Evaluation of the Performance of Eight Record-Extension Techniques Under Different Levels of Association, Presence of Outliers and Different Sizes of Concurrent Records: A Monte Carlo Study

Abstract

In this study, eight record extension techniques were described, their properties were explored, and their performances were examined and compared. The eight record extension techniques are the Ordinary least-squares regression (OLS), Kendall-Theil robust line (KTRL), Maintenance of variance extension techniques (MOVE1, MOVE2, MOVE3, and MOVE4), and two recently developed techniques, the modified KTRL (KTRL2), and the robust line of organic correlation (RLOC). The two recently proposed techniques have the advantages of being robust in the presence of outliers, and they are able to maintain the variance in the extended records. Monte-Carlo experiments were conducted to evaluate the performance of the eight record extension techniques under different levels of data contamination (presence of outliers), different sizes of concurrent records, and different levels of association. The performance of the eight techniques was examined for bias, and standard error of: individual records, moment estimates, and the full range of percentiles. Results showed that for individual record estimates, with uncontaminated data (no outliers), OLS, MOVE3 and MOVE4 provide comparable results and outperform the other techniques under any level of association or size of concurrent records. However, under any level of data contamination, the KTRL outperforms other techniques, for the estimation of individual records, under any level of association or size of concurrent records. In the case of uncontaminated data, MOVE techniques, RLOC and KTRL2 were almost similar and outperform OLS and KTRL in preserving the characteristics of the entire distribution, while MOVE techniques were slightly more precise than RLOC and KTRL2 for small sizes of concurrent records. However, under any level of contaminated data, RLOC and KTRL2 outperform other techniques in preserving the characteristics of the entire distribution, under any level of association and/ or size of concurrent records.