Insights: Using Mixture Models to Examine Group Differences Among Jurors

INSIGHTS

Using Mixture Models to Examine Group Differences Among Jurors:

An Illustration Involving the Perceived Strength of Forensic Science Evidence

OVERVIEW

It is critically important for jurors to be able to understand forensic evidence,
and just as important to understand how jurors perceive scientific reports.
Researchers have devised a novel approach, using statistical mixture
models, to identify subpopulations that appear to respond differently to
presentations of forensic evidence.

Lead Researchers

Naomi Kaplan-Damary
William C. Thompson
Rebecca Hofstein Grady
Hal S. Stern

Journal

Law, Probability, and Risk

Publication Date

30 January 2021

Publication Number

IN 116 IMPL

Goals

1

Use statistical models to determine if subpopulations exist among samples of mock jurors.

2

Determine if these subpopulations have clear differences in how they perceive forensic evidence.

THE THREE STUDIES

Definition:

Mixture model approach:
a probabilistic model that detects subpopulations within a study population empirically, i.e., without a priori hypotheses about their characteristics.

Results

  • Data from the three studies suggest that subpopulations exist and perceive statements differently.
  • The mixture model approach found subpopulation structures not detected by the hypothesis-driven approach.
  • One of the three studies found participants with higher numeracy tended to respond more strongly to statistical statements, while those with lower numeracy preferred more categorical statements.

higher numeracy

lower numeracy

Focus on the future

 

The existence of group differences in how evidence is perceived suggests that forensic experts need to present their findings in multiple ways. This would better address the full range of potential jurors.

These studies were limited due to relatively small number of participants. A larger study population may allow us to learn more about the nature of population heterogeneity.

In future studies, Kaplan-Damary et al. recommend a greater number of participants and the consideration of a greater number of personal characteristics.