Unwarranted inferences from statistical mediation tests

• Verfasst von: Fiedler, Klaus [VerfasserIn]
• Verfasst von: Harris, Christoph Alexander [VerfasserIn]
• Verfasst von: Schott, Malte [VerfasserIn]
• Titel: Unwarranted inferences from statistical mediation tests
• Titelzusatz: an analysis of articles published in 2015
• Verf.angabe: Klaus Fiedler, Chris Harris, Malte Schott
• Jahr: 2018
• Jahr des Originals: 2017
• Umfang: 8 S.
• Fussnoten: Available online 8 December 2017 ; Gesehen am 16.04.2020
• Titel Quelle: Enthalten in: Journal of experimental social psychology
• Ort Quelle: Amsterdam [u.a.] : Elsevier, 1965
• Jahr Quelle: 2018
• Band/Heft Quelle: 75(2018), Seite 95-102
• ISSN Quelle: 0022-1031
• DOI: doi:10.1016/j.jesp.2017.11.008
• Datenträger: Online-Ressource
• Sprache: eng
• K10plus-PPN: 1694704149

Abstract

Recent attempts to improve on the quality of psychological research focus on good practices required for statistical significance testing. The scrutiny of theoretical reasoning, though superordinate, is largely neglected, as exemplified here in a common misunderstanding of mediation analysis. Although a test of a mediation model X➔Z➔Y is conditional on the premise that the model applies, alternative mediators Z′, Z″, Z‴ etc. remain untested, and other causal models could underlie the correlation between X, Y, Z, researchers infer from a single significant mediation test that they have identified the true mediator. A literature search of all mediation analyses published in 2015 in Sciencedirect shows that the vast majority of studies neither consider alternative causal models nor alternative mediator candidates. Ignoring that mediation analysis is conditional on the truth of the focal mediation model, they pretend to have demonstrated that Z mediates the influence of X on Y. Recommendations are provided for how to overcome this dissatisfying state of affairs.