« Extreme carcinogenic doses for rats | Main | Adverse selection is NOT the problem »
Risk vs. uncertainty
Have you ever read Frank Knight, or the Austrians, and wondered what this distinction is all about? Neuroscience comes to the rescue:
In the experiment, test subjects made ambiguous bets while their brains were scanned using a functional magnetic resonance imager (fMRI).
In one example, the subjects were given the choice between betting money on the chances of drawing a red card from a "risky" deck that had 20 red cards and 20 black cards—that is, where the probability of choosing either color was 50:50—and making the same bet with an "ambiguous" deck where the color composition of the cards was unknown.
In most cases, the subjects chose to make the risky bet. Logically, however, both bets would have been equally good because in both cases, the chance of pulling a red card on the first draw was 50:50.
The brain scans revealed that ambiguous wagers were often accompanied by activation of the amygdala and orbitofrontal cortex (OFC), two areas of the brain that are involved in the processing of emotions. In particular, the amygdala has been found to be closely associated with fear.
A correlation between aversion to ambiguous decisions and activation of emotional parts of the brain makes sense from an evolutionary point of view, Camerer said. "Freezing in the face of danger is an old, emotional response which probably was evolutionarily adaptive in our ancestral past."
In the modern human brain, this translates into a reluctance to bet on or against an event if it seems at all ambiguous.
Could this help explain the absence of various long-term insurance markets? Thanks to Chris Masse for the pointer.
Posted by Tyler Cowen on December 13, 2005 at 06:20 AM in Economics, Science | Permalink
TrackBack
TrackBack URL for this entry:
http://www.typepad.com/t/trackback/3576/3828178
Listed below are links to weblogs that reference Risk vs. uncertainty:
» The Blog Herald: more blog news more often from Newsblog
An issue or news story to discuss? Submit a blog entry... [Read More]
Tracked on Dec 21, 2005 7:55:16 AM
Comments
Since the test subject had to bet on red, the offer to bet on a draw from the "ambiguous" deck sounds a lot like a sucker bet. If someone offered that bet in a bar, I'd assume the deck was full of black cards.
No wonder the subjects chose the "risky" bet.
Posted by: The Other Brock at Dec 13, 2005 7:01:46 AM
Good point, Mr. Other Brock.
I've often wondered if people in experiments tend to behave the way they would in real life, instead of the way experimenters tell them to behave (i.e. "assume the number of red cards in the deck is random." If you allow this possibility, a lot of the "irrational behavior" found in experiments goes away.
To me the most interesting question in behavioral economics is whether subjects consciously mistrust the experimenters, or whether human rationality is genetically programmed to assume a certain level of mistrust at a subconscious level.
Posted by: DK at Dec 13, 2005 7:51:48 AM
I think most people would assume you are maximizing information by choosing the deck with the known distribution (That was my gut feeling). I am sure it would take some explanation and convincing that the the maximum likelyhood criterion assumes the distribution of the unknown deck at 50:50 given no observations and therefore both choices are equal. I imagine the counter argument would be "but we don't know so perhaps it is worse..."
Posted by: sprice at Dec 13, 2005 9:18:45 AM
This example actually does not quite illustrate the difference between "risk" and "uncertainty." In this example one can write out the state-space and assign probabilities in both cases. In the case of true "uncertainty" the state space itself is not known or ill-defined. The real-world is uncertain. By definition we cannot know what kind of inventions and discoveries are going to change the world and in what way. Of course, this kind of uncertainty is somewhat nihilistic for economic analysis. I do think however that in modeling human behavior it is important to recognize that most decisions are taken in the frame of uncertainty rather than risk, which is why expected utility models need to be taken taken with a grain of salt.
Posted by: tea at Dec 13, 2005 9:52:02 AM
Is it just me or is the main lesson from the current fad for MRI experiments in economics just that we know amazingly little about how the brain works? Different stimuli, different lights go on. My computer is that way too - I press different buttons and different things happen on the screen. Somehow I thought we were farther along than this.
Jeff Smith, still working on making my web page as cool as Gary King's, so that I can get a link too. :)
Posted by: Jeffrey Smith at Dec 13, 2005 10:59:33 AM
I think fMRI is way overdone - basically you can publish a trivial economics experiment if you repeat it with fMRI. We already knew people were risk-averse; after this study we know...people are risk-averse, and the aversion is associated with an emotional response (duh).
Posted by: Paul N at Dec 13, 2005 11:08:37 AM
I have a pretty big problem with their characterization of the experiment:
"In one example, the subjects were given the choice between betting money on the chances of drawing a red card from a "risky" deck that had 20 red cards and 20 black cards, ”that is, where the probability of choosing either color was 50:50" and making the same bet with an "ambiguous" deck where the color composition of the cards was unknown.
"In most cases, the subjects chose to make the risky bet. Logically, however, both bets would have been equally good because in both cases, the chance of pulling a red card on the first draw was 50:50."
That's just plain wrong unless they are leaving out very important information: what's the process from which the "ambiguous" deck is created? In other words, what is the distribution of expected number of red cards in that deck?
Sure, if the "ambiguous" deck is built by 40 coin flips, then the statement is true. But there are plenty of other processes, not least of which is an evil experimenter secretly loading it with all blacks, that vary from this result. In fact, I think the process to generate the deck needs to provide a distribution of expected values of red cards with a median and mean of 20 and no skew (with a 40 card deck, that is). That cuts out plenty of real world processes.
I think the psychological insight might just be that people realize correctly that there are two levels of risk in the "ambiguous" deck. First, the risk that you've correctly estimated the distribution of expected red cards in the "ambiguous" deck, and second, the risk of picking a red card from that deck.
The relatively simple 50:50 deck doesn't sound so bad in comparison.
Posted by: Richard Vermillion at Dec 13, 2005 1:55:20 PM
I had the same first reaction as Richard Vermillion. How do you know that
the odds on the ambigious deck are 50-50. They can be as high as 39/40 or
as low as 1/40. Yes, you can argue that the sum of the probabilities of
the possible ambigous decks is 50-50, but for that specfic deck the odds are
essentially unknown. I, too, would select the deck I know to 50-50
as opposed to one that may be 50-50.
Posted by: Murphy at Dec 13, 2005 2:34:19 PM
A mathematical friend pointed out to me that the only requirement is that the mean of the expected value be 20 red cards (the median is not required to be 20, nor is skew an issue). On second thought, that makes sense -- a process that generates an "ambiguous" deck with 10 red cards twice as often as 40 red cards would still give you a 50/50 chance.
He also points out that if the subject can choose the color he/she wants to pick, and does it by flipping a coin, then you've removed the need to know anything about how the "ambiguous" deck is generated.
Posted by: Richard Vermillion at Dec 13, 2005 4:16:02 PM
Are they told who gets the money if they lose the bet? Is it the experimentor?
That would effect my decision.
Posted by: josh at Dec 13, 2005 6:48:50 PM
Since the test subject had to bet on red, the offer to bet on a draw from the "ambiguous" deck sounds a lot like a sucker bet. If someone offered that bet in a bar, I'd assume the deck was full of black cards.
According to the study, subjects were allowed to bet on either red or blue. This is the Ellsberg Paradox. A rational decision-maker should not prefer BOTH bets from the risky urn to those from the ambiguous urn.
Posted by: athxu at Dec 14, 2005 8:50:56 PM
athxu alone seemed to have read the paper, good on ya. The rest of you are making a sensible but incorrect guess from a press release or abstract. Indeed, the subjects can pick either color so that should provide some guarantee that they cannot be cheated (Newcomb's nutty paradox excepted)...but wait, even better, a third condition in the paper actually does stack the deck against them (they pick a color and their bet only counts if they bet the opposite color of a player who peeked at the deck). So the inference is that there is a general warning system in the brain that is activated by both playing a better-informed player *and* choosing in the face of distributional ambiguity. The key point in the paper is this: the fMRI evidence points to orbitofrontal cortex (OFC) as an important of an uncertainty-evaluation circuit. So what if you have a lesion in OFC? then you should be neutral toward ambiguity? And such people are. Jeff Smith makes a good point-- does this just tell us how little we know? Yes it tells us these systems are complicated; it also tells us we can conceivably make rapid progress. It boggles my mind to think that people are not hopeful about promise of fMRI, genetics, TMS, EEG, PET, DTI and other tools to teach us amazing things. Very soon? No. Some day? Certainly. Especially for social science. Keep in mind that Frank Ramsey speculated about having a "psychogalvanometer" to measure utilities directly...and now we have one.
Other thoughts: DK, most behavioral economics evidence now is not from experiments-- and no, subjects don't distrust us, we go through copious IRB approval more rigorous than editorial review for peer-reviewed journals for honesty--, it is from field data. Don't equate experimentation in economics, which is a tool to establish boundaries of phenomena and rule out bad theories, with behavioral economics, which is an approach that uses psychological data to make better assumptions for the eventual purpose of making better predictions.
Posted by: Colin Camerer at Feb 2, 2006 12:49:07 AM
leerling dronken in bed ^^^ amateurs anal ^^^ esotiche kino ^^^ ludicrous asiatiche ^^^ plus froid sauvage vierge ^^^ le plus frais surnaturel vierge ^^^ mman quarante cinq ^^^ froid pere trente ^^^ crave ragazze dildo ^^^ pretty segretaria sex ^^^ deltagande asiatisk ligga ^^^ fager servitris grupperna ^^^ caldo ragazze prostituta ^^^ sconfinato cowgirl dildo ^^^ hetere svak mork ^^^ fotsex klipp ^^^ pa kjokkenet 3kant ^^^ nett orgasme ^^^ kalos neos ^^^ asiatis piomenos sto domatio ^^^ papa americain ^^^ papa galerie ^^^
Posted by: levan at Sep 5, 2006 7:46:51 AM
