Talk:Subjective expected utility

Savage Paradigm
Could we perhaps create a page for Savage Paradigm that redirects here? 66.99.13.225 (talk) 14:17, 27 March 2008 (UTC)

Descriptive theory, then normative. Correct?
The article claims that “Savage proved that, if you adhere to axioms of rationality, if you believe an uncertain event has possible outcomes $$\{x_i\}$$ each with a utility to you of $$u(x_i)$$ then your choices can be explained as arising from a function…”

First of all, how can something like this be “proved”? It seems to be an empirical matter. But more importantly, the article later says, with regards to the failure of the theory to predict actual choices, that “Savage's response was not that this showed a flaw in his method, rather that applying his method allowed individuals to improve their decision making.”

The first paragraph claims a descriptive role for the theory, that it explains actual choices, but according to the second quote it is a normative theory about how people should reason. This problem is so flagrant that I have to question whether Savage ever actually made the first claim, since this one also contains the problematic claim of a proof. Could we either get a description more in line with Savage's own claim, or a reference for this flaw? Kronocide (talk) 13:22, 30 June 2008 (UTC)


 * I don't see how it's an empirical matter. Once you have axiomatized what it means for an individual to be "rational" and to hold "beliefs", then the question becomes a matter of mathematical logic. The crucial condition in the first sentence that you quoted is "if you adhere to the axioms of rationality", this means the rest is a normative statement of logic, rather than a positive statement of empirical observation. 202.36.179.66 (talk) 13:53, 26 September 2009 (UTC)

No Definitions Given For Terms Before Use
You write : "Savage assumed that it is possible to take convex combinations of decisions and that preferences would be preserved. So if you prefer $$ x(=\{x_i\})$$ to $$y$$ and $$s$$ to $$t$$ then you will prefer $$\lambda x + (1-\lambda )s$$ to $$\lambda y + (1-\lambda )t$$, for $$0<\lambda<1$$."

But you have to first tell readers what quantities $$s$$ and $$t$$ are. . ..