Webinars

A Replication Study of Operations Management Experiments in Management Science

Between-prospects comparisons: Regret Theory and Probability Dominance

 SDP-DAS Webinar-The Multivariate Metalog Distributions: An Introduction With Application to Strategic Decision-Making in Golf

SDP-DAS Webinar - Thinking about Process, Elicitation and Modelling in Decision Analysis 

Abstract: Decision analysis is much more than maximising an expected utility or value function: we all know that. Nonetheless, we may not reflect as much as we should on the wider, more qualitative processes that surround the analytical and quantitative steps. For instance:

• How and where are different types of knowledge drawn into the process?
• How do we work with and elicit not just numerical subjective probabilities, values and utilities, but softer more structural issues such as the context, models, available data, and appropriate computational methods?
• How do we recognise and address all the uncertainties that relate to a decision and its context?
• How do we report and archive details so that necessary information is available for audit?
• How do we recognise and reflect the wider political and Political processes surrounding the decision?

In discussing these and related questions, I will draw on several other disciplines and schools of thought than usually associated with (Bayesian) decision analysis. It is through the softer, qualitative process steps that we recognise, address and communicate those issues and uncertainties that are not modelled in the quantitative decision analysis, but still need communicating to the decision-makers so that they have a full understanding of the issues when they take their decision.

Multi-Multi-Attribute Utility: the Prettiest Axiomatization of Expected Utility You ever Saw & the Million-$ Question: Row-First or Column-First Aggregation?  

Abstract:

Keeney & Raiffa’s (1976) multiattribute utility aggregates U(X₁,…,X_n). Here X_j can refer to persons (welfare), timepoints (time preference), events (decision under uncertainty), commodities (consumer theory), etc. In applications one often has to aggregate over two or more kinds of attributes: risk & time, persons & commodities, etc. In classical models with complete separability the order of aggregation then does not matter. This gives our first result: the prettiest axiomatization of discounted expected utility you ever saw. Paradox 1: this is too simple to be true?

Modern behavioral generalizations relax complete separability (= sure-thing principle = time separability etc.) but maintain weak separability (= stochastic dominance = Pareto optimality etc.). Paradox 2: for two or more kinds of attributes this is just impossible.

A century-old forgotten theorem from macro-economics resolves our two modern paradoxes: Nataf (1948). He provided diagnoses and remedies for many ongoing debates, on ex-ante/ex-post fairness, incentive compatibility of random incentives, hedging in ambiguity measurements, equity in Harsanyi’s veil of ignorance, monotonicity in Anscombe-Aumann’s framework, etc. They all amount to the same million-$ question: “row-first or column-first aggregation?” Nataf resolved! 

CO-AUTHORS: Chen Li & Kirsten Rohde

Watch replay here

SDP/DAS joint webinar

Decision Making for Enhanced Health Security

Many internet platforms that collect behavioral big data use it to predict user behavior for internal purposes and for their business customers (e.g., advertisers, insurers, security forces, governments, political consulting firms) who utilize the predictions for personalization, targeting, and other decision-making. Improving predictive accuracy is therefore extremely valuable. Data science re- searchers design algorithms, models, and approaches to improve prediction. Prediction is also improved with larger and richer data. Beyond improving algorithms and data, platforms can stealthily achieve better prediction accuracy by “pushing” users’ behaviors towards their predicted values, using behavior modification techniques, thereby demonstrating more certain predictions. Such apparent “improved” prediction can unintentionally result from employing reinforcement learning algorithms that combine prediction and behavior modification. This strategy is absent from the machine learning and statistics literature. Investigating its properties requires integrating causal with predictive notation. To this end, we incorporate Pearl’s causal do(.) operator into the predictive vocabulary. We then decompose the expected prediction error given behavior modification, and identify the components impacting predictive power. Our derivation elucidates implications of such behavior modification to data scientists, platforms, their customers, and the humans whose behavior is manipulated. Behavior modification can make users’ behavior more predictable and even more homogeneous; yet this apparent predictability might not generalize when customers use predictions in practice. Outcomes pushed towards their predictions can be at odds with customers’ intentions, and harmful to manipulated users.

Listen to replay here

Utility over Risky Payoff Streams: Normative and Descriptive Approaches

Co-authors: Manel Baucells, Michal Lewandowski, and Krzysztof Kontek

 Abstract: We provide behavioral foundations of a preference model for risky payoff streams, a broad domain having timed lotteries, income streams under certainty, and repeated lotteries as special cases. Following expected utility and the notion that time is perceived as inherently uncertain, we inadvertently rediscover Bell's (1974) model. To this bedrock, we add the notion that preferences are affected by range effects. The result is a behavioral model with a broad domain and consistent with a plethora of phenomena (bias towards short payback periods, the four-fold patterns for risk and time, preference reversals for risk and time, temporal patterns of decreasing or increasing impatience, and magnitude effects).

Listen to a replay here

Properties of Utility Functions for Money

This talk is intended as the first of many in which decision analysts give casual talks on topics of interest to them. I will review, in a non-detailed way, three papers that I have written (during my career, i.e. not recently, and for the most part uncelebrated) that concern utility functions for money. An appropriate audience would be decision analysts with some research interest in the properties of utility functions. Though most of my published work has concerned multiattribute utility, this talk will not, except tangentially. It will also not be a general review of decision analysis.