Decision Theory

Decision theory provides a formal framework for how agents should make choices under certainty, risk and uncertainty. It distinguishes normative prescriptions—how an ideal rational agent ought to choose, typically via expected‐utility maximisation—from descriptive and prescriptive accounts that explain actual human choices and support real‐world decision processes. Core normative foundations rest on axioms of coherent preferences, leading to utility functions and Bayes’ rule for belief updating. Behavioural extensions relax assumptions of perfect rationality, incorporating prospect‐theoretic weighting, limited attention and heuristic sampling. Epistemic decision theory treats belief accuracy as a payoff, unifying action and credence through scoring rules. Non‐Bayesian learning models study how individuals fuse local observations and peer opinions across networks without full Bayesian computation. Imprecise‐probability and Dempster–Shafer frameworks generalise additive probabilities to credal sets and belief functions, offering robust choices when odds are vague or conflicting. Multi‐criteria decision‐making techniques aggregate heterogeneous criteria via weighting schemes, fuzzy sets or distance measures, supporting applications from resource allocation to supply‐chain design. Across these strands, advances in algorithms, uncertainty quantification and elicitation have broadened the reach of decision theory into engineering design, public policy and artificial intelligence.

Research from Nature Portfolio

New evidence‐theoretic methods have enhanced the modelling and fusion of imprecise assessments. An improved risk‐analysis framework combines triangular fuzzy numbers with the negation of belief assignments and evidence‐distance measures to weight expert opinions, yielding more balanced failure‐mode rankings in engineering applications. Extensions of Dempster–Shafer’s rule to incomplete “open” frames of discernment enable real‐time situation assessment in dynamic air‐combat scenarios, accommodating unknown threat categories and fusing sensor reports under uncertainty. A novel correlation belief function redistributes mass from focal propositions to related ones, mitigating conflict in classification tasks and improving robustness when fusing heterogeneous data sources.

Research from all publishers

Imprecise‐probability approaches have seen systematic developments in both foundations and applications. A subjective reinterpretation of credibility measures employs fair‐betting coherence to ground Liu–Liu’s credibility expectations on behavioural axioms, uniting consonant belief functions and Choquet‐integral expectations. Methods for inner approximation of coherent lower probabilities transform general credal sets into subclasses with favourable supermodularity or monotonicity properties, recasting decision problems as tractable linear or quadratic programmes without loss of informativeness. Distortion‐model estimators integrate imprecise probabilities with Bayesian networks and distortion functions to yield resilient human‐error‐rate estimates in safety‐critical systems, demonstrating stability under data perturbations and missing entries.

Decision Theory publication trend

The graph below shows the total number of articles in decision theory across all publications each year (not limited to Nature Index journals).

Technical terms

Expected utility: The probability‐weighted average of a utility function over uncertain outcomes, used to rank risky prospects.

Credal set: A convex family of probability distributions representing imprecise beliefs about uncertain events.

Dempster’s rule of combination: A mechanism for aggregating independent belief assignments by normalising and redistributing conflicting mass.

Choquet integral: A non‐additive integration operator for expectations under capacities or belief functions.

Prospect theory: A behavioural model that describes decision‐weight distortions and reference‐dependent utility in choices under risk.

Non-Bayesian learning: Distributed update rules that blend local observations and neighbours’ beliefs without full posterior computation.

Credibility measure: A function quantifying the uncertainty of evidence assignments, generalised from classical entropy concepts.

References

1. Decision Theory.
2. Fusion of expert uncertain assessment in FMEA based on the negation of basic probability assignment and evidence distance. Scientific Reports (2022).
3. Situation assessment in air combat considering incomplete frame of discernment in the generalized evidence theory. Scientific Reports (2022).
4. A new correlation belief function in Dempster-Shafer evidence theory and its application in classification. Scientific Reports (2023).
5. A subjective interpretation of Liu–Liu’s credibility measures and expectations. Fuzzy Optimization and Decision Making (2023).
6. Inner approximations of coherent lower probabilities and their application to decision making problems. Annals of Operations Research (2023).
7. Distortion models for estimating human error probabilities. Safety Science (2023).
8. Accuracy-First Epistemology Without Additivity. Philosophy of Science (2022).
9. Adaptive Social Learning. IEEE Transactions on Information Theory (2021).