Zaznavanje diskriminacije v dvostopenjskih odločitvenih procesih : delo diplomskega seminarja

Pristovnik, Jan (author), Smrekar, Jaka (mentor)

Težave dosedaj uporabljenih metod pri zaznavanju diskriminacije v dvostopenjskih odločitvenih procesih obsegajo nekonsistentnost rezultatov ob uporabi različnih testov na istih oziroma podobnih podatkih, visoko medsebojno koreliranost uporabljenih testov, problem multikoreliranosti ter ne optimalno aplikacijo specifičnih testov. Moč sedajšnih testov je močno odvisna od vzorcev, na katere jih apliciramo. Tekom diplome skonstruiramo PW test ter predstavimo Pt test. Testa ohranjata velikost na željeni $\alpha$ stopnji, imata visoko moč in konsistentne rezultate. Simulacijska raziskava pokaže, da sta primerna izbira, kadar se standardna odklona razlikujeta.

• Language: Slovenian
• Keywords: diskriminacija, dvostopenjski statistični test, Brunner-Munzel modificirani Wilcoxonov test, linearna ranžirna statistika
• Work type: Final seminar paper
• Typology: 2.11 - Undergraduate Thesis
• Organization: FMF - Faculty of Mathematics and Physics
• Year: 2019
• PID: 20.500.12556/RUL-110002
• UDC: 519.2
• COBISS.SI-ID: 18740313
• Publication date in RUL: 11.09.2019

PRISTOVNIK, Jan, 2019, Zaznavanje diskriminacije v dvostopenjskih odločitvenih procesih : delo diplomskega seminarja [online]. Bachelor’s thesis. [Accessed 17 May 2025]. Retrieved from: https://repozitorij.uni-lj.si/IzpisGradiva.php?lang=eng&id=110002

Secondary language

• Language: English
• Title: Detecting disparate impact arising from two-stage selection processes

Abstract

Problems of statistical methods used so far in detecting disparate impact arising from two-stage selection processes include inconsistency of results while using different tests on the same or similar data, high correlation between used tests, multiple comparison problem and not optimal application of specific tests. Power of current tests strongly depends on the samples, on which they are applied. During bachelor's thesis we construct PW test and present Pt test. They keep their size at the nominal $\alpha$ level, have high power and consistent results. Simulation study results indicate that PW and Pt test are appropriate choice when the standard deviations are different.

• Keywords: discrimination, two-step statistical test, Brunner-Munzel modified Wilcoxon test, linear rank statistics