Open AccessThe Shape of ROC Curves in Shooter Tasks: Implications for Best Practices in Analysis
Author(s) -
Caren M. Rotello,
Laura Jane Kelly,
Evan Heit
Publication year - 2018
Publication title -
collabra. psychology
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 1.444
H-Index - 10
ISSN - 2474-7394
DOI - 10.1525/collabra.171
Subject(s) - identification (biology) , task (project management) , phone , suspect , psychology , receiver operating characteristic , variance (accounting) , computer science , cognitive psychology , statistics , artificial intelligence , mathematics , machine learning , engineering , criminology , linguistics , philosophy , botany , systems engineering , accounting , business , biology
Four experiments addressed the widely studied issue of the association between racial groups and guns, namely shooter bias, as measured in the first-person shooter task or the weapon identification task, in which participants judge whether a suspect has a weapon or some other item such as a phone (Correll, Park, Judd, & Wittenbrink, 2002; Payne, 2001). Previous studies have employed various analyses that make conflicting, and indeed untested, assumptions about the underlying nature of the data: Analyses of variance and model-based analyses assume linear receiver operating characteristics (ROCs) and signal detection (SDT) analyses assume curved ROCs. The present experiments directly investigated the shape of the ROCs for the weapon identification task, demonstrating that they are curved, and that the majority of previous studies are at risk for inclusion of inappropriate analyses, because they assume linear rather than curved ROCs.