Contemporary Cognitive Approaches to Decision-Making
Cogsci 2018, Madison, WI

Organized by Daniel Bartels (University of Chicago), Oleg Urminsky (University of Chicago),
Todd Gureckis (NYU), and Jennifer Trueblood (Vanderbilt University)

Wednesday, July 25

Monona Terrace Convention Center, Level 4, Lecture Hall

Start End Speaker
9:00 AM 9:01 AM


9:01 AM 9:30 AM

Jennifer Trueblood (Vanderbilt University) - computational models of dual process theory

9:30 AM 10:00 AM

Jared Hotaling (University of New South Wales) - the effects of outcome information during sampling on decisions from experience

10:00 AM 10:30 AM

Falk Lieder (Max Planck Institute for Intelligent Systems) - an integrative framework for modeling and improving people's decision strategies

10:30 AM 10:45 AM COFFEE AND TEA
10:45 AM 11:15 AM

Abby Sussman (University of Chicago) - causal scope and judgment

11:15 AM 11:45 AM

Danny Oppenheimer (Carnegie Mellon University) - decision making in nonlinear environments

11:45 AM 1:15 PM LUNCH
1:15 PM 1:45 PM

Oleg Urminsky (University of Chicago) - what factors make decisions seem more similar or different

1:45 PM 2:15 PM

Hang Zhang (Peking University) - a decision-theorectic model for the temporal dynamics of visual priming effects

2:15 PM 2:45 PM

Nick Reinholtz (University of Colorado) - reasoning about numeric distributions

3:00 PM 3:30 PM Discussion -- Bring topics and burning questions... We could talk about this video, for example.
3:30 PM 4:00 PM

Paula Parpart (NYU) - reinterpreting heuristics as Bayesian inference

4:00 PM 4:30 PM

Neil Stewart (University of Warwick) - how machine-recorded data can be used to study cognitive questions


Presentation Abstracts


Many phenomena in judgment and decision-making are often attributed to the interaction of two systems of reasoning. While these so-called dual process theories can explain many types of behavior, they are rarely formalized as mathematical or computational models. Rather, dual process models are typically verbal theories, which are difficult to conclusively evaluate or test. We present a dynamic dual process model framework of risky decision-making that provides an account of the timing and interaction of the two systems and can explain both choice and response time data.

Back to top


We investigated the roles of attention and memory in decisions from experience by comparing decision procedures where monetary value information is (a) present during sampling versus (b) revealed after sampling. In three experiments participants made a series of choices between pairs of risky gambles represented as urns containing different mixtures of blue and red balls, starting with a sample task before making a consequential decision, and varying visual and auditory salience. Individuals placed greater weight on rare events when outcome values were absent during sampling. Highlighting a rare reward increased its salience when outcome values were present, but not when they were absent. Parameter estimates from a hierarchical Bayesian prospect theory model supported the conclusion that value-absent choices involved greater overweighting of rare events.

Back to top


A substantial literature has demonstrated systematic violations of expected utility theory, but the underlying cognitive mechanisms are not well understood. I present a theoretical framework that integrates the psychological realism of heuristics and biases with the unifying power and mathematical precision of normative principles. The assumption that decision-makers should make optimal use of their finite time and limited cognitive resources (resource-rationality) yields an automatic method for deriving rational heuristics from first principles. Combining this method with process tracing allows us to answer descriptive, normative, and prescriptive questions about people’s strategies for multi-alternative risky choice and planning. I illustrate its utility for characterizing and quantifying the irrationality of decision strategies relative to a realistic normative standard. Preliminary results suggest that deriving and teaching rational heuristics is promising for improving decision-making.

Back to top


Often, we must make decisions about how to bring about a desired outcome. Making these decisions, we frequently have information about how broad a range of potential outcomes that particular action could cause (e.g., a cold could cause sneezing while the flu could cause sneezing, headache and fever). We find that beliefs about which action will lead to a stronger outcome varies predictably based on the action’s intentionality and valence. People believe the narrow scope action is more powerful when it is intentional (vs. unintentional) and when it is a cure (vs. a cause), and there is no interaction between the two. This research opens a line of inquiry into how the scope of a cause can influence judgments and decisions.

Back to top


Although people are notoriously poor at decision making in non-linear systems (e.g. exponential, logarithmic, etc), there has been limited research as to the cognitive strategies that people use when engaging with such systems, nor why those strategies break down. In a series of studies, we find that people use a two step approach when reasoning about nonlinear systems: 1) Treat non-linear systems as though they were linear to gain a rough estimate of the outcomes and then 2) (insufficiently) adjust to account for non-linearity. This anchoring and adjustment strategy yields predictable error, and opens the door to interventions for improving outcomes.

Back to top


While academic researchers categorize decisions into literatures, less is known about how lay people view types of decisions. We develop a framework to predict when people will see decisions as similar or different. Participants rate decisions on attributes identified in an initial exploratory study as reasons for perceiving decisions as different. Relating these attributes to perceived similarity of decisions, we identify the key factors that make decisions seem similar or different. We develop a framework for thinking empirically about generalizability in research, and discuss the implications for people’s choices of decision strategies.

Back to top


In visual priming tasks, participants respond to the identity of a visual target following a prime. Three priming effects—positive priming, negative priming, and oscillated priming—were observed and considered by previous theories as consequences of automatic sensorimotor processes. Here we developed a decision- theoretic model for the response time of visual priming tasks. The basic idea is that the brain has a constantly updated, probabilistic expectation for the arrival time and identity of the incoming target, which determines the motor preparing rate at each moment and thus the response time in an adaptive way. The model could quantitatively predict all three priming effects and how they vary with the temporal structure of the environment. It also offers new insights to a range of related phenomena.

Back to top


To make optimal decisions, people often need to know aspects of a numerical distribution beyond its central tendency. For example, optimal stopping requires understanding the dispersion of potential payoffs. I discuss how people learn (or don’t learn) and use (or don’t use) distributional information in simple tasks. We find that participants learn the central tendencies of two simultaneously encountered distributions with high accuracy, but show predictable biases in learning about dispersion. This suggests a puzzle about how people represent distributional information.

Back to top


Simple heuristics are typically regarded as tractable decision strategies because they ignore a great deal of information in the input data. One puzzle has been why simple heuristics can outperform more complex full-information models, such as linear regression. I will demonstrate that these less-is-more effects do not portray any inherent advantages of ignoring information, but rather that discarding information is never optimal (at the computational level). I place heuristics into a Bayesian inference framework, where popular heuristics are equivalent to Bayesian inference under the limit of infinitely strong priors. Interestingly, down-weighting information with the appropriate prior is always preferable to entirely ignoring it. I will discuss recent work on the idea that high covariance environments may be responsible for the success of many heuristics.

Back to top


One project uses transactions from 180,000 individuals on a stockbroking platform, to show that people frame outcomes at the level of individual days, splitting new investments 1/N over purchased stocks. Despite 1/N purchasing, the 1/N pattern is obscured by aggregation at the portfolio-level frame--- revealing that the frame is the individual purchase and not the portfolio. Another project uses supermarket purchase history from millions of customers to explore the attraction, similarity, and compromise effects in real, every day purchases, and the implications for process models of multi- alternative choice.

Back to top


This webpage has been visited