Externally funded project

Training Study on Bayesian Reasoning (TraBaD) (TrainBayes)

Project Details
Project duration: 06/202011/2022


many professions, experts have to make decisions of a probabilistic nature on
the basis of uncertain evidence. Medicine and law are two important domains where
this kind of decision-making is part of a professional's daily routine. Many decisions
made by medical doctors or judges in a courtroom can be mathematically
justified with the help of Bayes’ Theorem,
a central formula in probability theory. This formula provides a model for an
altered probability estimation with regard to a certain hypothesis, for example
about the condition of a patient or the guilt of a defendant, after new data
(e.g., medical test results or legal evidence) has been presented.

Bayesian reasoning in this sense of updating hypotheses on the basis of
relevant data can be very difficult, not only for laypeople but also for medical
and legal experts, who often have to make momentous decisions. Misjudgments can
have tragic consequences, many instances of which have been documented.

to the compelling relevance of Bayesian reasoning, many studies in the fields
of psychology and mathematics education have investigated potentially
successful strategies supporting Bayesian reasoning. These studies indicate
that two strategies—representing statistical information in the form of natural
frequencies and certain visualizations of statistical information—have proven successful
for Bayesian reasoning.

Therefore, this project
aims at systematically combining the already developed promising strategies,
namely the aforementioned natural frequencies and visualizations, and at
investigating them experimentally in a training for students of medicine and
the law. In this study, an experimental design is chosen, in which two optimized
trainings based on natural frequencies and two different visualizations (unit
square and double tree diagram) are compared in a pre-post-follow-up-design to
control conditions, such as a traditional probability training (following
standard statistics textbooks) or a pure frequency training (without any
visualization). The comparison is based on people’s performance, their ability
to understand the influence of changed parameters in a Bayesian situation
(covariation) and people’s ability to communicate the results. Hence,
conditions for successful Bayesian inferences, both within and across domains,
are to be derived, in the sense of an investigation oriented towards basic
research. The innovation in our approach is to examine not only performance but
also the assessment of effects of parameters changes (covariation) as well as
the appropriate communication of a re-evaluated hypothesis in the light of new
information as aspects of Bayesian reasoning. A further innovative approach is
to analyse not only problem
solutions (products), but also cognitive strategies (processes) that were
documented by non-Bayesian tasks, failure analyses and write-aloud protocols.

Last updated on 2020-25-05 at 11:19