share
interactive transcript
request transcript/captions
live captions
download
|
MyPlaylist
VALERIE: Welcome to the next installment of the Center for Behavioral Economics and Decision Research and our series. So glad you could all come. Today I'm, again, not going to spend too much time introducing the speaker. Because if I covered everything in the vita, it would take us too long, but a few highlights.
He got his PhD in 2010 from Stanford University. But I came to know his work before that when he was a graduate student. Yes, I had heard of him and his work when he was a graduate student, and I'm not alone. He did after that, though, a post-doc at Vanderbilt University with David Zald and finished that in 2013.
He's received a number of awards and distinctions, including being identified as a rising star from APS. He got a distinguished dissertation award in the social sciences. And he received one of the top 10 Scientific Advances of 2007 Award as well.
Now, remember, I said his degree was in 2010. And he got the Top Science in-- you might have just retired after that. Quit while you're ahead. So he does work on a number of things, including-- I think he sent around a paper on the affect integration motivation framework.
So he integrates aging, and neuroscience, and motivation, and affect, and how people feel about losses. He has a beautiful paper, for example, in psych science on this parametric response in the insula to losses. Just a beautiful paper. Really, really enjoyable. Frontal parietal activity and white matter. Although maybe that's--
GREGORY R. SAMANEZ-LARKIN: I'm going to talk about some of that.
VALERIE: Yeah, you'll talk about some of that. Very good. Very good. And in short, he does careful scientific work. He carefully entertains up opposing hypotheses, which is extremely important. He's one of those good guys that I talk about who are the process-oriented neuroscientists who aren't just about the brain lights up. And isn't that wonderful? Sizzle, but no beef. You got the sizzle and the beef, which is good.
AUDIENCE: I never heard that in an introduction.
VALERIE: Yes. Well, hey. And we have it on tape. We could edit it later, though. We'll do a soliloquy from Hamlet or something later on the real tape, right? And that's extremely important, a process-oriented hypothesis driven approach in neuroscience that's open to alternative views. And it's really commendable to take that approach and very, very, very impressive.
And I would say, too, despite the early stage of his career, if you thought about it, a basic textbook in aging or what people are going to remember 25 years from now and 50 years from now, I cannot imagine that this field would omit his work. So let's give a warm Cornell welcome to Greg Samanez-Larkin.
[APPLAUSE]
GREGORY R. SAMANEZ-LARKIN: So thanks so much, Valerie. That was really nice. And also, just thanks for the invitation. I'm always happy to visit Ithaca. It's a lovely place and had fun meetings and looking forward to talking to you more today. Everyone can hear me OK, right? Good. OK.
So one thing I'll say first is please interrupt me. So at any point, you don't have to raise your hand or anything. You can talk out. You can whatever. This will be more fun for me if I get a little feedback from you guys. I brought a very flexible talk. So I can just run through this thing and take up all the time or we can-- what would be even more fun for me is if you stopped me along the way, and we have some discussion about it.
So I'm not committed to presenting everything I brought today. So what I thought I would do is give you a little bit of an introduction to what I do and who I am academically. And so I'll give you a little bit of a framework for the kind of work that we do and then two example studies and these are-- so I debated.
Should I do this? Or should talk about some really fun, new stuff that we're doing? And I decided to do the first thing. And I'll come back some time and talk about some of the latest stuff. But I thought that this was a good place to start. So in general, what do I do? I study what happens to your brain after you're fully formed, I usually say.
So after the mid-20s on, for the rest of adulthood, what happens to your brain and how that affects specifically financial decision-making. So a lot of what we do is connecting what we think of as pretty low-level neural systems. And actually, in some cases, we're looking at the molecular level with PET imaging to very broad, cumulative lifetime outcomes like financial health, and wealth, and credit scores, and things like that.
And so we're really trying to cover a lot of territory, which I think is ambitious in a way. I've also gotten a lot of career advice that this is a terrible thing to do, which I've mostly ignored. But mostly, what we're not doing is just thinking that making the assumption that these two things can be directly connected.
Instead, we're looking at the psychological processes through which these things are connected. So that's what we're trying to do, covering a lot of territory, but making sure we cover all the territory in between. So can I just get a-- so who primarily-- I just want to get a sense of the audience.
Who primarily identifies as a psychologist? Is this everyone? No. OK. And then who identifies mostly as a business school or econ person? Oh, good. And then what about neuroscience? Does anyone mostly identify as a neuroscientist?
Wow, it's like, what a great group. It's so well balanced. But this also makes it a little tough for me. Because I can either give it like a business school talk or a psychology talk or neuroscientist. Anyway, so I'll just do my best. OK, so right. So what we're interested in-- mostly financial decision-making, so saving and spending kinds of decisions.
But lately, we've started getting more interested in how to use this framework to study health-related decision-making and social decision-making as well. So let's get into it. So I'm interested in aging. So at the core of all of the research I do is this issue of aging.
And so just to motivate this a touch, if you haven't-- probably lots of you are well aware of this. But there's a lot of-- there's increasingly more and more older people in the world, in the US and across the world, right? So both because of increased longevity or increased life expectancy, but also because of low fertility rates.
There's just a larger proportion of older people in the world. Why does this matter for financial decision-making? Well, that's where all the money is. So 40% of the wealth in the US is held by people over the age of 65. And actually, a large share of that is in arguably risky investments.
But this is actually even an underestimate if you think about it. Because this is combining the very end of a saving period in life with a long drawn out period of decumulation where people are spending down all their earnings. So there's actually even more money in this age group than it seems, right?
Fraudsters and con artists know this. And this is potentially why they go after older adults, because that's where all the money is. And there's also this perception of them being gullible, and suckers, and all this stuff. And we do a bit of research on that, too. But I won't talk about that today, unless it comes up at the end.
But either way, there's a lot of money here. So shouldn't we know something about the financial decision-making of this age group? Because potentially, there could be a big impact on the economy, right? So the behavior of this group of individuals could have a big effect on the economy.
But also, independent of these broad economic implications, the other reason we study this is because we really care about the well-being of individuals, right? So I'm a psychologist primarily. And these are really, really critical decisions late in life.
If you make a mistake, it could be horrific, right? So if you make a big financial mistake late in life, you're not going to recover a 50-year investment in any span of time, relative to the time that you have left in life. So these are just really critical decisions.
One of the reasons we started getting into studying financial decision-making, honestly, was because it's easy to study in the lab. You can give people little money. You can take it away. You can give them more money or less money. People mostly care about making money. Although, we're doing some stuff, looking at other domains that hopefully I can talk about at the end, too.
But it's a convenient experimental tool. So not only are these important decisions, but this is a nice tool that we can use in the lab. So most of what we do-- so I think in the business school, some people think of what I do as experimental finance or neurofinance. Because we have people play these little games. And it's a bit of an artificial environment, even though it's for real money. And so I'll show a couple of examples of that today. OK.
So there's a lot of interesting things happening around this topic lately. And so I'll just give you an example here. So this is Surya Kolluri who-- I was on a conference call with him a few months ago. So the financial services industry is now really getting interested in aging. And this has been sudden in the last few years.
And so one comment he made on this call was this issue, speaking about aging and decision-making is red hot in the industry right now. So financial services really cares about this and mostly for the reasons of trying to prevent financial fraud. So they see a lot of money flying out the door, and they want to prevent that.
And so now, I think-- I don't know of another time in recent history where people in the financial services industry came to talk to psychologists, but this is actually happening now. So people are asking for advice for people like me, like, what do we know about aging and decision-making?
And so I think this is a really cool thing that's happening. Historically, if anyone from the industry wanted an academic perspective, they'd probably go to a business school or a financial economist. And I think now they're starting to talk to psychologists which, of course, I'm biased. But I think that's a great thing.
And then a couple months ago, Schiller, who is my colleague at Yale, speaking at this Future of Finance Conference. And part of what he said was, we need an industry that recognizes that people are complex, psychological animals. I think we're starting to get an industry that's realizing this, or at least pieces of the industry are starting to realize this.
So I think we're getting there. But what makes people so complex psychologically? Right? So I think this is true. I totally agree with this. But what makes them complex psychologically? So when I started doing this work-- so as Valerie mentioned, when I started as a graduate student at Stanford, previous to that I had studied emotion, emotional experience, emotion and aging, and a little bit of emotion/cognition interaction stuff.
But I really got into the decision-making stuff when I started graduate school. And there wasn't much out there. There wasn't much literature out there. And so at that time-- so there's been a lot of activity in the last 10 years. But if we just think historically, what do we know about the psychology of aging that might predict whether older-- whether financial decision-making will get better or worse? Or what will happen with financial decision-making across adulthood with aging?
So in the most broad sense, what we might point to are things like experience increases, right? As you move through life, you get more and more experience with financial decision-making. So you should get better, right? This predicts a link, basically linear increase with age. The more experience you have, the better your decisions get.
And I'm citing here strategically some more economics-oriented papers that are telling this story. And I'll come back to this. So there is experience on the one hand, which might predict improvement with age. But there's also some cognitive decline that happens just normatively or with normal, healthy aging, things like selective attention, inhibiting interference, working memory.
These kinds of skills seem to decline relatively linearly averaging across lots of people. There are tremendously important individual differences. But in general, we see these things declining. And this is a gradual decline, right? So it's not like it happens very suddenly. It takes decades to unfold.
But you might predict that to the extent that there's lots of information to process and often you're in maybe a bit of a novel situation, that some of these skills, some of these cognitive abilities would make it harder to make financial decisions or make wise financial decisions.
So this side of it would predict that older adults would get worse at financial decision-making. And just as a preview, both of these things are true. So sometimes the decisions of older adults are worse, sometimes they're better. I'll show you an example of each today. All right.
So just a little bit more of-- so like I said, we do these behavioral tasks. We also do a bit of brain imaging. So why do we do all this combination of stuff? And what do we think we are bringing to the table that the industry is potentially asking for to help explain age differences in financial decision-making? So we think we have something unique.
And I'm just going to point to Greenspan for a second here, who in 2007 said, I've been dealing with these big mathematical models of forecasting the economy. And I'm looking at what's going on in the last few weeks and I say if I could figure out a way to determine whether or not people are more fearful, or changing to euphoric, and have a third way of figuring out which of the two things are working, I don't need any of this other stuff. I could forecast the economy better than any way I know.
This makes it look like I just follow everything Greenspan-- like fed chief says all the time. This was on The Daily Show. So I was just watching The Daily Show once. And I was like, whoa, this is what we do, though. This is what my job is, essentially, right?
This is what we're expert at doing as psychologists is measuring that stuff, so both in asking people how they feel, looking at physiological responses, looking at neural responses. This is what we do. This isn't some thing that-- wouldn't it be amazing if we could find out how people felt and how that affected their decisions? That's exactly what we do, right?
And so I think-- and which of the things we're working. So it's there's looking at mechanisms of coordination between different kinds of processes, right? We also do that. So I think what Greenspan is talking about is, could we do this at the population level? Right? This is what would be really interesting. And I'll come back to that at the end of the talk today.
So I don't think we need to do it at the population level to make inferences about the population. But, OK, so how do we do this stuff? How are we linking these feelings potentially with financial decision-making? So we study something that's probably even lower level than things that you would call fear and euphoria.
So we study affect, so just even the most low-level feelings about whether you feel an investment option is a good thing or a bad thing, right? Do you feel pretty good about this? Do you feel bad about this in more of a dimensional than categorical way, even?
And we think that these feelings, these initial feelings, responses you might have when you're presented with different, let's say, investment options are related to the function of some systems. So I'm not going to go through all the anatomical details here.
But if you just look across the top row, we think there are these deep systems in the brain that are playing a role in generating or are representing these initial feelings you have when some options are in front of you. But then there's lots of other relevant information, right?
And so then maybe you think about other features of these options. Or you think about, what happened in the past when I did this thing? Or maybe someone gives you some advice. Or maybe you do some more research on this and you get some more information.
So there's this period which has a varying time scale of integration, right? So you get other information that might be relevant to the decision. And we think this integration involves coordination between the prefrontal cortex, especially the medial prefrontal cortex and regions of the striatum.
So for people that aren't brain people, this is-- so think about my head facing like this. If I cut this part of my face off and moved it to the side, it would look exactly like that. So this is the striatum. This is the caudate and the putamen. This is the ventral striatum, so I'll talk about that a lot.
And then this is a side view. If you took this half of my face off, this is what this would look like. And so these are these medial frontal regions that we'll talk a lot about. So mostly, just talk about these two parts of the brain today. So anyway, so then we think that there's potentially other relevant information that gets integrated.
And it's not just that we keep these feelings and these cognitions potentially separate, right? There's other bits of information-- update and modulate and modify how we feel about these things. So really, at the core of this model are these feelings, these affective states. And we think this is what drives choice.
So then once you have this integration, then basically these signals are modulated in these striatal regions, which then-- and then at some point, you do something, right? You decide to invest in something. Yeah, you make a choice. So this action part is at the end.
But really, I think what's unique about this model is that it starts with emotion in a way, right? It starts-- it says that affect is foundational for decision-making and in the context I normally apply it for financial decision-making, although we think it's a more global framework. OK. All right.
So that's a little bit where we're operating from and what's slightly different about our approach. All right. So let's walk through a couple of examples. So the first part of the talk, I'm going to talk about learning and risky financial decisions. Let me just-- it's 11:00. When does this end?
AUDIENCE: 1:15, technically.
GREGORY R. SAMANEZ-LARKIN: 1:15. So I've got a little over an hour. OK, just trying to pace myself. OK, so I'll actually start. I'll tell a touch of a story then. So when I started doing this stuff-- so my first year in graduate school, I mentioned that I was really wanting to get into studying aging and decision-making.
And the first thing I thought-- well, what's the most obvious thing about aging and decision-making? Well, older people are risk averse, right? So that seems to be just generally true, that people seem to get more and more risk averse as they get older.
And I was making generalizations based on things like, oh, you don't see a lot of older adults skateboarding and bungee jumping. So I was thinking about recreational risks, actually, when I was thinking about this a lot. And that later turns out to be really important.
So I just had this idea that this was a true thing in the world, right? And so I did my first study. And I designed the study. We're in the lab. People are playing this game where they're investing in stocks and bonds. And I thought, we're going to get-- the older people are going to be very risk averse.
And we're also going to do some brain imaging. And we're asking them about their feelings about these different options. And we're really going to figure this out. We're really going to nail this risk aversion thing. And so we did the first behavioral study. Older adults are more risk seeking. I was like, you know what? I'm just not that good of a researcher yet, I guess.
So we thought we had this really cool study, but it just didn't work. And I put that study aside. I thought it was wrong and didn't really work. And I did another study that actually worked out really well when it was-- anyway, so I did something else for a couple of years.
And then later as I had been reading the literature more, I realized, you know what? Not every paper shows that there's a risk aversion effect with age. In fact, this Deakin 2004 paper that everyone has cited is the most extreme risk aversion effect in the literature. And no one even comes close to it.
So then a few years later or so, somewhere along the way, I ended up doing a meta-analysis with a great colleague, Rui Mata, who is in Basel now, was formerly at the Max Planck Institute in Berlin. And what we did was we looked at all of the behavioral literature, so comparing younger and older adults on risky decision tasks.
And this is with Ralph Hertwig, also, who has done a lot of stuff on decisions made from description and experience. So what we did was we divided the types of tasks that people use and then looked at the age differences within clusters of tasks. And so at the highest level, we divided tasks where decisions are made from description and where decisions are made from experience.
So descriptions like you could have a 50/50 chance of win $10 or win $0, or you could have a certain chance of $5, right? All the information is sitting there right in front of you. And you can compute expected value. You can do everything you need to do, right there on that one screen, right?
And that's in contrast to these other tasks where decisions are made from experience. So this Iowa Gambling Task is most of the studies where you've got a deck of cards and you're drawing. And you're basically getting little bits of information as you're playing this game. And you're updating what you think the value of something is as you get information in the game, right? So there's a heavy learning component to this. This is why I say recent learning here.
So experience here means very recent experience. This is not the life experience that I was talking about at the beginning, right? This is you're in a new situation. You're trying to collect information really fast and figure out what's going on. So what did we find out?
So what's shown here is-- positive numbers here would mean that the older-- and all of these studies for the most part were comparisons of young and old. And by young, I mean people in their 20s and 30s and old in their 60s, 70s, and 80s, comparing those two age groups.
And so a positive number would mean older adults are more risk seeking. A negative number would mean older adults are more risk averse. The individual studies are these gray boxes. And then the colored diamonds are the summary effect across a bunch of studies.
So if you look at the description side, as far as I'm concerned, there's nothing interesting going on. There's effects on both sides. So where is this? I guess this drawer one is super, highly cited, or I had seen cited a lot. And I'm like, that doesn't even capture all the other-- anyway, so basically, these colored diamonds here are essentially zero.
There's almost no consistent age difference. It's not that older adults are more risk averse or more risk seeking in these kinds of tasks. Where we do see something slightly more systematic is in these tasks where you have to figure out what's going on as you go, right?
It's a new environment. You've got to learn really quickly. So in the Iowa Gambling Task, it looks like there's a slight positive effect where older adults are a little too risk seeking. Or I guess I shouldn't say too risk seeking-- are a little more risk seeking than the younger adults. And in this bias task, which I'll tell you about in a second-- same effect. And we've actually done it two more times since then. It's almost exact same effect.
And then in this other task-- the BART-- it's going the other direction, so not even something that's systematically explained by an age-related shift in risk preference, right? Even if it's a domain specific kind of thing, it's not that this is explained by, oh, older adults are just more or less risk tolerant.
In fact, there are effects on opposite sides. And it turns out if you look at these specific tasks, in the Iowa Gambling Task, the way to make the most money is actually to avoid the high variance decks. So you should be a little bit risk averse in this context, right? You should stay away from the risky stuff, because that's how you make the most money.
And In the BART, you actually are a little better off by pushing things a little bit further. So this is a task where you pump up a balloon. And every time you pump it, you get another point. But at some point, it might just blow up and you lose everything. But it turns out that people are a little too conservative in this, and that if you actually just push it a little bit, you could make more money. And older adults don't really do that as much.
So it turns out that basically older adults are just doing slightly worse in these tasks that are heavily learning dependent. And so I'll talk about exactly that today. So when we did-- so there's not a totally chronological story that I'm going to tell. But basically, when we saw this, we're like, oh, this is about learning. So these age differences that we're seeing in risky decision-making are really about learning, not necessarily about risk preference at all. OK.
So this was the first study I did that I thought was a failure. So what we did-- and this was in collaboration with Camelia Kuhnen who, when I started graduate school, she was just finishing her PhD in finance at Stanford. And there's a wonderful collaboration. We continue to collaborate now.
And so she had just developed this task with Brian Knutson where people are investing in stocks and bonds and repeatedly making these investments. And so this is just how an individual trial would play out. And so in this data that I'm about to show you, half of the people played this task while undergoing fMRIs. They're heads in a magnet. And they're playing this game.
And other half are just sitting at a computer in their behavioral lab playing the game. And they actually weren't screened for fMRI or anything. And I'll tell you why that's important in a moment. And so we tell them there's going to be this-- so here's $20, stick it in your pocket So you get a little bit of an endowment.
And here, you're going to play this game. And then you're going to be repeatedly asked to invest in one of these three assets. And so you could choose one of the stocks or the bond. You're going to make a choice. It's going to highlight you picked that stock. You just got $10 from choosing that, investing in that stock. And overall, your account is at $20.
Here's what would have happened had you invested in these other things. If you picked the bond, you would have gotten $1. If you picked the other stock, you would have gotten zero. So this is really critical. So we gave full market feedback, basically. And at this time, especially, was pretty uncommon in these kinds of tasks. Usually, you just get a bit of information about what you just chose, but not what you could have chosen.
The reason this is important is because this minimizes the demands for exploration in the task, right? So it's not that you have to choose everything to get information about it. So you can learn about all three options totally independent of your choice. And we thought that that was really important. Because there might be differences in the exploratory tendencies of different age groups. So that was a feature.
The other bit of information you get before you play this is that like the real world, the stocks are riskier, but you could potentially make a lot more money by figuring out which stocks are good. So there's a good stock, which has a positive expected value.
So half the-- 50% chance of winning $10, 25% chance of losing $10 and $0. And the non-optimal stock or the bad stock is the opposite expected value of that. So it's a minus $250 expected value, so 50% chance of losing $10 and 25% chance of winning $10 and $0.
The bond will always pay you a dollar. If you ever want to just pick up a dollar, you can choose the [INAUDIBLE]. This is the information we give people before they play it. And we actually let them practice a little bit. And we talked to them through the practice session and all that stuff.
So pretend you're a subject in this study. I just gave you all this information. It's the first trial. What do you pick? Someone just tell me what they would pick. A stock? You'd choose one of the stocks. Anyone else? Bond? Why?
AUDIENCE: The expected payoff.
GREGORY R. SAMANEZ-LARKIN: In what? So say more.
AUDIENCE: I'm sorry?
GREGORY R. SAMANEZ-LARKIN: You said the expected payoff.
AUDIENCE: So you have equal probabilities--
GREGORY R. SAMANEZ-LARKIN: What's your background, by the way?
AUDIENCE: [INAUDIBLE]
[LAUGHTER]
GREGORY R. SAMANEZ-LARKIN: So you're saying [INAUDIBLE]?
AUDIENCE: [INAUDIBLE] bond.
GREGORY R. SAMANEZ-LARKIN: And because?
AUDIENCE: Of the expected payoffs.
GREGORY R. SAMANEZ-LARKIN: For the buyer? Yeah. Because its expected value on-- and what is a good stock?
AUDIENCE: [INAUDIBLE]
GREGORY R. SAMANEZ-LARKIN: That's right. OK, so the expected value of the good stock is plus $250. The expected value of the bad stock is minus $250. And there's a 50/50 chance that this triangle stock is going to be the good one. So 50% of the-- anyway, so that turns out to be $0, so the bond is $1. All right.
So then what do you do as you move through the task? And that's turns out to be the optimal strategy. So the other thing I'm going to show you today are two examples of ways that we, I guess, score behavior, right? So this first example is going to be we write down a model of the way to maximize earnings in this game.
And then we look at how people are not like that. And when they deviate from that model, we're going to classify that in a few different ways. And we're going to relate that to some specific brain signals. The second approach, which I'll talk about in the next study, is that we write down a model, but it has some flexible components.
And then our exercise is defining what the value of those knobs are. And that'll become more clear when I talk about it. So we use these combination of approaches in different settings. So let's see. Do I even have this? OK. So is there anyone with a finance-- this may not even be the right screen.
Maybe what I should just say is the optimal strategy is essentially you choose the bond and you pay attention to the market, right? So you start out choosing the bond and you watch. And you just keep picking bond until it becomes relatively clear to you.
And there's a very specific number on what relatively clear is in this model, that one of the stocks is the good one. You switch to it and maximize. You just pick that stock every time, unless the expected values change. And then maybe you'd switch back. But that's the idea. That's the optimal strategy-- bond, bond, bond, bond, bond. And then it's like, oh, that looks like the good stock. I'm just going to keep picking that.
Do people do that? No, not really. People do that, actually. So basically, this is just saying when the expected value of one of the stocks exceeds expected value of the bond exceeds 1 or the same way of saying that is when the expected value of the good stuff-- the chance that one of the stocks is the good stock exceeds 0.7. The probability is at least 0.7, then you switch.
AUDIENCE: Do people know that the [INAUDIBLE]? So they know--
GREGORY R. SAMANEZ-LARKIN: Yeah. So they get all that information about the probability distributions. They're also told that the task is stationary. Not in those words, but they're told that when they're in a round of trials, one of the stocks will be randomly assigned to be good. And one will be bad. And it will stay that way. There will not be shifts. There won't be changes.
But periodically, we stop the task. We ask them a few questions, and then things get reshuffled. And it might be the same stock, but it might be the other one. You just have to start over with it. So we start them over, learning many, many times in this. All right. But is the optimal strategy clear to people, what you should do? OK. All right.
So what do people do? All right. So these traces-- I don't know why I did dashed lines with dots in them. But anyway, so these dash lines are basically an average of across a bunch of sets of blocks, what the optimal, what this-- oh, by the way, this little-- this risk neutral agent updates according to Bayes' rule, so it's a Bayesian learner.
And so what this model predicts if you average across a bunch of different types of blocks is that you start out with a very strong initial preference for the bond. But that preference falls off as it becomes clear which stock is a good one, so if you just look at that background where the dash lines are.
And then your preference for the good stock increases over time. And then you hopefully never choose that bad stock at any point, but people do. So anyway, the younger people-- and again, this is a sample between the ages, I think, of 20 and 85. And it's a continuous age range.
And we've just semi arbitrarily split them into young, middle age, and older adults. And we just did this in thirds. But this is approximately people in their 20s and 30s, 40s and 50s, 60s, 70s, something like that. It's slightly shifted, but that's about the idea.
And so we just plotted these three age groups to show you how well their behavior corresponds to what this model is doing. This is a little bit misleading, because the model actually is pretty stepwise. So on any individual given block, people will choose bond, bond, bond, bond, bond and then switch completely to stock. But when you average across a bunch of different trial types, you get these curves.
So anyway, the younger people start out with strong bond preference. They're really good at detecting the difference between the good and bad stock as they learn in the task. The middle-age adults-- not quite as a strong initial bond preference, but they're really good at detecting the difference between the good and bad stock.
The older adults-- even weaker initial bond preference and aren't quite as good at detecting the difference between these good and bad stocks. There's a lot of things going on here in differences in behavior. So we're going to try to dig into some of this a little bit. All right.
So the one thing we can do is just look at in total the percentage of time that people deviate from this model, the choices that they make that aren't in line with what this model is doing. So this is going to be the proportion of total choices. And "optimal" is in quotes, because we could have a big debate about what's optimal. Because if you're acting according to your preferences-- and that could be optimal, even though-- anyway, so hopefully, we don't have that debate right now.
So we're just going to talk about deviations from this model, right? And we're going to reserve judgment. So basically, at older ages, they're more often-- their choices are more frequently not lining up with what this model would do. Well, we can break each one of these little gray dots.
So each dot here is a person. And we can break each one of these little gray dots into three little dots, which are three different kinds of-- three different ways of deviating from this model. And I'll talk about those. This is another cool thing about this task.
AUDIENCE: [INAUDIBLE] to make sure-- risk neutral is assumed and linearity?
GREGORY R. SAMANEZ-LARKIN: Linearity in what sense? Of the probability?
AUDIENCE: You're not putting any kind of curve.
GREGORY R. SAMANEZ-LARKIN: That's right.
AUDIENCE: Linear? OK.
GREGORY R. SAMANEZ-LARKIN: That's right.
AUDIENCE: And it's risk neutral?
GREGORY R. SAMANEZ-LARKIN: It's risk neutral. But, yeah. There's no non-linear probability weighting function or any-- yeah. OK, so in that previous plot, every single gray dot was a person. Now, every person has three dots, one in each color, right?
So now, we're dividing those deviations into three different categories. And the three different categories are bond mistakes. This is what Camus initially called a risk aversion mistake. This is choosing a bond too late. So you should be in the stocks and you pick a bond.
So this is, I think, pretty clearly risk aversion. This is you just don't want to take a chance. You're going to be a little conservative. And that shows no age effect whatsoever. So this is, again, why I thought that I was doing the wrong thing here. But there's no effect there of these kinds of risk averse-- there's huge variance, also, by the way, if you at that.
But where we do see the age differences are in both of these types of stock deviations, so both early in the round. So early would be older adults are picking the bad stocks sometimes, which is pretty bad, but then also are just picking one of the stocks in a later round. But then late in the round, they're picking actually the bad one. These are the really bad mistakes in this task.
And actually, if you see these blue dots, there are lots of people that never do that ever. They never pick that bad stock, which is good. This person is doing really bad in the task. They're losing lots of money. So anyway, we get a lot of variance in this.
So I think this is a feature of this task. You can get an independent-- two different measures of how risk seeking someone is in some moments and how risk averse they are in other moments, right? So it's not like you just get one score. This task allows you to be risk seeking or risk averse at different moments in time. OK, so I mentioned that half of the people did this study. So this is 110 subjects plotted right here.
Half of the people did this study while having their brain scanned. Half just came in and played it on a computer. And part of the reason we did that was because we thought, well, the average older person in the community who is going to come and stick their head in a magnet and play this game-- maybe they're just not representative of the risk preferences that were-- and so we thought we'd recruit a subgroup of people that just came in to do it behaviorally.
And essentially, I could have also just plotted this as the behavioral subset in the fMRI subset. But all the slopes are essentially identical. The effects are basically the same. So it didn't really seem to matter. We don't think we got an especially risky subset of older adults, at least specifically in the fMRI part. All right.
So this next part is something a little different. So I mentioned we collect MRI data. And we did something which, at the time, I thought was a little bit crazy with the fMRI data. It turns out that at least some people think it's not so crazy now. But at this time, I was in grad school getting this data.
I was halfway through grad school. And I was reading tons of stuff, reading lots of stuff. And I found these really interesting papers from [INAUDIBLE] and [INAUDIBLE] on what's going on, how some of these fluid cognitive changes might be related to changes in neuromodulatory systems in the brain, specifically dopamine.
And so one theory was that the dopamine system just gets noisier. So the firing is noisy. Or there's just more noise, more signal variability, potentially, in the dopamine system as you get older. And this accounts for some of these cognitive errors that you see. Purely theoretical. Almost zero data on this, so beautiful simulations. It looks really convincing, almost no data.
And I just thought it was a compelling idea. So we're doing fMRI. We're not measuring dopamine, right? So that should be made very clear. But I thought, wouldn't it be really interesting if there was some way we could take these ideas and test them empirically?
And so the idea we had was, well, we've got signal over time, right? Why don't we just look at the variability of that signal? And this is usually something that's essentially discarded in fMRI analysis, right? So people compare the mean for different conditions of, what's going on in this brain region when this is happening versus that's happening? And do direct tests of those means.
But people don't look much at how the signal varies, right? You just think of the variance as the noise component. And you're going to-- that's what you're dividing your signal by or something. You just don't care about that stuff. But we think there's maybe something interesting there.
And so what we did was if you think about-- so in fMRI, you basically have a Rubik's Cube of data, essentially. You've got all these little cubes of data in each one of these little cubes. You've got to measure a brain signal over time, right? So people are playing this game. And every two seconds, we get a measure in each one of these little parts of the brain. And so we've got this time series of signal from each of these parts of the brain.
And what we decided to do was quantify the variability of that signal using something like standard deviation, but not exactly. The problem-- I don't know if I should get into the technical reasons why we didn't use that. But essentially, we found a better statistic for this application, which is mean squared successive differences-- old von Neumann proposed statistic.
And by the way, a little bit of-- this is another just career aside. At the time I was working on this separate project on emotional experience in everyday life, we had longitudinal data where people were rating how they felt throughout the day. And then every five years, they were sampled.
So it's also time series data, right? But there are these mean fluctuations. So the feature of this statistic is that it can handle fluctuations in mean changes over time. And so we were doing this totally separate project where we were using this signal to look at variability. And I was like, oh, this is a cool statistic that I could use in this completely other situation.
So anyway, just keep an eye out for those things. All right. So what we did was we quantified in each little part of the brain, in each voxel of the brain. We quantified the signal variability over time. So this is just a toy example for a part of the brain that would get a high score versus a low score. So one place where there's a lot of signal variability and one place where there's less signal variability. OK.
So at the time, no one was really-- we hadn't found a paper on this. So we thought we were just coming up with this. It turns out that actually in Toronto, they were doing the exact same thing. And I'll tell you more about that in a second. So what we did was we looked at just-- so set the behavior aside for a second.
We looked at how the signal variability in different parts of the brain might change with age. And so what I'm going to show you are some brain maps where hotter colors would mean that older adults have more signal variability in these regions, significantly more single variability compared to younger people. And then cooler colors would mean older adults have less signal variability compared to younger adults.
And what we found was that in these striatal regions, where we think there are reward-related signals and learning-related signals that are relevant for these kinds of tasks, we saw increased signal variability with age, so potentially excessive. Although, we don't know exactly what the right level of variability is.
So older adults-- more single variability. Well, is this related to the behavior at all? Maybe this is just an interesting age thing that's happening that's independent of the task. So I told you that older adults are more likely to make these stock mistakes. But if we take a measure of the signal variability from the specific part of the ventral striatum, it actually is also correlated with making these stock mistakes controlling for age. So this signal variability in the brain is directly related to making these errors.
So as I mentioned, I actually went to human brain mapping one year. And there was this other grad student, Doug Garrett, that had his paper on bold variability and aging. I was like, what? We're doing that right now. And we went and we talked to each other. Actually, both of our papers were in review at Journal of Neuroscience at the same time. And we were really hoping they would be published together, but they were published three weeks apart.
So anyway, Doug's findings are interestingly similar, but also different in some ways. So here are effects where we see age-related increases in signal variability in parts of the mid-brain and striatum. They also get those same effects. So in the mid-brain and parts of the striatum, you see increased signal variability with age. But you also get lots of this other stuff where there's less signal variability with age in older adults. And we didn't see that in our data.
So for one thing, it looks like replication, right? At least one other group is doing something similar. And they're seeing a similar kind of effect. That's always good to see. But I think maybe even more interesting thing about this, is this is rest. These people aren't doing anything. There's no money at stake. There's no stocks and bonds. They're not even pressing a button.
This is, I think, from a fixation baseline like a block design task. So there's just nothing. This is essentially resting state data in a way. People aren't engaged in any kind of task. So there seems to be something general about potentially these age differences in signal variability. OK.
Any questions about that or comments on that? I mean, there's a lot of things we could talk about. So one thing is that there's signal in that variability, but then there's also potentially noise. And we're not doing a good job of-- so one thing we could do is regress out all of the signal that we can assume and then see if there's residual variability. But anyway, since then we've thought of a lot of good ideas.
AUDIENCE: [INAUDIBLE] question-- what are the predictions of the simulation [INAUDIBLE]?
GREGORY R. SAMANEZ-LARKIN: So it's basically the positive correlation. It's that there'll be excessive there. Yeah, sorry. I should have tied that back together. So-- right. So this looked pretty consistent with that theory. So if we're going to make some assumption that-- I mean, this is a part of the brain that's covered in dopamine receptors.
So if we want to link back to that theory, which we were hoping that we could do, is, yeah, that's the idea. That theory predicts that there's increased signal variability. And that's what's accounting for these potential cognitive errors and similar thing in our task. We have increased signal variability with age related to more mistakes in the task.
AUDIENCE: Two quick questions. With the [INAUDIBLE] data, did you also look at signal complexity in addition to variability? And also, in the sense of frequency decomposition, so doing a [INAUDIBLE] and looking at what the content was and frequencies in the regions.
GREGORY R. SAMANEZ-LARKIN: No, we didn't do that. I mean, we do a little bit of temporal filtering, but I think out of the range. There's still probably lots of interesting stuff there.
AUDIENCE: Yeah, exactly. [INAUDIBLE] But then also, did you do basic resting state function [INAUDIBLE]?
GREGORY R. SAMANEZ-LARKIN: No. So we have no resting-- so I've only recently started collecting resting state data in the last two or three years. And in none of these studies did we have resting state data. Yeah. OK, so-- oh, good. I'm doing good time wise. All right.
So the other thing-- so as I mentioned-- so this isn't-- there's nothing really very specific. Even though this is related to the task behavior, there's nothing necessarily specific about the signal variability. It's not that we looked at it while people were thinking about making a choice, when they're anticipating what may happen, or when they get feedback. Is there just lots of feedback-related variability?
We didn't do a good job of really figuring out which part of the process of this kind of learning-based risky decision-making was this affecting, necessarily, right? There's this general link between the signal variability and behavior. So after that, we did other studies which were similar where we used similar tasks, but we didn't have a bond.
So let's think about it as you just have two stocks. We actually didn't call them stocks. Because after a while, we realized there were some problems with that. When we called things stocks and bonds, people walking with all of this outside knowledge of what stocks are like and what bonds are like.
And this is also-- we ran these studies between 2006 and 2010, so there's all kinds of crazy-- and people are like, bonds aren't even safe anymore. So you're telling them like, well, they are in this specific study. So there was sometimes these periods we had to negotiate.
So anyway, we just got rid of the stock label at some point. And so we were just looking at these two different options that had different probabilities of paying different amounts of money, right? Same exact context, but without all that baggage that the labels bring.
So in those kinds of tasks, we looked at how well-- so we looked-- we went specifically after a component of this process that we thought was relevant, which is the learning component of this. Who's heard of Rescorla-Wagner learning in here? Oh, good. So who's familiar-- who knows anything about reinforcement learning or has heard about reinforcement learning? Good. OK, a little bit of familiarity. OK.
So the concept of reinforcement learning is pretty similar to what we just described. You do something. You get some feedback. And you make-- you update your expectations about what will happen the next time you do that thing, or see that thing, or encounter that thing, or whatever it is.
And so at the core, this model basically specifies how this learning process works, that you get a little bit of information and you qualify that information with respect to what your expectation was at that moment in time. And this is called a prediction error.
So you get some specific outcome like, you just won $5. But the expected value of choosing that at that moment in time might have been $2.20 or something. So then there's $5 minus $220. That's the size of your prediction error. So there's some deviation between what just happened relative to your expectations.
You're going to take that quantity and incrementally update your future predictions about what will happen when you encounter this thing or do this thing or whatever. And the way that's updated is according to some learning rate, which can vary freely across different individuals. Some people might update really fast. Some people might update really slowly. And we allow people to have different learning rates.
And so we basically have this series of two equations that describe how people play these kinds of games with this little dial, which is learning and then another dial, which I won't describe, of how quickly people update. And then what we look for in the brain specifically is this learning-related signal, this prediction error signal. Because we think this is the teaching signal.
This is the core of the learning part of this kind of a task. It's like, what just happened? How much better or worse did the world just get, based on what I just did? And how am I going to update the future? So we think this is really a key quantity in learning.
So when we look for whether there are age differences in this kind of a learning-related signal, we actually find that there's less of this learning-related signal in older adults. So especially in these medial prefrontal regions, but then also in ventral striatal regions, we see less of this prediction error-- so this is a prediction error-related signal in older adults, so less of this teaching signal.
And in fact, what we do see there is something that looks more just like a magnitude signal. It's like, sweet. I just won $5. Not like, how good is $5 relative to what you were expecting on this trial, right? And so it's that qualification of that signal that is not as strong in the older adults. All right. Well, it turns out that just if you're going to invest in scanning older people, sometimes you want to rescan them, because they're good subjects.
And so quite a few subjects did both of these studies. So the study that basically just has the two stocks, but they're not called stocks-- done a really careful job of looking for a specific learning-related signal in the brain and find out that there's less of this learning-related signal, especially in this medial prefrontal region and then this other task where there are stocks and bonds.
And people are making more mistakes choosing the stocks, which is a learning dependent choice. And we see more single variability in the striatum. So there are-- I don't know-- something like 20-something subjects, I think, that did both parts of this. And it turns out that these things are pretty decently well correlated. So a quarter of the variance in one is accounted for by the other.
So essentially, people that have a better discrimination between positive and negative prediction errors, so more sensible of this learning-related signal, have lower levels of the signal variability in the stadium. And on the other side of this being people with high levels of signal variability have a less accurate or a stranger learning-related signal in the medial prefrontal cortex.
The interesting thing about this is that these parts of the brain are really highly interconnected. So from the midbrain, there are direct projections, up into these striatal regions. There are also projections and then from these striatal regions, back into the pallidum and thalamus, and into medial frontal cortex, and from medial frontal cortex back into striatum.
So there's this interconnected series of loops between these frontal and striatal regions. And so we're seeing effects in both places, right? And so we thought, maybe there's something about the communication between these two parts of the brain that's accounting for these age differences in learning.
So we did diffusion tensor imaging to look at white matter connectivity or white matter tract integrity. I don't know the right word I'm supposed to use these days for it. But so this measure-- fractional anisotropy, so a proxy we hope for, like the strength of the connections between these different parts of the brain.
And where we found age effects so that where older adults had potentially weaker connectivity or reduced integrity of the connections was between the thalamus and prefrontal cortex and prefrontal cortex and the striatum. So this little bit of the loop that goes through the medial prefrontal cortex was lower in older ages and also was associated with learning.
So having stronger connections, if that's what this signal is, is associated with learning better in these two stock tasks. And this effect holds, controlling for age. So it seems like there is something about the communication between these two different parts of the brain and something about the role of the medial prefrontal cortex that's potentially messing things up a little bit in this environment.
So this first part of the talk is really about where older adults are doing slightly worse than younger adults. And I'll say slightly. Because older adults actually learn just fine if you give them enough time. In most of these tasks, you've got 10, 12, 15, 20 trials to try to figure out a relationship. And depending on the version of the task, that's sometimes harder easier.
And so we notice that older adults are doing slightly less well. But actually, we've done since then versions where we just give them three times as many trials. And they actually catch up just fine and do just as well as the younger adults. So there's just a little bit of a slowness in learning. It's not that this is some wildly horrific cognitive loss.
It's a pretty subtle effect. But basically, what we've shown is that we think that this is really about learning, not really about risk taking. But it affects risk taking, right? It affects the risky decisions that you make. And we think it really-- so we've identified a few things as functional signal variability. Is it related to structural connectivity?
And then what I haven't talked about today-- but if we have a little time at the end, I can talk about. It was one of my favorite studies to run, which was a study which I was working on my dissertation. And one of my committee members said, oh, you know what you really should do? And I was like, I don't want to do that. And it sounds interesting. But I have too many other things I want to do right now. And he's very insistent that I do-- so this is Anthony Wagner.
And I selected him as a committee member, because I knew he was tough. And then I was like, oh, he's being tough on me. This is what I wanted, but I don't really want it. So we did the study anyway. So part of this study was, how can we-- all right. What time is it? I can say a little bit about it. OK.
So part of the idea was that maybe there's nothing new here. Previous to doing all this work, there were a couple decades of research in the cognitive neuroscience of aging, showing that, like I started out talking about, these fluid cognitive deficits.
So let's take working memory, for example. So older adults do a little worse on working memory tasks, sometimes much worse. And that's been linked to function of lateral prefrontal cortical systems and various other parts of brain. And so maybe you're seeing these differences in these striatal systems. And it's totally-- it's just because older adults are trying to do this stuff with working memory.
They're not doing this kind of reinforcement learning stuff that you think they should be doing in this task. Instead, what they're trying to do is, all right, how many times does this one win? OK. Now, it's won three times. And it lost twice. And then this other one won two times.
And then that's a pretty taxing way of doing it. But doesn't mean that people aren't doing it that way, right? And so a hypothesis was that this could just be another working memory deficit, and that these systems aren't even what are being used. And so you're seeing less of this learning-related signal in the striatum, because people aren't doing striatal learning. The older adults aren't doing striatal learning in this task. All right.
So that was an idea. So we thought, well, If this is potentially are working memory dependent, let's just wipe out working memory and see what happens. So we had this hypothesis that there might be something we could do to keep working memory busy and occupied. That would just wipe out performance and we thought maybe disproportionately for the older adults.
But then instead of just being mean to people, we thought, why don't we also try at the same time to see what information we could provide to people to help them make better decisions? So if we're going to be mean, we should also try to do something nice.
And so we have this-- I think I have some slides in this buried in the end. And we have time right now. I'm talking fast. You guys aren't interrupting me enough. But that means you get to see more data. OK, so here's what we did. And this is a total detour, but I think it's a fun one. OK.
So they had people play the same task. But while they're playing it, they're listening to this series of tones-- (SINGING) doo, doo, doo, doo, doo, doo, doo, doo, right? This is it. This is pretty obnoxious. It sounds like that. And so what you're doing, while you're playing this game is keeping track of how many low and how many high tones you hear.
Because at the end of the trial, it says, how many high tones were in that series? And then you say-- and we constrain it to three different choices. And they have to pick which thing-- which one they thought it was. And they're incentivized to do this. So there's a $10 bonus that they can-- oh, I don't know if I said this. But everything we do in the lab is totally incentive compatible.
So it's all for real money. This is a real $10, a real $22. All this stuff is real. It's small stakes, right? People make at best $100 and something dollars on these tasks, which is not insignificant. But it's not whether you're going to take your $800,000 investment and get an annuity, right? So it's maybe not that.
But either way-- OK, so people are making money on these choices, but they're also making money on this. So there's a $10 bonus. And your accuracy in this condition can determine how much of that $10 bonus that you get, so equally incentivized. I mean, I guess a little bit less money is at stake for those. But I don't know that they have a sense of that when they start the game.
They're basically told to do both of those things as best you can and try to do them at the same time. So don't try to alternate where you really focus on one or the other. We really want you to be deeply engaged in both of those things at the same time. All right.
And then the thing that we did that we hoped was nice was that we called-- this is hard to see, but we called this the ticker condition, which is essentially you see-- this is you open up-- now, I used to say a newspaper, but now, you just have an app.
But you can see what's been going on with these assets recently, right? And all the information-- so this is probably a little more like real life, actually. It's like here's the historical performance of these different things is right here. You don't have to keep this in mind.
Now, we don't think these striatal learning systems are doing this thing anyway. We think they're just carrying one estimate of value updating it, carrying that single estimate through time. But either way, this might be how we make financial investment decisions in life. And so we thought, let's just show them the history and then let them make the decisions.
The other thing we tried to do was have a simplified version of this so that they don't have to see all these little bits. So you can't really see it. But there's a big plus, green plus if it won $10, a big red minus if it lost $10, and then a flat line if it was a $0. And then this little flat line with a tiny plus on it is a $1 for the bond.
So it's still quite a bit. And if anything, it's actually adding more information to the screen, right? And so it's unclear what people will do with this information. So we try to do a summary version of it. And so we had this little thermometer, which is the higher the green bar, the higher the expected value, the positive expected value. The lower the red bar, the worse the negative expected value at this moment in time.
And this is hard to explain to subjects, actually. So this is exactly our little Bayesian updater, this little thermometer. It's exactly doing Bayesian updating. So this should be the best predictor of what you should do. I mean, you should just follow this thing, but people were skeptical of it.
And they told us that it led them astray sometimes. And it's really hard to communicate independence to people. But I also teach statistics at Yale. But, yeah, it's really hard to tell people that. Even the good ones are going to lose sometimes. So it's not like we were misleading you. It's like, if you just stick with it, you'd still end up-- anyway, I don't want to get into all these discussions of subjects all the time. All right.
So what did we find? So here's the baseline condition. I plotted things a little bit differently for you guys. So the older adults made fewer choices in line with that model. And so then what we're going to see is that dual task condition. But now, there's going to be positive and negative pieces of the scale.
And then zero would be no change compared to baseline. But it's basically like, what's the effect of the dual task? So is it the case that the younger, older adults get better in the dual task on making these stock and bond choices or get worse? What do people think? And I'll tell you, everyone on the committee predicted this inaccurately. So I was wrong about this. Everyone was wrong about this. What do you think? You maybe-- seen these data before as well.
AUDIENCE: Can you selectively [INAUDIBLE] brain and memory and force people to [INAUDIBLE]
GREGORY R. SAMANEZ-LARKIN: So you predict they get better. Yeah, I wish. So basically, here's what happened-- nothing. Didn't do anything. It did absolutely nothing. Yeah, so the fact that we thought would be the coolest effect was if this made the older adults get better. Because potentially, they're trying to use multiple systems.
And actually, if we occupy this one that isn't really optimized for this kind of task, you can let the striatum do its thing. But that didn't work. But anyway, this is super interesting. Because people came out of this condition and were really irritated. So if anything, this was-- if you want a negative mood induction or if you want to annoy the neighbors around your lab, have this task going all day long with (SINGING) doo, doo, doo, doo, doo.
Afterwards, we were like, we should have used headphones for that study, I think. Because the neighboring labs were like, when is this study going to be over? This is tones all day. So anyway, people said that they just couldn't keep track of things, and that they did really bad.
The interesting thing is that they were engaged in that secondary task. So accuracy on counting tones was not significantly different between the two age groups. This is just behavioral, by the way. There's no brain imaging here. But this is the interesting part. So periodically, we stopped them and we say, all right, which one is the good one and which one is the bad one right now? Right? So they played a bunch of trials.
And then we say-- so this is a test of their declarative knowledge at this moment in time. This gets messed up. So people are less able to tell you, don't tell you as well. They're right when they communicate that, I couldn't figure things out. That's what they're communicating. They have the sense that they can't figure things out. They're not doing as well, explicitly identifying which is the good thing.
But actually, it doesn't matter at all for their individual choices that they make. So there's this interesting disconnection between your awareness of how you're doing and how you're actually doing. So we thought that was pretty interesting, even though we didn't get these cool age differential effects like we thought we'd see.
And then basically, these two other conditions helped both age groups, which was nice. And we call these things decision aids. The reason I don't make too big of a deal about this is because these things that we used-- I mean that ticker thing is what could be used in the real world.
But the idea of developing some simple little graphical meter is pretty unrealistic in most real life situations. So basically, we think that even though we think these are a hint at potential decision aids, we don't think they're very applicable. We think this is basically many, many, many, many steps removed from actually coming up with some helpful intervention for these kinds of decisions. All right. All right.
Let me go back to where I was. OK. Here's where I was. OK, so we've done quite a bit of stuff on what's going on with older adults. It's these learning-related risky decisions and older adults being slightly worse at them in trying to identify the neural circuitry and exactly what's going on in these natural systems. All right.
And we think this lines up pretty well at the very beginning when I was talking about these fluid deficits. And this isn't these classic fluid deficits that you think of as working memory. And I hope that was convincing when I just showed you that. This is maybe not just a working memory deficit or some attentional deficit, that really there's even a different part of the brain where we still think this is a bit of-- this is a fluid cognitive deficit.
Because this is a situation where you dropped in. You get some information, but you've got to figure out things really fast, right? And so you need these flexible, fast cognitive skills to do it. So we think this lines up more with that cognitive decline side of the story that I was telling at the beginning, these fluid cognitive deficit.
But there are lots of situations where older adults make better decisions than younger adults. So I started out saying that experience would predict that you'd get better and better and better at making financial decisions as you get older. And one example of where older adults make better quantitatively better decisions is an intertemporal choice.
So this is you could have $4 now. Or if you wait two weeks, I'll give you $9. Right? So these kinds of things-- there's different amounts of money at stake. And you have to tolerate different lengths of delays to decide which of those things you'll go for.
And [? Carinna ?] actually has this really nice review from a few years ago, basically showing that lots of studies show that older-- and this totally baffles biologists and economists that especially in these monetary contexts, older adults are more likely to take the delayed option.
So they'll tolerate the delay and wait it out a little and make a little bit more money. Does that seem like it makes sense or doesn't make sense to anyone? Or is it clear why that would be totally counterintuitive? Does anyone have any thoughts on it? Yeah?
AUDIENCE: [INAUDIBLE] So just to make sure that I'm understanding, so an older adult is going to be more tolerant of an intertemporal delay is what we're saying. So I mean, this goes into just the way that the brain constructs perceptions of time. But if it's not based on one's own recorded experience, then the relative slice of time that you're talking about-- the denominator is bigger. And so the overall fraction is going to be smaller if that's-- I mean if--
GREGORY R. SAMANEZ-LARKIN: Oh, so you're thinking about a lifetime account of it. This delay relative to the life you've experienced or something [INAUDIBLE].
AUDIENCE: Yeah. Well, I mean, just that's the way that the brain is constructing senses of time, right? But it's based on--
GREGORY R. SAMANEZ-LARKIN: Someone was saying something about mortality. Sorry, sorry.
AUDIENCE: Let's have fun, [INAUDIBLE]
[LAUGHTER]
GREGORY R. SAMANEZ-LARKIN: Yeah, yeah, yeah. Yeah. Yeah. So say what you were going to say about that.
AUDIENCE: I was going to say I was thinking more about mortality, that people who are older-- that they have more patience. They may not have as much time left at their disposal.
GREGORY R. SAMANEZ-LARKIN: Well, right.
AUDIENCE: [INAUDIBLE]
GREGORY R. SAMANEZ-LARKIN: So that's exactly right. So that's why--
AUDIENCE: [INAUDIBLE]
GREGORY R. SAMANEZ-LARKIN: Yeah, so that's why I'm saying the biology economics prediction is that you should be super present oriented.
AUDIENCE: [INAUDIBLE]
GREGORY R. SAMANEZ-LARKIN: Yeah, exactly. So your time horizons are short. You should be extremely present oriented, so you should have a huge "now" bias. You should be a tremendous temporal discounter, right? You should really discount any delay, but we see the opposite. And I think we have a decent story about it.
So I just want to point out that these kinds of intertemporal choices are this side of things where you get all of the information you need on one screen. You don't have to necessarily within this task learn the value of these things. So they're simpler in the way that you could just see a choice and make a decision. You don't have to go through this learning process.
And actually, it turns out that maybe some of these decisions are somehow dependent on your accumulated experience with these things. Maybe you're hinting at some of this, so depending on your experience with these kinds of things. And we'll come back to this as a potential account of this.
And I'll also tell you an opposite account of the findings that I'm trying-- that we're battling out in the literature right now. So here's what the tasks look like. So again, we have people. And this is actually just younger and older people. We didn't really have much funding for this when we started it. So we just did this extreme groups comparison-- people in their 20s and 30s to people 65 to 85.
But people are making these choices. You could have $17.63 today or $20.27 in two weeks. And you look at these choices. And so here is an example, again, like this reinforcement learning example. We've got this. We're going to call this a value function, so the value of some option. So for a given reward and a given delay is going to be a function of-- actually, this model doesn't really matter too much. Maybe I'll just show you the data.
But basically, the idea is that there's going to be a component of the model which has a high discount rate that's really sensitive to the present and a component of the model that has a low discount rate, which is a little more tolerant of time delays, and that your choices are a function of the combination of these two different functions.
There's been a big argument in the neuroscience literature about whether this is the right model or whether another model, which is just-- so basically, if you think about the combination of a super steep exponential and a shallow exponential, ends up with like a hyperbolic function, right? So people call this a quasi-hyperbolic model. But some people have argued that a hyperbolic model better characterizes brain function.
Turns out, that totally doesn't matter whatsoever for our findings. So this was a study that we did with Sam McClure who had historically been a fan of this model. And so that's what we used. But I'll show you it doesn't matter which model you use. Actually, did fit a hyperbolic model too, but--
So I'm actually going to skip this. Because it's better to show you the signal. Things are much more clear. Basically, we saw more delta related to low discounting-related value signals in the striatum in older adults. But what that-- so how we ended up with that-- if you just look at the signal. So this is just a time course on an individual trial.
Some trials-- there's a today option, so you could get some money today. And other trials-- even the sooner option is delayed by two weeks. On trials where there's a today option, the young people show a positive signal change, some more signal in this part of the ventral striatum.
But even if things are delayed even just two weeks, it's nothing. I don't care about that. It's not even real money-- two weeks away. And so if you look at the older adults, you see something pretty different. So both for options available now and things available later, you see signal increase.
Basically, what you're seeing here is no time discounting, right? So this is just, again, a magnitude signal, more of a pure magnitude signal. It's like, oh, it's either $18 or this is $22. And I'm not going to worry too much about the delay. And actually, people told us that afterwards after playing the game. It was like, I didn't pay as much attention to that, the delay of things.
Because it's like, what's two weeks, or four weeks, or all that, right? And we heard that disproportionately from older adults. So we're not the only people to show these age by delay interactions, so both-- this is from our study. The reason these have-- this is-- maybe look weird to you guys, because there's people here.
This comes from another figure of a larger figure I made, which has data from rats. So there's a little rat in another one that you don't see. But I don't always just put weird people on figures. So these are two human studies. You actually see the same thing in rats, which is another story.
But basically, you see it. So here are our effects that I just showed you, but plotted the way that Ben Eppinger plotted his, just to show you how comparable they are. And so I actually like Ben's effects even better. I wish our data looked like his.
So he gets this huge effect in young people where there's for sooner options, much more signal than for later options. And then for older adults, he gets almost the opposite. And this would be consistent with no time discounting, because that delayed option is higher in magnitude.
So this is just a magnitude signal. This is what you'd predict exactly. The funny thing about this is that our discussion sections are completely different. We have the exact same data, essentially. I mean, there's some little qualitative differences, but this story is older people have no dopamine, lost dopamine. And they've lost this dopamine initiated impulsivity that we see in young people.
So they don't have this impulsivity. So it's a cognitive decline story, which just happens to serendipitously result in people making better-- making decisions that result in earning more money, right? So it's like, oh, isn't that great? You get a little stupider. But at least you make a little bit more money, or you can tolerate more time delays.
But there's a bunch of problems with that argument. So one problem is in many cases if you give people a dopamine agonist, they get more patient, right? So that doesn't really make sense. Well, anyway, there's a few other reasons. So Ben and I are great friends, but we argue about this stuff all the time.
So our account, our discussion section told a little bit different story, which is people have a lot of experience with delayed rewards as you get older. As you get older, you've realized interest rates, right? You made investments early in life. You watched them grow. You got to spend that money later. Or maybe you're spending that money now and enjoying it. So you just have more experience with delayed rewards.
And there's some data to suggest that older adults, maybe because of these experience-related reasons, get better at affective forecasting, so get better at predicting the hedonic impact of something that's going to happen in the future. And so we think this might be related, that basically older adults are just making better future value predictions about these different [AUDIO OUT]. And we think this might be related to experience.
Again, neither of us, Ben nor I, have the data to really establish which one of these stories is right. But anyway, I like this story. So I actually went to one of my colleagues at Yale that does studies with rodents. And we've talked about doing an experience manipulation to see how these things would change, but we haven't done that yet.
But anyway, I think our general interpretation is maybe the subjective value of these rewards is already established. You gain experience with this. And you learn to value delayed rewards even more as you gain experience with them. And so that might be what's accounting for a lot of what we see in the literature. Yeah?
AUDIENCE: Do you have any measure of the subjects' assets?
GREGORY R. SAMANEZ-LARKIN: We do.
AUDIENCE: Because-- I mean-- so the standard economic-- the economic story here would be the time shouldn't matter unless you're [INAUDIBLE]. And it could be the case that younger people are more likely to be included in [INAUDIBLE]
GREGORY R. SAMANEZ-LARKIN: Yeah, that could be. Yeah, could be. So, yeah. Yeah, that could be. So we can't deal too well with that. So we don't have a good measure of liquidity, I don't think. I mean, we have a measure of income. We have a measure of wealth overall.
I think in this specific study, we weren't looking at-- so more recently, we've broken down assets and debts into different kinds of categories, which maybe could give us a sense of liquidity. Because we look at housing, wealth, and all this other stuff. So we haven't been able to do that well.
You're bringing up a more general point which is a good one, which usually people stop me in the first five minutes to say-- whenever I talk about financial decision-making aging in this experimental stuff, say, yeah, but older people have more money. They're wealthier. So that's just-- could explain everything. And it could explain a lot of it. And we don't know. So we never have enough.
What we really need is lots of variance in wealth in a sample that doesn't vary much in age, right? And we never have that kind of power to really look at these things. But I think in some cases, a wealth effect would go in the opposite direction, in fact.
So in that first study with the stocks and bonds task, people who do better in that task actually have more assets. So people that-- they have-- so if we look at their long-term savings and just how much wealth they have at that moment in time-- so I don't know which direction it goes or whatever the causality is there.
But basically, there's this correlation. It's reasonably strong. And so older people have more money. So you'd expect if this is wealth related or something. Anyway, so I think sometimes it's not clear. Basically, the short answer is we haven't done a good job of measuring those things, so part of that could have to do with liquidity.
AUDIENCE: And the other way to go would be to go would be to do primary cohorts. So certainly, the behavioral economics research right now is very big on stop using money in [INAUDIBLE] tasks. Because money should not-- money trade-off should not [INAUDIBLE] preferences, unless [INAUDIBLE].
GREGORY R. SAMANEZ-LARKIN: Right. Oh, you mean because-- not because it's not the right-- I mean, the problem with that is that there are different patterns of behavior in different domains, right? It's not that your preference should be domain general, necessarily. That's one issue with this.
AUDIENCE: If you're not [INAUDIBLE], and you ask these questions, all that matters is the interest rate you get [INAUDIBLE].
GREGORY R. SAMANEZ-LARKIN: That's right.
AUDIENCE: If you're doing anything [INAUDIBLE]. People don't do that. I agree. But we don't know why they do that. And thus, it's hard to know how to interpret their behavior.
GREGORY R. SAMANEZ-LARKIN: Yeah. But I just want to say that I don't know that looking at a different domain, a non-financial domain is the right solution to that problem. But anyway--
AUDIENCE: Yeah, I agree on that point.
GREGORY R. SAMANEZ-LARKIN: So I mean, there is at least one aging study where they look at primary rewards. And there's no age effect in discounting, basically. I mean, I also have a different account of this, which is that your time horizon for financial investment extends beyond your lifetime, right? So you're normally thinking about passing your money on. And so you're really-- we're talking about a different time horizon now, which could affect these things. Anyway, [INAUDIBLE].
AUDIENCE: Yeah, I share your information. Because when I explain in class that your preferences shouldn't matter, [INAUDIBLE]
GREGORY R. SAMANEZ-LARKIN: Yeah. OK, all right. So I think these
AUDIENCE: [INAUDIBLE]
GREGORY R. SAMANEZ-LARKIN: Yes, go ahead.
AUDIENCE: [INAUDIBLE] Sorry. I know you said you wanted to get interrupted. One other way of getting at that, that I was thinking of might be to look at reward sensitivity more generally. I don't know if there's a way you could control for how much of the older adults and the younger adults that you're [INAUDIBLE] amount of money [INAUDIBLE]
GREGORY R. SAMANEZ-LARKIN: People have done that.
AUDIENCE: [INAUDIBLE]
GREGORY R. SAMANEZ-LARKIN: So, yeah. So they study them. I'm a co-author on one paper where we look separately at a task that just looks at your sensitivity to different magnitudes of rewards and then tries to see whether that sensitivity to magnitude is at all related to your preferences in a discounting task.
And so the first couple-- at least, first, one study and I think there's at least one or two other examples where people have done a similar thing. They found that people with more sensitivity to reward magnitude were more impatient in these kinds of tasks.
We find the opposite, actually, that people that are more sensitive to the magnitudes are more patient. So the literature is a bit mixed on this. And the reason I think-- is because at least two things matter when you're making these decisions, right? It's how magnitude sensitive you are, but also how time sensitive you are.
So even though we've got these nice models that have different discounting components, we really don't do a good job of trying to figure out which piece of it-- that magnitude piece or the time piece of it is what's really driving your sensitivity. Differential sensitivity to those two things is really driving it, which I think is a hard thing to do.
AUDIENCE: [INAUDIBLE] is one of your reward sensitivity measures. [INAUDIBLE]
GREGORY R. SAMANEZ-LARKIN: I mean, there's just--
AUDIENCE: [INAUDIBLE] that?
GREGORY R. SAMANEZ-LARKIN: In those examples where we look at the--
AUDIENCE: [INAUDIBLE]
GREGORY R. SAMANEZ-LARKIN: So we don't have a behavioral measure of reward--
AUDIENCE: [INAUDIBLE]
GREGORY R. SAMANEZ-LARKIN: --sensitivity in that example. It's just the neural measure, Yeah. Yeah-- in those other studies. This is some of the adolescent stuff that I did with Meg. So I don't always study adults. I guess sometimes I study kids, too. All right.
But I think that-- I hope that these are two different examples. I laid out at the beginning that sometimes there are these experience-related improvements or cognitive-related deficits. And so hopefully, these are two examples of that and some of the progress that we've made looking at the neural systems.
So what? Does this tell us anything about financial behavior in the real world? This is what's just been killing me lately. So this has really been bothering me. And every year it bothers me more and more and more. And I actually can't really take it anymore. And so I mentioned we did this one study. In that one study, we looked at how well people's investment choices in this little task are related to their long-term savings in life. And we got this nice little correlation.
But for the most part, if you look at the literature, it seems that no one really-- I shouldn't put words in people's mouth, but it seems that people don't really care much about this. Or at least nobody is really doing much of it. So we have these great tasks in the lab that isolate these very specific processes.
And have these beautiful models that break down all this processing. And we can quantify all these different components of how people learn, and how they decide, and all of these things. But we don't know if any of it has anything to do with how people make financial decisions with big piles of money.
And I don't even mean to emphasize the magnitude difference. Because I don't think that's the biggest issue. We just don't know how well any of this stuff relates to what people do with their money in the real world. And the reason I say that, I'm sorry, I can't take this anymore, is because this research is tremendously expensive, right? We spend a ton of money on this.
And so I didn't even talk a lot of what I do lately in the last few years is PET imaging. If you think fMRI is expensive, you haven't seen anything. So our fMRI sessions cost about $1,000. Each one of our PET sessions cost $4,000. And often, we do two PET scans for each subject, so we're talking over $10 grand a person. And a lot of this stuff is taxpayer funded-- NIH research.
And I just-- I don't mean to be dramatic. But I can't keep doing this stuff and spending millions of dollars on this stuff if it doesn't relate to real life behavior. So maybe this is a stupid career decision. But I have to pause. And I'm not really going to totally pause. My work is going to keep going. But for the next couple of years, what I'm really going to invest a lot of my time in is trying to connect what we're doing in the lab to what people are doing in real life.
So does this tell us anything? I don't know. Maybe. We're trying to do that. So we have a little bit of hints. We have a few papers. I think a lot of people assume that these are real financial decisions. This is incentive compatible. We hand out real cash. So this is real. Why would we see how it relates to real life decisions? These are already real decisions, right?
And I think that's just an assumption that needs to be verified. OK, so we're working on that. And maybe I'll say more about it if we have some time. I think we will have some time. All right. This other issue about the scalability of this. So I started with this idea from Greenspan that, oh, if we only had a way of telling how people felt and predict.
And then we could use that to predict how-- first of all, I think one thing I should have said about that is what he's saying is if we could do that, we wouldn't need these models. In fact, I think he said exactly that. But I think that's wrong. First of all, I love the models, too. Right? So I think we should use the models and the feelings. And we should just put everything together.
But I think what he's saying, really, is if we could get a population sense of what's going on with people, how people are feeling, then we could make better predictions about where the market is going to go. And so as I mentioned, with the stuff that we do is just with a handful of people, basically. At best we're scanning a hundred people or something like that in some of these studies. And this is just a random sample of people from the community.
And so can we make generalizations about what these people are doing, especially in the case of financial decision-making about what people would do at the population level in these kinds of situations? And I think there aren't any good examples, really, if we look in the branch of neuroeconomics, which is the more neurofinance-oriented thing.
But I think there are a couple of really inspiring examples that suggests we could do this on the more neuromarketing public health side of things. So a couple of examples. Does anyone know this? There's this Greg Berns' study about music. Has anyone heard of this? So you have? Oh, good. All right. Good.
So what they did-- and I don't even know what this study was initially on, but they had-- this is-- when was this? Early 2000s, or mid-2000s, or something. They had adolescents listening to little music clips. Who remembers Myspace? I mean, Myspace still exists, right? But when Myspace was really a thing, when every time you went to the movie theater, it'd be myspace.com/starwars or whatever. It was like every website was on Myspace or something for some reason.
But so when Myspace was a thing, people used to go and listen to music. And it was a good place to discover new music, right? Or your friend's band had their things on there, whatever. And so they took clips from Myspace from bands that were unsigned. And they had kids listen to the clips. And then they would rate. How much do you like it? How much do you think your friends would like it? How much do you think-- do you think this band is going to get a deal?
It wasn't that kind of question, but it was really predicting the future, like market impact of these sorts of things. And they did this study. And I think there was even a component of this study where then they got feedback about what their friends might have said.
Anyway, there was a different point of the study. But then later, they realized that some of those songs got hugely popular that they just happened to have in their study. And they're like, oh, cool. So now, we could go and look back at people's ratings on these things. And we can also look at their brain signal.
I don't know if I said this, but they were being scanned while they listened to these clips. And they could see if there was anything in their brain that's going to predict popularity at the population level. Turns out, it works pretty well. So people's ratings predict pretty well how well these-- how much money these different songs make eventually. But what adds significantly more variance to that prediction are these brain signals.
And so both behavior and these brain signals are explaining unique variance, potentially, in how well these songs did in the marketplace, which is pretty cool. It's 28 kids listening to some music in a brain scanner. And you can predict it. So it's like hit detector. Anyway, I don't know if you were there. The year that Greg Berns presented this at neuroeconomics, this was a talk everyone was like-- this is what everyone wants to do.
AUDIENCE: [INAUDIBLE]
GREGORY R. SAMANEZ-LARKIN: It was unbelievable. Yeah, it was-- people were blown away. So, yeah, that's a good point. So Valerie just said this was the high point for neuromarketing. But, yeah. So anyway, it's pretty compelling. But what's even cooler, I think-- so that's great.
You can sell things to people. And you can use this brain imaging stuff to predict what people will buy and all that stuff. There's really, really beautiful work from Emily Falk. And Elliot Berkman has been involved in this and Matt Lieberman, basically taking that same kind of approach, a neuromarketing approach to try to help people live healthier lives, right?
Instead of sell them stuff-- so we're going to-- I'm not going to make any judgments about the field of marketing. Marketing just makes transactions efficient and matches people with the products that they maybe sometimes don't even know that they want. But maybe we can use that framework to try to help people exercise more, or not smoke, or things like that. So there's a series of amazing sites. So look up Emily Falk some time. I think these are the coolest brain imaging studies lately.
AUDIENCE: [INAUDIBLE] We have about five minutes.
GREGORY R. SAMANEZ-LARKIN: Five minutes? Oh, OK, [INAUDIBLE].
AUDIENCE: In case anybody wants to--
GREGORY R. SAMANEZ-LARKIN: Is it 1:10? Yeah, I should. OK, so basically, very, very briefly, what they show is-- and there's three different anti-smoking ads. They show them to people in the scanner. The brain response predicts the population uptake of these different [INAUDIBLE] things, way better than the behavior does.
So I think there's a few examples that basically you can look at a small group of people and make a generalization about the population. There's a few examples. All right. So basically, just give me one minute to summarize. So essentially, what we're trying to do is figure out how the brain works, how it changes with age.
And, yeah, I'm not going to talk about this, because I don't have any time. One quick point is that I think the extent to which older adults or younger adults will be disadvantaged or advantaged in decisions really depends on the context. So there's massive context dependency here.
So the extent to which of the decision that you're making requires you to use these fluid cognitive skills or these crystallized cognitive skills. You're going to predict that younger or older adults will have different advantages there. So really, the goal of what we're doing is trying to improve health and well-being.
As I mentioned, some of the stuff we're doing lately is looking at how well some of these things generalize to different kinds of rewards, like social rewards and stuff that we're doing with [? Carinna ?] that's related to a little bit more health related in some ways.
And then as I mentioned, for the next couple of years, we're really trying to develop interventions and work together with industry to try to help people make better decisions. So I have to thank the National Institute on Aging, which has made massive investment in my career so far. And I hope that continues. But also, we've been supported by the FINRA Investor Education Foundation for a lot of this work. All right. Thanks.
[APPLAUSE]
AUDIENCE: So I think [INAUDIBLE]
GREGORY R. SAMANEZ-LARKIN: Thanks for stopping me, because I was getting excited about the [INAUDIBLE].
AUDIENCE: We love that.
AUDIENCE: So compared to a lot of [INAUDIBLE] how is the dopamine story hanging in?
AUDIENCE: Yeah, how's the dopamine?
GREGORY R. SAMANEZ-LARKIN: So our last-- what time is it? Our last subject is going to be run four hours for this first-- for this project. So we got this R01 to look at. So basically, yeah. We saw all of these signals that looked in places that might have something to do with dopamine.
So I actually went and did a post-doc with David Zald to learn how to measure dopamine, instead of just talking about it. And so while I was a post-doc, we wrote a grant. And so we're just now finishing data collection. So we'll have our last PET scan tonight. And so we don't know yet. But basically, we'll have PET dopamine receptor data and some fMRI data on some discounting tasks and stuff like that. So, yeah. We don't know. I'll know soon. In a few weeks, maybe.
AUDIENCE: So I really like your statement, the intention of taking what you've done here and tying it into actual [INAUDIBLE] effectiveness in financial decision-making. I have one concern about it. And I'm wondering if you share this, that is I'm worried that those individual decisions that you've analyzed don't contribute much to the variance in your ultimate financial success [INAUDIBLE]. Often the judgment is [INAUDIBLE] people as they get older all the time, financially or any other domain. I just can't do that anymore, which means--
GREGORY R. SAMANEZ-LARKIN: What do you mean?
AUDIENCE: Well, all sorts of things. Oh, I can't make those decisions anymore. I can't go shopping.
GREGORY R. SAMANEZ-LARKIN: So they'll defer.
AUDIENCE: So they defer. And so the question is, what do you do? You turn it over to index funds. Therefore, it's active. You trust this person [INAUDIBLE] turns out to be trustworthy or not. So it's vetted decisions that are [INAUDIBLE]. And you made some reference earlier on to scanning. I just wondered.
GREGORY R. SAMANEZ-LARKIN: Yep. No, I think you're totally right. The other-- a related-- I'll ask myself another question related to that, which is I think sometimes people bring up that it seems the most important financial decisions are almost made early in life, right?
So to determine your future financial well-being, it's really there's this critical phase potentially between 30 and 55 or something, where you really have to get your stuff together and get stuff invested so that you have time to-- anyway, so I think people have pointed that out. And I think that's totally right.
But I think-- so I'm not just studying-- so a couple things. I don't just study decision-making in old age, right? This is an adult lifespan approach. I think studying this stuff in young people is just as important. In fact, half of what FINRA Foundation's funds lately is trying to help young people make better financial investment decisions. And so I'm all about that.
So the other issue you're mentioning is that what's happening in the real world is all of this interesting stuff where maybe things get deferred or you ask other people to do things. So we are doing this stuff lately. So I have a data set, looking at financial fraud. It's actually funded by FINRA Foundation where we look at individual differences in older adults and susceptibility to investment fraud and things like that.
And the short answer is that it's super complicated so that there's not-- so I think there's this perception. So basically, the industry has gotten really involved in these. The financial services industry has gotten really interested in this stuff, because they think this is all dementia. And they think, oh, older people make mistakes. We're in this unique opportunity to-- these are all people that are-- have almost Alzheimer's disease or something like that, right?
And so not only can we help them, but we can also help their family figure out that this person has Alzheimer's disease and stuff like that. And I think when we've looked at this, actually-- so we've looked at actual fraud victims and tried to figure-- it's not that at all.
It's actually this interesting constellation of cognitive, emotional, motivational things that make you more susceptible. And we have a very small data set. So I don't know if we can make-- I shouldn't say too much about it. But, yeah, I think it's super complicated. And I think there's lots of factors that we aren't measuring right now, right?
So you're saying it's not we should just necessarily take this research program and try to get real world measures and expect things to correlate, right? Because there are other ways that people make decisions that we should be trying to account for. And I totally agree, so thanks.
AUDIENCE: The delegation part.
GREGORY R. SAMANEZ-LARKIN: Right. Yeah. And the question is-- especially related to the delegation thing, it's who's the right-- so this is a question. So the Office of Older Americans within the Consumer Financial Protection Bureau-- I've had a couple of calls with them. And this is one thing that they're really concerned about lately is like, who's the right person to help in some of these situations? And are people good at selecting that right person to help? And there's no research on that whatsoever right now.
AUDIENCE: [INAUDIBLE]
GREGORY R. SAMANEZ-LARKIN: Yeah? Sorry. Sorry.
AUDIENCE: I wanted to point out-- by the way, see that [INAUDIBLE] for the undergraduates in the audience? That was a pre-doctoral award. And so if any of you are interested in thinking about those things, those things are actually possible.
GREGORY R. SAMANEZ-LARKIN: Yeah, talk to me about funding if you want. I've written so many grants in my life.
AUDIENCE: [INAUDIBLE] Thank you so much.
Gregory Samanez-Larkin, Assistant Professor, Department of Psychology, Cognitive Science & Neuroscience, and Co-Director, Scientific Research Network on Decision Neuroscience and Aging, Yale University, presents a review of his research on the psychological processes of financial decision-making as the brain ages.