Original Research

The applicability of fair selection models in the South African context

G. K. Huysamen
SA Journal of Industrial Psychology | Vol 21, No 3 | a591 | DOI: https://doi.org/10.4102/sajip.v21i3.591 | © 1995 G. K. Huysamen | This work is licensed under CC Attribution 4.0
Submitted: 20 June 1995 | Published: 20 June 1995

About the author(s)

G. K. Huysamen, The University of the Orange Free State, South Africa

Full Text:

PDF (5MB)

Abstract

This article reviews several models that are aimed at achieving fair selection in situations in which underrepresented groups tend to obtain lower scores on selection tests. Whereas predictive bias is a statistical concept that refers to systematic errors in the prediction of individuals' criterion scores, selection fairness pertains to the extent to which selection results meet certain socio-political demands. The regression and equal-risk models adjust for differences in the criterion-on-test regression lines of different groups. The constant ratio, conditional probability and equal probability models manipulate the test cutoff scores of different groups so that certain ratios formed between different selection outcomes (correct acceptances, correct rejections, incorrect acceptances, incorrect rejections) are the same for such groups. The decision-theoretic approach requires that utilities be attached to these different outcomes for different groups. These procedures are not only eminently suited to accommodate calls for affirmative action, but they also serve the cause of transparency.

Opsomming
Hierdie artikel verskaf 'n oorsig van verskeie modelle om billike keuring te verkry in situasies waar onderverteen-woordigende groepe geneig is om swakker op keuringstoetse te vaar. Terwyl voorspellingsydigheid 'n statistiese begrip is wat betrekking het op stelselmatige foute in die voorspelling van individue se kriteriumtellings, het keuringsbillikheid te make met die mate waarin keuringsresultate aan sekere sosiaal-politieke vereistes voldoen. Die regressieen gelyke-risiko-modelle maak aanpassings vir verskille in die kriterium-op-toetsregressielyne van verskillende groepe. Die konstante-verhoudings, voorwaardelike-waarskynlikheids- en gelyke-waarskynlikheidsmodelle manipuleer die toetsafkappunte van verskillende groepe sodat sekere verhoudings wat tussen keuringsresultate (korrekte aanvaardings, verkeerde aanvaardings, korrekte verwerpings, verkeerde verwerpings) gevorm word, vir verskillende groepe dieselfde is. Die besluitnemingsbenadering vereis dat nutwaardes geheg word aan die verskillende gebeurlikhede by verskillende groepe. Nie alleen is hierdie prosedures by uitnemendheid geskik om oproepe tot regstellende aksie te akkommodeer nie, maar hulle voldoen ook aan die eis van deursigtigheid.


Keywords

Fair selection models; South African context

Metrics

Total abstract views: 3663
Total article views: 3025


Crossref Citations

No related citations found.