to preserve the internal distributional characteristics of the observed intervals rather than relying on aggregated summaries. The proposed method is evaluated through a simulation study and applied to wind farm data from South Africa to model variability in wind direction. In both the simulated and real data settings, the proposed approach demonstrates superior performance compared with the classical centre range method, highlighting its potential as a more effective tool for analysing interval valued circular data.
