Organizers
Emmanouil Konstantinidis
University College London
Contact details:
Department of Cognitive, Perceptual & Brain Sciences
Room 204c, 26 Bedford Way, London, WC1H 0AP
UK
Telephone: (+44) 020 7679 5364
E-mail: emmanouil.konstantinidis.09@ucl.ac.uk
Neil Bramley
University College London
Contact details:
Department of Cognitive, Perceptual & Brain Sciences
Room 201, 26 Bedford Way, London, WC1H 0AP
UK
E-mail: neil.bramley.10@ucl.ac.uk
LJDM website http://www.ljdm.info
Web administrator:
Dr Stian Reimers ( Stian.Reimers.1@city.ac.uk
)
LJDM members’ (Risk & Decision) list
Contact: Dr Marianne Promberger ( marianne.promberger@kcl.ac.uk
)
15 th
January
22 nd
January
29 th
January
05 th
February
12 th
February
Performance of healthy participants on the Iowa Gambling Task and a comparison of Reinforcement-Learning Models
Helen Steingroever
University of Amsterdam
Relative Rank Theory and Valuation
Gordon Brown
University of Warwick
The Importance of the Probability of Losing in Repeated Decisions
Stefan Zeisberger
University of Zurich
Reason-Based Rationalisation
Christian List
London School of Economics
The value of irrationality: violations of regularity and transitivity mirror statistically optimal brain processes
Konstantinos Tsetsos
University of Oxford
19 th
February
26 th
February
05 th
March
NO SEMINAR – UCL READING WEEK
Something on Darwinian neurodynamics
Chrisantha Fernando,
Queen Mary University
A unified framework for analysing and improving individual and group judgement
Henrik Olsson
University of Warwick
12 th
March
19 th
March
Causal-knowledge and Information Search During Categorization
Bob Rehder
New York University
Knowledge sharing as social dilemma: Status and feedback moderate expert contributions
Karin Moser
Roehampton University
26 th
March Bayesian sampling to cluster and explore
David Leslie
University of Bristol
15.01.2014
Helen Steingroever
University of Amsterdam
Performance of Healthy Participants on the Iowa Gambling Task and a Comparison of
Reinforcement-Learning Models
Decision-making deficits in clinical populations are often studied using the Iowa gambling task
(IGT). A review of healthy participants’ performance on this task is the focus of the first part of my talk. I illustrate that –in contrast to the assumptions underlying the IGT– healthy participants
(1) do not learn to prefer the good decks; (2) do not show a systematic decrease in the number of switches across trials; and (3) show idiosyncratic choice behavior. These findings question the prevailing interpretation of IGT data. The second part of my talk focuses on reinforcementlearning models that aim to decompose IGT performance in its constituent psychological processes. I compare the absolute model performance of the Expectancy Valence (EV) model, the
Prospect Valence Learning (PVL) model, and a hybrid version of both models –the PVL-Delta model– using two different methods. These methods assess (1) whether a model provides an acceptable fit to an observed choice pattern, and (2) whether the parameters obtained from model fitting can be used to generate the observed choice pattern. I show that all models provide an acceptable fit to two data sets; however, when the model parameters were used to generate choices, only the PVL-Delta model captures the qualitative patterns in the data. Thus, a model’s ability to fit a particular choice pattern does not guarantee that the model can also generate that same choice pattern.
22.01.2014
Gordon Brown
University of Warwick
Relative Rank Theory and Valuation
How do people value states of health, decide on a fair price for a product, or determine the appropriate amount of damages to award against a polluting company? Here I describe and discuss a process I refer to as “relative rank matching”. The subjective magnitude of quantities such as prices, health states, or crimes are assumed to be determined by contextual comparison involving rank-based principles such as those embodied in Range Frequency Theory and Decision by Sampling. However such models are often silent on the question of how comparisons are made across incommensurable dimensions. Subjective judgements are assumed to be entirely relative, yet we have no difficulty rejecting a “relatively good” bottle of wine in favour of a “relatively bad” house. Although relative judgements cannot themselves provide a common currency for comparing options, it is suggested that relativity-matching translation into a common distribution
(such as market prices) is often possible and enables comparison across different dimensions.
When a suitable matchable dimension such as a market price distribution is unavailable, however, our valuations are inconsistent and unreliable.
29.01.2014
Stefan Zeisberger
University of Zurich
The Importance of the Probability of Losing in Repeated Decisions
In a series of experiments we demonstrate that the overall probability of losing is a very important criterion for the valuation of risky prospects. Our results contradict mean-variance and
(Cumulative) Prospect Theory predictions. We significantly extend previous research as our dynamic experimental setting is less complex and additionally allows for learning through feedback. Furthermore, this is the first study to compare decisions under risk and ambiguity while analysing the importance of the overall probability of losing.
Paper download link: http://ssrn.com/abstract=2169394
05.02.2014
Christian List
London School of Economics
Reason-Based Rationalisation
We introduce a “reason-based” way of rationalizing an agent’s choice behaviour, which explains choices in terms of “motivationally salient” properties of the options and/or the choice context, thereby explicitly modelling the agent’s conceptualization of any choice problem. Reason-based rationalizations can explain non-classical choice behaviour, including boundedly rational and sophisticated rational behaviour, and predict choices in unobserved contexts, an issue neglected in standard choice theory. We characterize the behavioural implications of different reason-based models and distinguish two kinds of context-dependent motivation: “context-variant” motivation, where different choice contexts make different properties motivationally salient, and “contextregarding” motivation, where the agent cares not only about properties of the options, but also about properties relating to the choice context.
12.02.2014
Konstantinos Tsetsos
University of Oxford
The value of irrationality: violations of regularity and transitivity mirror statistically optimal brain processes
Years of research within the cognitive and decision sciences has converged to the idea that human decisions are systematically at odds with rational choice theory and utility maximisation. This thesis, although empirically well-supported, is somewhat paradoxical: why did humans evolve to be irrational and thus suboptimal? To address this question, it is necessary to understand the form and scope of the cognitive processes that promote irrational decisions. Unveiling the processes of higher cognitive functions, such as economical decision-making, is particularly challenging because the flow of input information is rarely under experimental control. As a result the inputoutput function is vague and mechanistic decision theories remain unconstrained and hardly
testable. To circumvent this problem, and inspired by research in perception and visual psychophysics, I will present a novel paradigm (termed “value psychophysics”) that abstracts complex decision problems into simple, well-controlled, information-integration experiments. I will show that classical paradoxes such as violations of regularity and transitivity as well as framing, risk and loss-aversion biases can be obtained in this simple, psychophysical task.
Surprisingly a single mechanism, based on selective attention towards the valuable samples of incoming information, underlies these phenomena. I will demonstrate that this empiricallyestablished selective mechanism outperforms the statistically optimal (and rational) choice algorithm in terms of discrimination accuracy under the assumption that moderate levels of uncorrelated (e.g. cortical) noise corrupt the integration process. I will conclude that violations of rationality reflect and neurally-constrained, optimal choice algorithms.
19.02.2014
UCL Reading Week
26.02.2014
Chrisantha Fernando
Queen Mary University
TBA – (something on Darwinian neurodynamics)
05.03.2014
Henrik Olsson
University of Warwick
A unified framework for analyzing and improving individual and group judgment
What determines the performance of group judgment, and can it be improved? After decades of research, the social sciences still lack an integrative theoretical framework that could help answering these questions. I propose that connecting group decision making research with insights from machine learning and statistical theory could help us develop such a framework and answer these questions. The framework, based on decomposing the prediction error of group judgment into bias, variance, and covariance, makes it possible to determine when and why groups perform better than individuals and to devise new ways of improving the accuracy of group judgment. I
use computer simulations to estimate the bias-variance profiles of cue- and exemplar-based strategies and the bias-variance-covariance profiles for groups of these strategies. The results show that exemplar-based strategies benefit the most from averaging, due to lower bias and higher variance compared to cue-based strategies. The insights from the bias-variance-covariance framework tell us that the success of group judgment depends on which strategies the individual members are using (e.g., low or high variance strategies), what information they are focusing on
(e.g., how much overlap there is between the information they are using), and how the individual assessments are used to reach a group decision (e.g., averaging or relying on the perceived best member). Using the bias-variance-covariance framework to recognize and analyze these aspects of group judgments can help us understand what factors contribute to successful group decisions in a variety of real world contexts.
12.03.2014
Bob Rehder
New York University
Causal-knowledge and Information Search During Categorization
This research assessed how causal knowledge influences the order in which classifiers seek information. Undergraduates learned two novel categories. One category’s features exhibited a common cause network (one feature causes two others), and the other exhibited a common effect network (one feature is caused by two others). One neutral dimension did not take part in any causal relations. Participants chose which of two feature dimensions they would like to see in order to classify an object. Participants preferred to query features involved in two causal relations over those involved in one, which in turn were preferred to those involved in none. In addition, when some features of the to-be-classified item were already known, participants chose to query causally-related dimensions. Existing models of causal-based classification failed to account for these results.
19.03.2014
Karin S. Moser
University of Roehampton London
Knowledge sharing as social dilemma: Status and feedback moderate expert contributions
Groups and organisations set cooperative goals for their members, yet in reality some team members contribute more than others towards these goals. Experts, in particular, face a social
dilemma: From the group’s perspective they should share their knowledge, whereas individually they are better off not sharing their knowledge, because acquiring knowledge is costly. Two experiments tested the hypothesis, derived from indirect reciprocity and competitive altruism theory, that experts contribute more if their status is being recognized. In two experiments with different designs (scenario and virtual team simulation) we manipulated expertise and performance feedback and examined the impact on people’s contributions in various informationsharing tasks. As predicted, experts contributed more when feedback was individualized and public, thus ensuring status rewards. In contrast, novices contributed more when performance feedback was collective, regardless of whether it was public or private feedback. Implications for theory and practice on knowledge sharing in teams are discussed.
26.03.2014
David Leslie
University of Bristol
Bayesian sampling to cluster and explore
I present results from two recent papers in which Bayesian ideas are shown to replicate either desirable or observed behaviour for sequential decision-makers. In the first part of the talk I present and discuss an old and inherently Bayesian idea of Thompson (Biometrika 1933) on how to balance exploration and exploitation in a sequential decision-making problem. When faced with a decision opportunity, a single sample should be drawn from the posterior distribution of the value of each available action, and the action with the highest sampled value should be selected.
This ensures that the action with the highest expected value is most likely to be selected, whereas other actions may also be selected, but with probability that decreases with their expected value, and increases with the uncertainty about this value. I will then present a Bayesian clustering model of learning and decision-making suitable for `jumpy but sticky' environments, and show that the model naturally replicates several of the classically paradoxical effects observed in rat decision-making. Joint work with Benedict May and Kevin Lloyd.