When dimensionality matters: revisiting local polynomial quantile estimators through simulation and real-data evidence
Communications in Statistics: Simulation and Computation, 2026 (SCI-Expanded, Scopus)
- Yayın Türü: Makale / Tam Makale
- Basım Tarihi: 2026
- Doi Numarası: 10.1080/03610918.2026.2716275
- Dergi Adı: Communications in Statistics: Simulation and Computation
- Derginin Tarandığı İndeksler: Science Citation Index Expanded (SCI-EXPANDED), Scopus, Applied Science & Technology Source, Compendex, MathSciNet, zbMATH, Academic Search Ultimate (EBSCO), Business Source Ultimate (EBSCO), Materials Science & Engineering Collection (ProQuest), Technology Collection (ProQuest)
- Anahtar Kelimeler: Finite-sample performance, Local polynomial smoothing, Mincer wage equation, Nonparametric estimation, Quantile regression
- Van Yüzüncü Yıl Üniversitesi Adresli: Evet
Özet
This study compares local polynomial quantile estimators (LLQR, NPQ, SWH, and DIRECT) using two- and three-dimensional simulation designs. The estimators are evaluated under normal and chi-square error distributions, across various sample sizes and quantile levels. In two-dimensional models, NPQ provides the most consistent results at both the median and tail quantiles. While SWH performs comparably at the median, its efficacy diminishes in the tails due to over-smoothing. LLQR shows low variance but falls behind because of its high bias. In three-dimensional models, LLQR substantially reduces its bias and becomes the most effective estimator. NPQ remains a strong alternative, while SWH fails to adapt to higher dimensionality. The robustness analyses further confirm that the main qualitative findings remain stable across alternative simulation settings, including boundary regions, higher quantile levels, and correlated covariate structures, although some variation in the magnitude of performance measures is observed. A real-data application based on the Mincer wage equation corroborates these findings. Overall, the results indicate that the choice of estimator depends on model dimensionality and quantile level: NPQ is recommended for lower-dimensional models, while LLQR is superior in higher-dimensional settings.