Pop-up workshop: Bias in Social Institutions

1200-1215 – Informal milling around with words of introduction from Prof. Gerit Pfuhl (Psychology, IPS) and freshly made waffles

1215-1300 – Michael Morreau, UiT: “Voting, deliberation and bias in expert panels”

Voting and deliberation are different methods for resolving disagreements among members of juries, panels and committees. They have, in general, different outcomes. In a case study of avalanche-danger panels in Switzerland, patterns of disagreement among forecasters are identified under which one or the other method tends to produce more-accurate outcomes. An observed “deliberation-voting reversal” is explained as the consequence of biases, common to forecasters, that push individual errors in the same direction. Likely implications are spelled out for the procedures of panels for policy evaluation, grant funding and shortlisting candidates in a job search.  

1315-1400 – Joanna Kreft, UiT: “Mandevillian intelligence: the role of epistemic trust, competence and bias in discovery”

It might seem only proper for us to trust the testimony of equally competent informants to the same extent. I demonstrate using a computational study that sometimes it is beneficial from an epistemic point of view that epistemic trust does not track competence. That is, it can be better that someone trusts different people to different degrees, even though they are equally deserving of trust. This result might be found surprising, even dismaying.  Properly understood, it is neither. The underlying reason for it is that epistemic distrust can maintain cognitive diversity within the group, and in so doing facilitate scientific discovery. It is an example of Mandevillian intelligence: individual vice (here, unjustified epistemic distrust) brings a collective advantage (here, a better chance of discovery).

Coffee & Tea & Cake and freshly made waffles

1415-1500 – Zeynep Burçe Gümüşlü, LMU, Munich: “Barriers to valid meta-analyses in the behavioural sciences”

Publication bias and the prevalence of underpowered studies are important sources of bias in meta-analytic estimates. I argue in this talk that available countermeasures are inadequate in the behavioral sciences today. I draw three conclusions: First, improving the validity of meta-analysis requires improving the object-level studies. Second, object-level studies with sufficient sample size and statistical power are needed for valid meta-analyses. Third, the received view that all available data must be included in a meta-analysis is untenable in the behavioral sciences. Considering these, the optimal way to conduct a meta-analysis in these fields is by compiling evidence only from large-N object-level studies. The optimal is not good enough currently in the behavioral sciences, as the number of such studies is scant. This can only be addressed by taking sample size and statistical power much more seriously as the criteria for research quality, both in the design and review processes. This suggestion must be embraced as a necessary step to increase the research quality and cumulative knowledge in the behavioral sciences by friends and foes of meta-analysis alike.

1515-1600 – Berna Kilinc, Bogazici University, Istanbul: “Debiasing judgment aggregation”

Oftentimes, truth escapes a single detector, be it the informed judgment of a person, a measurement device or an experiment.  Thus are measurements repeated, observers multiplied and experiments rerun when the warrant for accepting a statement requires some decent standards.  Not only in everyday contexts but also in scientific practice, a proposition may have as its sole warrant for belief the information gathered from several fallible detectors rather than one or two nearly unfailing ones.  In order to model these situations, I derive a generalized version of the classical Condorcet Jury Theorem, relaxing the competence conditions on single detectors, while still securing unbiased estimations of the aggregate in the long run.  I use this theorem to examine first a real project in earthquake detection and then the wide-ranging practices of meta-analysis. 

1600-1700 – Open ended and freewheeling discussion


Når: 29.11.22 kl 12.00–17.00
Hvor: Møterom Innsikten (TEO-H1 1.524)
Sted: Tromsø
Målgruppe: Ansatte, Studenter, Gjester / eksterne
Ansvarlig: Gerit Pfuhl
Legg i kalender