Jump to content

Talk:Regret (decision theory)

Page contents not supported in other languages.
From Wikipedia, the free encyclopedia

Under Minimax Regret you have "The aim of this is to perform as closely as possible to the optimal course." Previously, this said "uniformly close to the optimal course". I think the earlier language was more accurate. Part of the problem here is that we don't have a good definition of "optimal course"; perhaps "uniformly close to optimal outcomes" would be better language?

Professor Heidi Burgiel, Bridgewater State University, hburgiel@bridgew.edu 207.206.237.139 (talk) 16:18, 21 February 2013 (UTC)[reply]

Dr. Schlag's comment on this article

[edit]

Dr. Schlag has reviewed this Wikipedia page, and provided us with the following comments to improve its quality:


There are three types of decision theoretic models based on regret.

The first is that of minimax regret that goes back to Wald (1950) and was axiomatized by Milnor (1954). It is a way to understand decision making when there are no probability distributions. It is based on axioms that are very similar to those that lead to maximin utility. The term regret arises as the axioms tell the decision maker in each state to compare realized payoffs to best payoffs. This comparison can be in differences as originally proposed or in ratios (leading to the concept of competitive ratio). Emotions play no role, hence the entry to the wiki article is inappropriate for this orginal theoy of regret.

The second is that of anticipated regret due to Loomes and Sugden. The presentation of the wiki article makes it appear like this is the main model. It is not. Here the emotions that arise due to regretting not having taken other actions is modeled as an add on to the experienced payoff.

The third is no regret theory where the decision maker looks backwards and compares own performance to performance of other benchmarks or constant actions. This approach is popular in machine learning and computer science, the main reference here is the book of Cesa-Bianchi and Lugosi. Recently there has also been some interest in this appraoch in economics, eg see the papers on regret matching by Hart and Mas-Colell.

Ideally there should be a separate article for each of these models, each starting with an intro paragraph that refers to the others. In this context, note that the wiki article on competitive regret is extremely poor.


We hope Wikipedians on this talk page can take advantage of these comments and improve the quality of the article accordingly.

Dr. Schlag has published scholarly research which seems to be relevant to this Wikipedia article:


  • Reference : Ludovic Renou & Karl H. Schlag, 2008. "Minimax regret and strategic uncertainty," Discussion Papers in Economics 08/2, Department of Economics, University of Leicester, revised Apr 2008.

ExpertIdeasBot (talk) 18:58, 27 June 2016 (UTC)[reply]

Lead rewrite

[edit]

Until a few minutes ago, the first word of the article was a bold link to a song about regret. The problem with this topic is setting the stage, because several items are stacked. Having taken one cut, I can now see perhaps a way forward to get the topic items into the first sentence.

In decision theory, regret aversion is a model of decision-making centered on the human emotion of regret as an aversive experience, to be consciously minimized in the decision process.

To avoid here is to fall into the naive cartoon metaphor of regret: sobbing and moping and hair pulling and rending of shirts and villainous vows punctuated with "next time!" It's a trope for a reason, but it's also not accurate in the large. Regret can also be quite self-possessed. See Superforecasting: "The Economist reports that superforecasters are clever (with a good mental attitude), but not necessarily geniuses." These people are highly attuned to regret, but their adjustments are incremental, and evenly considered, and undramatic—yet many of these people are at the same time intensely competitive. I'm even hesitant to pose regret as an inherently negative emotion, because the pain of it is often in the mind of the perceiver, especially if the perceiver has a traditional sore-loser western psychology (which makes this depiction non-global, among other failings). — MaxEnt 15:52, 18 March 2017 (UTC)[reply]

Strange warning

[edit]

The page starts with Not to be confused with the loss metric of "regret" used in online machine learning/reinforcement learning (which is the difference between made decisions and the optimal decision). That looks incorrect as this page is all about the difference between the decision actually taken and what later becomes known to have been the optimal decision. So I have removed it.18:20, 11 January 2020 (UTC) — Preceding unsigned comment added by 2A00:23C6:1489:9900:7592:E59F:3FF1:2D3 (talk)

uniqueness of regret functional

[edit]

I would like to add some information on Fisburn's theorem on bilinear decision functionals with further generalization of the result showing the unique form of transitive regret function. (namely It was shown in the following article *Reference : A.E. Allahverdyan & V.B. Bardakhchyan, "Regret theory, Allais’ paradox, and Savage’s omelet," Journal of Mathematical Psychology, 2023.) — Preceding unsigned comment added by Kolobokish1 (talkcontribs) 19:20, 11 March 2024 (UTC)[reply]