De-Biasing Role Induced Bias Using Bayesian Networks

Mark Schweizer, De-Biasing Role Induced Bias Using Bayesian Networks, Law, Probability and Risk, Volume 18, Issue 4, December 2019, Pages 255–273, Doi.org/10.1093/lpr/mgz015 (revised version)

22 Pages Posted: 27 Aug 2018 Last revised: 3 Feb 2020

Date Written: April 26, 2018

Abstract

The merits of using subjective probability theory as a normative standard for evidence evaluation by legal fact-finders have been hotly debated for decades. Critics argue that formal mathematical models only lead to an apparent precision that obfuscates the ad-hoc nature of the many assumptions that underlie the model. Proponents of using subjective probability theory as normative standard for legal decision makers, specifically proponents of using Bayesian networks as decision aids in complex evaluations of evidence, must show that formal models have tangible benefits over the more natural, holistic assessment of evidence by explanatory coherence. This paper demonstrates that the assessment of evidence using a Bayesian network parametrized with values obtained from the decision makers greatly reduces role-induced bias, a bias that has been largely resistant to de-biasing attempts so far. This shows that using Bayesian networks as decision aids can benefit legal decision making.

Keywords: evidence law, bayesian network, role-induced bias

JEL Classification: K41

Suggested Citation

Schweizer, Mark, De-Biasing Role Induced Bias Using Bayesian Networks (April 26, 2018). Mark Schweizer, De-Biasing Role Induced Bias Using Bayesian Networks, Law, Probability and Risk, Volume 18, Issue 4, December 2019, Pages 255–273, Doi.org/10.1093/lpr/mgz015 (revised version). Available at SSRN: https://ssrn.com/abstract=3231859 or http://dx.doi.org/10.2139/ssrn.3231859

Mark Schweizer (Contact Author)

Swiss Federal Patent Court ( email )

St. Leonhardt-Strasse 49
St. Gallen, 9023
Switzerland

HOME PAGE: http://www.bundespatentgericht.ch

Here is the Coronavirus
related research on SSRN

Paper statistics

Downloads
41
Abstract Views
294
PlumX Metrics