Skip to main content

Faculty & Research

Decision Sciences - Academic Areas

Decision science

Decision Sciences is an interdisciplinary field that draws on economics, machine learning, statistical decision theory, operations research, forecasting, behavioral decision theory and cognitive psychology. Broadly speaking, Decision Sciences at INSEAD addresses three fundamental and inter-related questions. First, how should a "rational" person make decisions? This question is at the heart of economics, and often serves as a baseline for evaluating human decision making. Second, how do people really make decisions? Recent research has explored the ways in which people are "boundedly rational," and utilize rules-of-thumb and shortcuts to formulate judgments and to choose among alternatives. Often these shortcuts do well, but equally often they lead to systematic biases and serious errors. Finally, given what we know about rational decision making and actual behavior, how can we help people, especially managers, improve their decision making? Decision researchers employ a variety of techniques to improve decision making, ranging from sharpening statistical intuition to quantitative decision analysis

Below are descriptions of some of the current research interests of our faculty:

Decision Making under Uncertainty

Ambiguity and delayed resolution of uncertainty

Most of the uncertainties in real life are resolved over time. For example, an organization that implements a project today will receive information about the underlying uncertainties -whether a recession will unfold or not - sometime later. Our work (Abdellaoui et al 2011, 2019) provides evidence that delays affect risk-taking in a systematic manner.

There are questions that we would like to explore further: how do the delays in the resolution of uncertainties affect the attitude to ambiguity (i.e., the case of imprecise probabilities)? How does an ambiguous delay of resolution affect decision making in general? How is the evaluation of compound lotteries under ambiguity affected by delays?

We plan to contribute to the growing field of ambiguity (Trautmann and van de Kuilen 2015) and to address these questions through modelling (axiomatic) and experimental research. For the experimental part we are interested in both behavioural data from the lab, and process data via eye tracking. 

  • Abdellaoui, Mohammed, Enrico Diecidue, Ayse Onculer (2011), “Risk Preferences at Different Time Periods: An Experimental Investigation.” Management Science, 57, 975-987.
  • Abdellaoui, Mohammed, Enrico Diecidue, Emmanuel Kemel, Ayse Onculer (2019), “Temporal Risk Resolution: Utility versus Probability Weighting Approaches.”
  • Trautmann, Stefan T., and Gijs Van De Kuilen. "Ambiguity attitudes." The Wiley Blackwell handbook of judgment and decision making 1 (2015): 89-116.

Regrets for irreversible decisions

Regret theory (Bell 1982, Loomes and Sugden 1982) is one of the most popular alternatives to the normative benchmark of expected utility. Our research has demonstrated that the measurement of such a theory is possible (Bleichrodt et al 2010), provided a behavioural foundation (Diecidue and Somasundaram 2017), and highlighted that risk attitudes under regret are far more complex than was initially postulated (Somasundaram and Diecidue 2017). There are several avenues of research that can be explored in the future: how to include in regret theory systematic behavioural aspects such as reference dependence and probability transformation? How do teams of decision makers deal with individual regrets? How does regret affect irreversible decisions?

We plan to address these questions through modelling (axiomatic) and experimental research. For the experimental part we are interested in both behavioural data from the lab, and process data via eye tracking. 

  • Bell, David E. "Regret in decision making under uncertainty." Operations research 30.5 (1982): 961-981.
  • Bleichrodt, Han, Alessandra Cillo, Enrico Diecidue (2010), “A Quantitative Measurement of Regret Theory.” Management Science, 56, 161-175.
  • Diecidue, Enrico, Jeeva Somasundaram (2017), “Regret Theory: A New Foundation.” Journal of Economic Theory, 172, 88-119.
  • Loomes, Graham, and Robert Sugden. "Regret theory: An alternative theory of rational choice under uncertainty." The economic journal 92.368 (1982): 805-824.
  • Somasundaram, Jeeva, Enrico Diecidue (2017) “Regret Theory and Risk Attitudes.” Journal of Risk and Uncertainty, 55, 147-175.

Almost stochastic dominance

The idea of Almost Stochastic Dominance is to bound the ratio of marginal utilities, thus providing a continuum of stochastic order – ranging from first-order stochastic dominance to a complete order based on the expected value. Recent representative papers are Tsetlin and Winkler 2018, Müller et al. 2017, Tsetlin et al. 2015. Our work in progress involves applications to the situations where only partial information about the distributions is available, both in the univariate and multivariate cases. We think this stream of research is important, as it shows how to rank the distributions in a way that a broad class of decision makers would agree. That allows to free up more time for looking for other alternatives and for more automated decision making (as choice among many alternatives can then be delegated to, e.g., AI-based systems).

  • Tsetlin, I., and R.L. Winkler, 2018, “Multivariate Almost Stochastic Dominance,” Journal of Risk and Insurance 85(2), 431-445.
  • Tsetlin, I., R. Huang, L. Tzeng and R.L. Winkler, 2015, “Generalized Almost Stochastic Dominance,” Operations Research 63(2), 363-377.
  • Müller, A., Scarsini, M., I. Tsetlin and R.L. Winkler, 2017, “Between First and Second-Order Stochastic Dominance,” Management Science 63(9), 2933-2947.

R&D/creative activity

The biggest challenge in decision making is not so much about how to choose from a given set of alternatives, but rather how to decide that one should stop exploring the situation (e.g., looking for other alternatives, learning about the environment, inventing new things, building a model) – i.e., when to move to the decision/choice stage. There are multiple angles from which one can look at that (ranging from philosophy to psychology), and search/optimal stopping models from decision analysis might be very useful here. Overall, how to manage and facilitate research and learning activity is an important and quite open question.

  • Massala, O., and I. Tsetlin, 2015, “Search before Trade-offs are Known,” Decision Analysis 12 (3), 105-121.
  • McCardle, K., I. Tsetlin and R.L. Winkler, 2018, “When to Abandon a Research Project and Search for a New One,” Operations Research 66(3), 799-813.
  • Zorc, S., and I. Tsetlin, 2019 “Deadlines, Offer Timing, and the Search for Alternatives,” Operations Research, forthcoming

Aggregating information, framing the decision problem, and decentralized decision making

On any non-trivial issue, there will be plenty of divergent opinions – related both to different probabilistic forecasts and to different views about relevant objectives. In addition, decision making is often decentralized, in the sense that there is no single person possessing all relevant information – this information is spread out within, e.g., an organization, and different decisions are made at different levels and based on different information sets. There are multiple research directions that we are interested in here.

Recent interest in the interaction of humans and machines (which is now happening in various contexts – from auto-pilots to medicine) is also relevant. It brings up new challenges to decision making, which have to be addressed from multiple angles, including ethical and legal ones.

Individual and Group Decision Making; Negotiations

Question-asking in organizations 

Asking questions, a classical problem solving tool since Socrates, has been argued recently to enhance leader effectiveness. We explore why people in leadership positions might nevertheless be reluctant to ask genuine questions, why and when they should do that, and how organizational culture may limit one’s willingness to engage in constructive question-asking.  

  • Cojuharenco, I., & Karelaia, N. (2019). When leaders ask questions: Can humility premiums buffer the effects of competence penalties? Working paper. 

Decision making and identity

How does people’s sense of self or identity affect their decisions and what are the consequences? For instance, how do conflicting elements of ourselves affect our choices? How does our identity shape our prosocial choices? What are the consequences of making authentic choices, i.e., acting in accordance to one’s true self, in organizational settings?    

  • Karelaia, N., & Guillén, L. (2014). Me, a woman and a leader: Positive social identity and identity conflict. Organizational Behavior and Human Decision Processes, 125, 204-219. 
  • Guillén, L., Karelaia, N., & Leroy, H. (2019). Authentic (Mis)Fit: When being oneself reduces conflict and improves performance (and when it does not). Working paper. 
  • Cojuharenco, I., Cornelissen, G., & Karelaia, N. (2016). Yes, I can: Feeling connected to others increases perceived consumer effectiveness and socially responsible behavior. Journal of Environmental Psychology, 48, 75-86. 

Prosocial behavior

What determines whether people make more prosocial choices, i.e., choices that benefit others or society in general, as opposed to considering personal benefit/cost only? How do people allocate their time or resources to help other individuals or causes? Do we tend to help the most needy or the most “deserving” of help (i.e., with more merit)? How can we nudge people into making more effective prosocial choices? 

  • Cojuharenco, I., Cornelissen, G., & Karelaia, N. (2016). Yes, I can: Feeling connected to others increases perceived consumer effectiveness and socially responsible behavior. Journal of Environmental Psychology, 48, 75-86. 
  • Cojuharenco, I., Karelaia, N., & Murad, Z (work in progress). Understanding helping: Why help-giving is more helpful than help-seeking.  

Decision making effectiveness 

When does relying on decision shortcuts lead to high-quality decisions? When should decision makers pay more attention to the information at hand? When should they collect further evidence? When should they make fast, intuitive decisions? What are possible observable collective outcomes of (often unobservable) individual simple decision strategies (e.g., excess entry in competitions)? How does mindfulness affect decision making effectiveness? 

  • Karelaia, N., & Reb, J. (2015). Improving decision making through mindfulness. In Mindfulness in Organizations, Reb, J., & Atkins, P. (Eds.), pp. 163-189. Cambridge University Press. 
  • Hogarth, R. M., & Karelaia, N. (2012). Entrepreneurial success and failure: Confidence and fallible judgment. Organization Science, 23(6), 1733-1747. 
  • Karelaia, N., & Hogarth, R. M. (2008). Determinants of linear judgment: A meta-analysis of lens studies.  Psychological Bulletin, 134(3), 404-426. 

Negotiation process (moves in time)

Ordering of moves: Do (and if so which) moves made earlier versus later affect the negotiation outcome? Which moves should be made before others to improve negotiation outcomes? Which moves if done before others deteriorate negotiation outcomes or lead to no deals? 

Ordering of issues: Does the order of different issues in an agenda increase or decrease value creation, and thus improve or deteriorate negotiation outcomes?

Negotiation, emotion, and gender

Do same or different gender pairings create and claim value differently? Do different emotional displays by different genders impact negotiation outcomes differently? What if the counterparty is from the same or different gender?

Sequential negotiations

Do people perform better or worse in sequential negotiations with similar partners? What behaviors lead to improved or worse subsequent negotiations? When does improved trust lead to lower transaction costs and psychological safety, and thus better deals, and when does it lead to relationship complacency and thus to worse deals (both in business and in romantic relationships)? When and how does negotiation fatigue take place? How does it happen and what is its impact?

Multiparty negotiation

What bilateral win-win negotiation moves work or not in multiparty negotiations? When? How? What are win-win multiparty negotiation best practices?

Machine Learning and AI; Optimization

Understanding the new AI risks

Unlike any other technologies, Artificial Intelligence has the ability to take increasingly complex decisions: from recommending products, to credit scoring, to medical diagnosis. With decisions comes liability and risks. What are new risks due to AI and how to better manage them? How should regulators think about, say, the approval of AI systems, such as AI enabled medical devices or autonomous vehicles? How to best monitor these technologies as they continuously evolve by learning from data through usage? 

Approximating intractable optimization problems

 Optimization problems appear in many different areas and can take on a variety of shapes and forms. Depending on their formulation, these problems can be tractable or intractable to solve. We focus on providing approximations to intractable problems and studying the power of these approximations, both in worst-case and average-case scenarios. Below, we describe two concrete applications of this line of work.

Example 1: Shape-constrained regression

In this set-up, we wish to fit a polynomial regressor to data such that the least squares error between my predicted value and my observed value is minimal. Unlike standard polynomial regression however, we seek to impose additional shape constraints (e.g. convexity, monotonocity with respect to a variable) to the regressor. We can show that this regression problem is intractable to solve. However, by using what is known as sum of squares techniques, we provide an approximation of the intractable regression problem. We are further able to show that this approximation delivers a “good” regressor nevertheless, in the sense that the regressor obtained via the approximation is consistent with the true underlying function that generated the data.

Example 2: Graph alignment

In the graph alignment problem, we assume that we are given two correlated non-identical graphs. Examples of such graphs are the bipartite graphs that encode which user has rated which movie on Netflix and which user has rated which movie on IMDb. We would expect these two graphs to be positively correlated (i.e., if a user rates a movie on Netflix, he or she is more likely to have done so on IMDb) but not exactly identical. The graph alignment problem amounts to finding a mapping of nodes from one graph to the other such that the overlap of the graphs is maximized. In terms of Netflix/IMDb graphs, this corresponds to identifying which user on IMDb corresponds to which user on Netflix.

It is not known whether the graph alignment problem is easy to solve, even in the case where the two graphs are exact replicas of one another. In our work, we attempt to answer this problem under the assumption that we have a specific generative model for the two graphs. We can then come up with algorithms that solve this problem efficiently under the assumption that the two graphs we observe are generated using this model and that the parameters we have chosen for the model satisfy certain conditions.

Making personalized decisions in highly relevant contexts using randomized experiments

Across contexts that include healthcare, marketing, and revenue management, it is key for the decision maker to personalize an action to features of the recipient. For example, physicians want to administer different courses of treatment to different patients, according to their genomic type and medical history; e-commerce platforms want to recommend different products to different customers, according to their estimated preferences; and retailers want to target different demographics with different promotions.

We look at several questions in this space that are both intellectually and mathematically challenging, and also substantively impactful. What is a good design of a randomized experiment in practice, in the offline and in the online setting? How can we use data from a randomized experiment in order to evaluate a targeting policy efficiently? How can we target in the face of non-stationarity, i.e. in cases when the training data is not representative of the implementation data? Research on these and other challenging questions is part of our efforts towards developing impactful algorithms and frameworks for highly personalized decision making.

  • D. Simester, A. Timoshenko, S. I. Zoumpoulis, Efficiently Evaluating Targeting Policies: Improving Upon Champion vs. Challenger Experiments. Management Science, forthcoming.
  • D. Simester, A. Timoshenko, S. I. Zoumpoulis, Targeting Prospective Customers: Robustness of Machine Learning Methods to Typical Data Challenges. Management Science, forthcoming.
  • T. Evgeniou, D. Simester, A. Timoshenko, S. I. Zoumpoulis, Using the Timing of Past Responses to Address Non-Stationarity in Targeting Models Due to Dilution. Working paper.
  • A. Alban, S. E. Chick, S. I. Zoumpoulis, A Value of Information Approach to Designing Sequential Clinical Trials for Personalized Health Care. Ongoing work.

Resident Faculty

Standing & Affiliate Faculty

Emeritus Faculty

Visiting Faculty


Post-Doctoral Researchers