Defense Date: 2026/18/11
Student

ABDULLAH ADNAN SAYAH

Department / Program Science | Statistics

Smoothing spline density estimation from doubly truncated data

Supervisor Soliman KHazaei

Abstract

In many scientific fields, such as astronomy, survival analysis, and epidemiology, doubly truncated data are frequently encountered. In this type of data, the variable of interest is observable only if it falls between two random bounds (such as \\(U\\) and \\(V\\)). This sampling mechanism naturally introduces bias into the data, and consequently, conventional density estimators designed for complete data provide biased and inconsistent estimates of the true distribution.The present study aims to propose a nonparametric method for estimating the density function in the presence of doubly truncated data, based on smoothing splines. In this method, by employing the logistic transformation in the form of\\[ f(x)=\\frac{e^{\\eta(x)}}{\\int e^{\\eta(x)}\\,dx} \\]which automatically satisfies the intrinsic properties of a density function (non-negativity and integration to one), the problem is transformed into the minimization of a penalized likelihood function in a reproducing kernel Hilbert space (RKHS). This approach makes it possible to adjust for the bias induced by double truncation through the penalty term.The proposed method is validated through Monte Carlo simulations under various scenarios, and its performance is compared with that of the kernel density estimator (KDE). Furthermore, the practical effectiveness of the method is demonstrated through the analysis of two real-world datasets from the medical and industrial fields. The results indicate that the smoothing spline approach provides smoother, more accurate, and less biased estimates compared with the kernel-based method and can be employed as an efficient tool for analyzing incomplete data with a doubly truncated structure.