share interactive transcript request transcript/captions live captions
download
|
MyPlaylist
GEOFFREY FISHER: This is some joint work with Vanessa Janowski and Antonio Rangel from Cal Tech. So many important decisions that we make, even on an everyday basis, often involve us making trade offs between risky and safe alternatives. So for instance, do you want to buy stocks or bonds? Do you want to buy insurance for particular life events or would you rather go gambling at some casino and risk money that way?
And one observation that's fairly prevalent in the literature is that a significant amount of variation is observed in these types of choices. And this variation is found both at the level of an individual and at the context that an individual is in. And because of this, we're really only beginning to understand some of the underlying mechanisms that are responsible for decision making under uncertainty.
And so the idea in this paper is that a significant fraction of this variance in decision making under uncertainty can be explained by differences in the allocation of attention at the time of choice. And so what that means is that as you're making a choice, if you tend to allocate more attention towards the upside of a risk, you may be more likely to accept that gamble. If instead you happen to allocate more attention towards the downside or the loss of a risk, you may be more likely to pick a safer alternative.
And so what we're going to do in this paper is find some evidence for this mechanism. And we'll do so in two laboratory experiments. The first experiment is going to be purely correlational. The idea is to really ask, are there any differences at all in risky decision making that are associated with any differences in attention? And we'll find an important link here. And then in the second experiment we'll ask whether this link is causal.
So what happens as we introduce exogenous variation in attention? Will doing so affect choices? And we'll find some evidence that, yes, in fact it will.
OK. So let me run through the first correlational experiment with you. So the idea here is to have subjects come into the laboratory. And they'll take part in about 300 trials that all roughly have the same structure. At the beginning of every trial, every subject sees a fixation cross at the center of the screen. After they look at that fixation cross for half a second, they see the main choice options for a specific trial.
And so in every trial, a subject has to decide whether they want a particular mixed valence gamble that they see on the screen in front of them or whether they want a reference option of $0 for sure. And so in this particular trial, a subject is deciding, do I want to choose a lottery that will pay me $10 with probability of 0.2, will lose me $5 with probability 0.5, or will give me $0 with probability 0.5. Or would I rather have a safe alternative that gives me $0 for sure?
Subjects have as long as they like to respond to this question. And critically, as they're responding to this question, we use an eye tracking camera that sits below the computer monitor to record precisely where each subject is looking on a screen so we have the order and duration of each fixation to each region of interest which here is a particular attribute. So the regions of interest here are the gain amount, the loss amount, the gain probability, and the loss probability.
And so after subject enters a response, they see a blank screen, and then move on to the next trial. And they repeat this for about 300 trials. So the big question here is, are there any differences in attention that are associated with any differences in behavior? And so let me show you a couple of analyzes that get at this.
The first is between subjects analysis. The idea here is to run a number of regressions where the dependent variable for every subject is the average amount of time or the amount of time that they chose the gamble, so the fraction of time they chose to gamble in the experiment. For all subjects, that's regressed on attentional data, which is the relative time that they spend fixating to each region of interest.
And so the one I want to draw your attention to here is in the last column, where it looks at the relationship between the propensity to choose the gamble and looking at any component of the loss choice. So any loss is either looking at the loss amount or the loss probability.
And what you see here is a significant negative relationship. So subjects who are more likely to look at the loss are also more likely to pick the safe option of $0 for sure. And you could see from this R squared that it does seem to explain a relatively high percentage of the variation in choices here.
So this is a between subject analysis. We could also look more within subjects. So here what we'll do is take a binary variable, 0.01 variable, for whether or not they chose the gamble and regress it on attentional data in each trial. And so again, let me just draw your attention to the any loss column. So again, here, it's replicating the previous between subjects finding where subjects who are more likely to look at any component of the loss are also more likely to choose the safe option and less likely to choose the risky gamble they see on the screen.
So this is interesting. It provides some correlational evidence, but it doesn't really get at the main question which is, is there any causal relationship from attention to choice? Right now it's this sort of chicken/egg problem. It's not clear whether preferences are affecting attention or whether attention can shift choices. And personally, I likely believe that it does go both ways. But what we want to establish in experiment two is that there is, in fact, this causal relationship from attention to choice.
And so it's a little bit tricky to come up with these experiments. So let me show you what we came up with here. So in this experiment, subjects are still making decisions over whether they want the same mixed balance gambles that they saw before, except this time once they attend to a particular option for a certain amount of time-- so once they attend to the gain component for a certain amount of time or the loss component for a certain amount of time-- that component will then leave the screen.
So each component has a maximum fixation duration that you can attend to it. Once you hit that maximum fixation duration, it disappears, and you'd be left with only the other option on the screen. Once you then look at the remaining option for a certain amount of time, that too will disappear. You'll see a question mark. And you're incentivized to enter your response as fast as possible after that.
But so how does that shift attention in a systematic way? So what we do is make one of these options a target option. So let's say, for instance, that the gain component here is a target option. Target options need to be fixated to for about 1.2 seconds. After target options are fixated to for 1.2 seconds, they disappear from the screen. The non-target option only needs to be fixated on for about 0.4 seconds. So once a non-target option is fixated on 3.4 seconds, that will then disappear from the screen.
So there's this clear exogenous manipulation in attention where attention is being shifted to the target option away from the non-target option. So let me just run through a sample trial so you could see how this would work here. So let's say that in this trial the gain component is the target option. The loss component is the non-target.
So you're free to fixate back and forth. But once one of the targets reaches its maximum fixation duration, then it will disappear from the screen. So likely in this case, since the loss component is the non-target option, that will reach its maximum fixation duration first because its maximum fixation duration is shorter than the target duration.
So once this is fixated to for about 0.4 seconds, it will disappear from the screen. And the screen will only have the gain component available left. Then once that component is fixated on until it reaches its maximum fixation duration of 1.2 seconds, then that will disappear from the screen and subjects then enter their response as fast as possible.
Now critically, every question that subjects see will have both a gain component that is the target and the loss component will be the target. So what that means is that in the experiment, one trial will have the gain components to target. Another trial somewhere in the experiment will have the loss component to the target. So it's very strong within subject's design that allows us to answer this question of whether exongenously varying attention can manipulate choices.
And so that's really what we want to test here. So what I'm showing you here is the average fraction of time that that subject's chose to gamble. And it's just split by whether the target attribute was the gain or the loss. And what you see here is that when the target attribute is the gain, subjects are significantly more likely to choose to gain than the loss, finding some evidence that this channel is, in fact, causal from intention to choice.
[APPLAUSE]
Geoffrey Fisher, assistant professor in the Charles H. Dyson School of Applied Economics and Management, presents current research findings regarding behavioral economics and human decision-making Sept. 8, 2015 as part of the Behavioral Decision Research Workshop Showcase. Sponsored by the Department of Human Development and the Center for Behavioral Economics and Decision Research at Cornell University.
