TY - JOUR
T1 - Deviation from Perfect Performance Measures the Diagnostic Utility of Eyewitness Lineups but Partial Area Under the ROC Curve Does Not
AU - Smith, Andrew M.
AU - Lampinen, James Michael
AU - Wells, Gary L.
AU - Smalarz, Laura
AU - Mackovichova, Simona
N1 - Funding Information:
This research was supported by a grant awarded to Iowa State University, from the Laura and John Arnold Foundation, 400-61-12C, and by a grant awarded to AMS from the Natural Sciences and Engineering Research Council of Canada, RGPIN-2018-05336. The views and opinions expressed in this report are those of the authors and do not necessarily reflect the views of the funders.
Publisher Copyright:
© 2018 Society for Applied Research in Memory and Cognition
PY - 2019/3
Y1 - 2019/3
N2 - When one lineup identification procedure leads to both fewer innocent–suspect identifications and fewer culprit identifications than does some other lineup procedure, it is difficult to determine whether the procedures differ in diagnostic accuracy. In an influential article, Wixted and Mickes (2012) argued that measures of probative value do not inform diagnostic accuracy in these situations but that the partial area under the receiver operator characteristic curve (pAUC) does. In more recent research, we have found that pAUC does not necessarily indicate which of two lineup procedures has higher expected utility. When two lineup procedures produce different innocent-suspect identification rates, it leads to differential truncation of the ROC curves. As a result, diagnostic utility as measured by the pAUC is confounded with witness confidence level. We introduce a novel receiver operator characteristic measure, deviation from perfect performance (DPP), that unconfounds diagnostic utility and witness confidence level and consistently indicates which of two lineup procedures has higher expected utility. Our findings suggest that eyewitness scientists should abandon pAUC as a measure of diagnostic accuracy and embrace deviation from perfect performance.
AB - When one lineup identification procedure leads to both fewer innocent–suspect identifications and fewer culprit identifications than does some other lineup procedure, it is difficult to determine whether the procedures differ in diagnostic accuracy. In an influential article, Wixted and Mickes (2012) argued that measures of probative value do not inform diagnostic accuracy in these situations but that the partial area under the receiver operator characteristic curve (pAUC) does. In more recent research, we have found that pAUC does not necessarily indicate which of two lineup procedures has higher expected utility. When two lineup procedures produce different innocent-suspect identification rates, it leads to differential truncation of the ROC curves. As a result, diagnostic utility as measured by the pAUC is confounded with witness confidence level. We introduce a novel receiver operator characteristic measure, deviation from perfect performance (DPP), that unconfounds diagnostic utility and witness confidence level and consistently indicates which of two lineup procedures has higher expected utility. Our findings suggest that eyewitness scientists should abandon pAUC as a measure of diagnostic accuracy and embrace deviation from perfect performance.
KW - Area under the curve
KW - Deviation from perfect performance
KW - Diagnostic accuracy
KW - Eyewitness science
KW - ROC analysis
UR - http://www.scopus.com/inward/record.url?scp=85054724894&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=85054724894&partnerID=8YFLogxK
U2 - 10.1016/j.jarmac.2018.09.003
DO - 10.1016/j.jarmac.2018.09.003
M3 - Review article
AN - SCOPUS:85054724894
SN - 2211-3681
VL - 8
SP - 50
EP - 59
JO - Journal of Applied Research in Memory and Cognition
JF - Journal of Applied Research in Memory and Cognition
IS - 1
ER -