Tversky

How do people think about probability?

pnrj cognitive science, critique of neoclassical economics , , , , , , , , , , , ,

Nov 27, JDN 2457690

(This topic was chosen by vote of my Patreons.)

In neoclassical theory, it is assumed (explicitly or implicitly) that human beings judge probability in something like the optimal Bayesian way: We assign prior probabilities to events, and then when confronted with evidence we infer using the observed data to update our prior probabilities to posterior probabilities. Then, when we have to make decisions, we maximize our expected utility subject to our posterior probabilities.

This, of course, is nothing like how human beings actually think. Even very intelligent, rational, numerate people only engage in a vague approximation of this behavior, and only when dealing with major decisions likely to affect the course of their lives. (Yes, I literally decide which universities to attend based upon formal expected utility models. Thus far, I’ve never been dissatisfied with a decision made that way.) No one decides what to eat for lunch or what to do this weekend based on formal expected utility models—or at least I hope they don’t, because that point the computational cost far exceeds the expected benefit.

So how do human beings actually think about probability? Well, a good place to start is to look at ways in which we systematically deviate from expected utility theory.

A classic example is the Allais paradox. See if it applies to you.

In game A, you get $1 million dollars, guaranteed.
In game B, you have a 10% chance of getting $5 million, an 89% chance of getting $1 million, but now you have a 1% chance of getting nothing.

Which do you prefer, game A or game B?

In game C, you have an 11% chance of getting $1 million, and an 89% chance of getting nothing.

In game D, you have a 10% chance of getting $5 million, and a 90% chance of getting nothing.

Which do you prefer, game C or game D?

I have to think about it for a little bit and do some calculations, and it’s still very hard because it depends crucially on my projected lifetime income (which could easily exceed $3 million with a PhD, especially in economics) and the precise form of my marginal utility (I think I have constant relative risk aversion, but I’m not sure what parameter to use precisely), but in general I think I want to choose game A and game C, but I actually feel really ambivalent, because it’s not hard to find plausible parameters for my utility where I should go for the gamble.

But if you’re like most people, you choose game A and game D.

There is no coherent expected utility by which you would do this.

Why? Either a 10% chance of $5 million instead of $1 million is worth risking a 1% chance of nothing, or it isn’t. If it is, you should play B and D. If it’s not, you should play A and C. I can’t tell you for sure whether it is worth it—I can’t even fully decide for myself—but it either is or it isn’t.

Yet most people have a strong intuition that they should take game A but game D. Why? What does this say about how we judge probability?
The leading theory in behavioral economics right now is cumulative prospect theory, developed by the great Kahneman and Tversky, who essentially founded the field of behavioral economics. It’s quite intimidating to try to go up against them—which is probably why we should force ourselves to do it. Fear of challenging the favorite theories of the great scientists before us is how science stagnates.

I wrote about it more in a previous post, but as a brief review, cumulative prospect theory says that instead of judging based on a well-defined utility function, we instead consider gains and losses as fundamentally different sorts of thing, and in three specific ways:

First, we are loss-averse; we feel a loss about twice as intensely as a gain of the same amount.

Second, we are risk-averse for gains, but risk-seeking for losses; we assume that gaining twice as much isn’t actually twice as good (which is almost certainly true), but we also assume that losing twice as much isn’t actually twice as bad (which is almost certainly false and indeed contradictory with the previous).

Third, we judge probabilities as more important when they are close to certainty. We make a large distinction between a 0% probability and a 0.0000001% probability, but almost no distinction at all between a 41% probability and a 43% probability.

That last part is what I want to focus on for today. In Kahneman’s model, this is a continuous, monotonoic function that maps 0 to 0 and 1 to 1, but systematically overestimates probabilities below but near 1/2 and systematically underestimates probabilities above but near 1/2.

It looks something like this, where red is true probability and blue is subjective probability:

cumulative_prospect
I don’t believe this is actually how humans think, for two reasons:

  1. It’s too hard. Humans are astonishingly innumerate creatures, given the enormous processing power of our brains. It’s true that we have some intuitive capacity for “solving” very complex equations, but that’s almost all within our motor system—we can “solve a differential equation” when we catch a ball, but we have no idea how we’re doing it. But probability judgments are often made consciously, especially in experiments like the Allais paradox; and the conscious brain is terrible at math. It’s actually really amazing how bad we are at math. Any model of normal human judgment should assume from the start that we will not do complicated math at any point in the process. Maybe you can hypothesize that we do so subconsciously, but you’d better have a good reason for assuming that.
  2. There is no reason to do this. Why in the world would any kind of optimization system function this way? You start with perfectly good probabilities, and then instead of using them, you subject them to some bizarre, unmotivated transformation that makes them less accurate and costs computing power? You may as well hit yourself in the head with a brick.

So, why might it look like we are doing this? Well, my proposal, admittedly still rather half-baked, is that human beings don’t assign probabilities numerically at all; we assign them categorically.

You may call this, for lack of a better term, categorical prospect theory.

My theory is that people don’t actually have in their head “there is an 11% chance of rain today” (unless they specifically heard that from a weather report this morning); they have in their head “it’s fairly unlikely that it will rain today”.

That is, we assign some small number of discrete categories of probability, and fit things into them. I’m not sure what exactly the categories are, and part of what makes my job difficult here is that they may be fuzzy-edged and vary from person to person, but roughly speaking, I think they correspond to the sort of things psychologists usually put on Likert scales in surveys: Impossible, almost impossible, very unlikely, unlikely, fairly unlikely, roughly even odds, fairly likely, likely, very likely, almost certain, certain. If I’m putting numbers on these probability categories, they go something like this: 0, 0.001, 0.01, 0.10, 0.20, 0.50, 0.8, 0.9, 0.99, 0.999, 1.

Notice that this would preserve the same basic effect as cumulative prospect theory: You care a lot more about differences in probability when they are near 0 or 1, because those are much more likely to actually shift your category. Indeed, as written, you wouldn’t care about a shift from 0.4 to 0.6 at all, despite caring a great deal about a shift from 0.001 to 0.01.

How does this solve the above problems?

  1. It’s easy. Not only don’t you compute a probability and then recompute it for no reason; you never even have to compute it precisely. Just get it within some vague error bounds and that will tell you what box it goes in. Instead of computing an approximation to a continuous function, you just slot things into a small number of discrete boxes, a dozen at the most.
  2. That explains why we would do it: It’s easy. Our brains need to conserve their capacity, and they did especially in our ancestral environment when we struggled to survive. Rather than having to iterate your approximation to arbitrary precision, you just get within 0.1 or so and call it a day. That saves time and computing power, which saves energy, which could save your life.

What new problems have I introduced?

  1. It’s very hard to know exactly where people’s categories are, if they vary between individuals or even between situations, and whether they are fuzzy-edged.
  2. If you take the model I just gave literally, even quite large probability changes will have absolutely no effect as long as they remain within a category such as “roughly even odds”.

With regard to 2, I think Kahneman may himself be able to save me, with his dual process theory concept of System 1 and System 2. What I’m really asserting is that System 1, the fast, intuitive judgment system, operates on these categories. System 2, on the other hand, the careful, rational thought system, can actually make use of proper numerical probabilities; it’s just very costly to boot up System 2 in the first place, much less ensure that it actually gets the right answer.

How might we test this? Well, I think that people are more likely to use System 1 when any of the following are true:

  1. They are under harsh time-pressure
  2. The decision isn’t very important
  3. The intuitive judgment is fast and obvious

And conversely they are likely to use System 2 when the following are true:

  1. They have plenty of time to think
  2. The decision is very important
  3. The intuitive judgment is difficult or unclear

So, it should be possible to arrange an experiment varying these parameters, such that in one treatment people almost always use System 1, and in another they almost always use System 2. And then, my prediction is that in the System 1 treatment, people will in fact not change their behavior at all when you change the probability from 15% to 25% (fairly unlikely) or 40% to 60% (roughly even odds).

To be clear, you can’t just present people with this choice between game E and game F:

Game E: You get a 60% chance of $50, and a 40% chance of nothing.

Game F: You get a 40% chance of $50, and a 60% chance of nothing.

People will obviously choose game E. If you can directly compare the numbers and one game is strictly better in every way, I think even without much effort people will be able to choose correctly.

Instead, what I’m saying is that if you make the following offers to two completely different sets of people, you will observe little difference in their choices, even though under expected utility theory you should.
Group I receives a choice between game E and game G:

Game E: You get a 60% chance of $50, and a 40% chance of nothing.

Game G: You get a 100% chance of $20.

Group II receives a choice between game F and game G:

Game F: You get a 40% chance of $50, and a 60% chance of nothing.

Game G: You get a 100% chance of $20.

Under two very plausible assumptions about marginal utility of wealth, I can fix what the rational judgment should be in each game.

The first assumption is that marginal utility of wealth is decreasing, so people are risk-averse (at least for gains, which these are). The second assumption is that most people’s lifetime income is at least two orders of magnitude higher than $50.

By the first assumption, group II should choose game G. The expected income is precisely the same, and being even ever so slightly risk-averse should make you go for the guaranteed $20.

By the second assumption, group I should choose game E. Yes, there is some risk, but because $50 should not be a huge sum to you, your risk aversion should be small and the higher expected income of $30 should sway you.

But I predict that most people will choose game G in both cases, and (within statistical error) the same proportion will choose F as chose E—thus showing that the difference between a 40% chance and a 60% chance was in fact negligible to their intuitive judgments.

However, this doesn’t actually disprove Kahneman’s theory; perhaps that part of the subjective probability function is just that flat. For that, I need to set up an experiment where I show discontinuity. I need to find the edge of a category and get people to switch categories sharply. Next week I’ll talk about how we might pull that off.

Prospect Theory: Why we buy insurance and lottery tickets

pnrj heuristics and biases, Top 10 things to know about economics , , , , , , , , , , ,

JDN 2457061 PST 14:18.

Today’s topic is called prospect theory. Prospect theory is basically what put cognitive economics on the map; it was the knock-down argument that Kahneman used to show that human beings are not completely rational in their economic decisions. It all goes back to a 1979 paper by Kahneman and Tversky that now has 34000 citations (yes, we’ve been having this argument for a rather long time now). In the 1990s it was refined into cumulative prospect theory, which is more mathematically precise but basically the same idea.

What was that argument? People buy both insurance and lottery tickets.

The “both” is very important. Buying insurance can definitely be rational—indeed, typically is. Buying lottery tickets could theoretically be rational, under very particular circumstances. But they cannot both be rational at the same time.

To see why, let’s talk some more about marginal utility of wealth. Recall that a dollar is not worth the same to everyone; to a billionaire a dollar is a rounding error, to most of us it is a bottle of Coke, but to a starving child in Ghana it could be life itself. We typically observe diminishing marginal utility of wealth—the more money you have, the less another dollar is worth to you.

If we sketch a graph of your utility versus wealth it would look something like this:

Marginal_utility_wealth

Notice how it increases as your wealth increases, but at a rapidly diminishing rate.

If you have diminishing marginal utility of wealth, you are what we call risk-averse. If you are risk-averse, you’ll (sometimes) want to buy insurance. Let’s suppose the units on that graph are tens of thousands of dollars. Suppose you currently have an income of $50,000. You are offered the chance to pay $10,000 a year to buy unemployment insurance, so that if you lose your job, instead of making $10,000 on welfare you’ll make $30,000 on unemployment. You think you have about a 20% chance of losing your job.

If you had constant marginal utility of wealth, this would not be a good deal for you. Your expected value of money would be reduced if you buy the insurance: Before you had an 80% chance of $50,000 and a 20% chance of $10,000 so your expected amount of money is $42,000. With the insurance you have an 80% chance of $40,000 and a 20% chance of $30,000 so your expected amount of money is $38,000. Why would you take such a deal? That’s like giving up $4,000 isn’t it?

Well, let’s look back at that utility graph. At $50,000 your utility is 1.80, uh… units, er… let’s say QALY. 1.80 QALY per year, meaning you live 80% better than the average human. Maybe, I guess? Doesn’t seem too far off. In any case, the units of measurement aren’t that important.

Insurance_options

By buying insurance your effective income goes down to $40,000 per year, which lowers your utility to 1.70 QALY. That’s a fairly significant hit, but it’s not unbearable. If you lose your job (20% chance), you’ll fall down to $30,000 and have a utility of 1.55 QALY. Again, noticeable, but bearable. Your overall expected utility with insurance is therefore 1.67 QALY.

But what if you don’t buy insurance? Well then you have a 20% chance of taking a big hit and falling all the way down to $10,000 where your utility is only 1.00 QALY. Your expected utility is therefore only 1.64 QALY. You’re better off going with the insurance.

And this is how insurance companies make a profit (well; the legitimate way anyway; they also like to gouge people and deny cancer patients of course); on average, they make more from each customer than they pay out, but customers are still better off because they are protected against big losses. In this case, the insurance company profits $4,000 per customer per year, customers each get 30 milliQALY per year (about the same utility as an extra $2,000 more or less), everyone is happy.

But if this is your marginal utility of wealth—and it most likely is, approximately—then you would never want to buy a lottery ticket. Let’s suppose you actually have pretty good odds; it’s a 1 in 1 million chance of $1 million for a ticket that costs $2. This means that the state is going to take in about $2 million for every $1 million they pay out to a winner.

That’s about as good as your odds for a lottery are ever going to get; usually it’s more like a 1 in 400 million chance of $150 million for $1, which is an even bigger difference than it sounds, because $150 million is nowhere near 150 times as good as $1 million. It’s a bit better from the state’s perspective though, because they get to receive $400 million for every $150 million they pay out.

For your convenience I have zoomed out the graph so that you can see 100, which is an income of $1 million (which you’ll have this year if you win; to get it next year, you’ll have to play again). You’ll notice I did not have to zoom out the vertical axis, because 20 times as much money only ends up being about 2 times as much utility. I’ve marked with lines the utility of $50,000 (1.80, as we said before) versus $1 million (3.30).

Lottery_utility

What about the utility of $49,998 which is what you’ll have if you buy the ticket and lose? At this number of decimal places you can’t see the difference, so I’ll need to go out a few more. At $50,000 you have 1.80472 QALY. At $49,998 you have 1.80470 QALY. That $2 only costs you 0.00002 QALY, 20 microQALY. Not much, really; but of course not, it’s only $2.

How much does the 1 in 1 million chance of $1 million give you? Even less than that. Remember, the utility gain for going from $50,000 to $1 million is only 1.50 QALY. So you’re adding one one-millionth of that in expected utility, which is of course 1.5 microQALY, or 0.0000015 QALY.

That $2 may not seem like it’s worth much, but that 1 in 1 million chance of $1 million is worth less than one tenth as much. Again, I’ve tried to make these figures fairly realistic; they are by no means exact (I don’t actually think $49,998 corresponds to exactly 1.804699 QALY), but the order of magnitude difference is right. You gain about ten times as much utility from spending that $2 on something you want than you do on taking the chance at $1 million.

I said before that it is theoretically possible for you to have a utility function for which the lottery would be rational. For that you’d need to have increasing marginal utility of wealth, so that you could be what we call risk-seeking. Your utility function would have to look like this:

Weird_utility

There’s no way marginal utility of wealth looks like that. This would be saying that it would hurt Bill Gates more to lose $1 than it would hurt a starving child in Ghana, which makes no sense at all. (It certainly would makes you wonder why he’s so willing to give it to them.) So frankly even if we didn’t buy insurance the fact that we buy lottery tickets would already look pretty irrational.

But in order for it to be rational to buy both lottery tickets and insurance, our utility function would have to be totally nonsensical. Maybe it could look like this or something; marginal utility decreases normally for awhile, and then suddenly starts going upward again for no apparent reason:

Weirder_utility

Clearly it does not actually look like that. Not only would this mean that Bill Gates is hurt more by losing $1 than the child in Ghana, we have this bizarre situation where the middle class are the people who have the lowest marginal utility of wealth in the world. Both the rich and the poor would need to have higher marginal utility of wealth than we do. This would mean that apparently yachts are just amazing and we have no idea. Riding a yacht is the pinnacle of human experience, a transcendence beyond our wildest imaginings; and riding a slightly bigger yacht is even more amazing and transcendent. Love and the joy of a life well-lived pale in comparison to the ecstasy of adding just one more layer of gold plate to your Ferrari collection.

Where increasing marginal utility is ridiculous, this is outright special pleading. You’re just making up bizarre utility functions that perfectly line up with whatever behavior people happen to have so that you can still call it rational. It’s like saying, “It could be perfectly rational! Maybe he enjoys banging his head against the wall!”

Kahneman and Tversky had a better idea. They realized that human beings aren’t so great at assessing probability, and furthermore tend not to think in terms of total amounts of wealth or annual income at all, but in terms of losses and gains. Through a series of clever experiments they showed that we are not so much risk-averse as we are loss-averse; we are actually willing to take more risk if it means that we will be able to avoid a loss.

In effect, we seem to be acting as if our utility function looks like this, where the zero no longer means “zero income”, it means “whatever we have right now“:

Prospect_theory

We tend to weight losses about twice as much as gains, and we tend to assume that losses also diminish in their marginal effect the same way that gains do. That is, we would only take a 50% chance to lose $1000 if it meant a 50% chance to gain $2000; but we’d take a 10% chance at losing $10,000 to save ourselves from a guaranteed loss of $1000.

This can explain why we buy insurance, provided that you frame it correctly. One of the things about prospect theory—and about human behavior in general—is that it exhibits framing effects: The answer we give depends upon the way you ask the question. That’s so totally obviously irrational it’s honestly hard to believe that we do it; but we do, and sometimes in really important situations. Doctors—doctors—will decide a moral dilemma differently based on whether you describe it as “saving 400 out of 600 patients” or “letting 200 out of 600 patients die”.

In this case, you need to frame insurance as the default option, and not buying insurance as an extra risk you are taking. Then saving money by not buying insurance is a gain, and therefore less important, while a higher risk of a bad outcome is a loss, and therefore important.

If you frame it the other way, with not buying insurance as the default option, then buying insurance is taking a loss by making insurance payments, only to get a gain if the insurance pays out. Suddenly the exact same insurance policy looks less attractive. This is a big part of why Obamacare has been effective but unpopular. It was set up as a fine—a loss—if you don’t buy insurance, rather than as a bonus—a gain—if you do buy insurance. The latter would be more expensive, but we could just make it up by taxing something else; and it might have made Obamacare more popular, because people would see the government as giving them something instead of taking something away. But the fine does a better job of framing insurance as the default option, so it motivates more people to actually buy insurance.

But even that would still not be enough to explain how it is rational to buy lottery tickets (Have I mentioned how it’s really not a good idea to buy lottery tickets?), because buying a ticket is a loss and winning the lottery is a gain. You actually have to get people to somehow frame not winning the lottery as a loss, making winning the default option despite the fact that it is absurdly unlikely. But I have definitely heard people say things like this: “Well if my numbers come up and I didn’t play that week, how would I feel then?” Pretty bad, I’ll grant you. But how much you wanna bet that never happens? (They’ll bet… the price of the ticket, apparently.)

In order for that to work, people either need to dramatically overestimate the probability of winning, or else ignore it entirely. Both of those things totally happen.

First, we overestimate the probability of rare events and underestimate the probability of common events—this is actually the part that makes it cumulative prospect theory instead of just regular prospect theory. If you make a graph of perceived probability versus actual probability, it looks like this:

cumulative_prospect

We don’t make much distinction between 40% and 60%, even though that’s actually pretty big; but we make a huge distinction between 0% and 0.00001% even though that’s actually really tiny. I think we basically have categories in our heads: “Never, almost never, rarely, sometimes, often, usually, almost always, always.” Moving from 0% to 0.00001% is going from “never” to “almost never”, but going from 40% to 60% is still in “often”. (And that for some reason reminded me of “Well, hardly ever!”)

But that’s not even the worst of it. After all that work to explain how we can make sense of people’s behavior in terms of something like a utility function (albeit a distorted one), I think there’s often a simpler explanation still: Regret aversion under total neglect of probability.

Neglect of probability is self-explanatory: You totally ignore the probability. But what’s regret aversion, exactly? Unfortunately I’ve had trouble finding any good popular sources on the topic; it’s all scholarly stuff. (Maybe I’m more cutting-edge than I thought!)

The basic idea that is that you minimize regret, where regret can be formalized as the difference in utility between the outcome you got and the best outcome you could have gotten. In effect, it doesn’t matter whether something is likely or unlikely; you only care how bad it is.

This explains insurance and lottery tickets in one fell swoop: With insurance, you have the choice of risking a big loss (big regret) which you can avoid by paying a small amount (small regret). You take the small regret, and buy insurance. With lottery tickets, you have the chance of getting a large gain (big regret if you don’t) which you gain by paying a small amount (small regret).

This can also explain why a typical American’s fears go in the order terrorists > Ebola > sharks > > cars > cheeseburgers, while the actual risk of dying goes in almost the opposite order, cheeseburgers > cars > > terrorists > sharks > Ebola. (Terrorists are scarier than sharks and Ebola and actually do kill more Americans! Yay, we got something right! Other than that it is literally reversed.)

Dying from a terrorist attack would be horrible; in addition to your own death you have all the other likely deaths and injuries, and the sheer horror and evil of the terrorist attack itself. Dying from Ebola would be almost as bad, with gruesome and agonizing symptoms. Dying of a shark attack would be still pretty awful, as you get dismembered alive. But dying in a car accident isn’t so bad; it’s usually over pretty quick and the event seems tragic but ordinary. And dying of heart disease and diabetes from your cheeseburger overdose will happen slowly over many years, you’ll barely even notice it coming and probably die rapidly from a heart attack or comfortably in your sleep. (Wasn’t that a pleasant paragraph? But there’s really no other way to make the point.)

If we try to estimate the probability at all—and I don’t think most people even bother—it isn’t by rigorous scientific research; it’s usually by availability heuristic: How many examples can you think of in which that event happened? If you can think of a lot, you assume that it happens a lot.

And that might even be reasonable, if we still lived in hunter-gatherer tribes or small farming villages and the 150 or so people you knew were the only people you ever heard about. But now that we have live TV and the Internet, news can get to us from all around the world, and the news isn’t trying to give us an accurate assessment of risk, it’s trying to get our attention by talking about the biggest, scariest, most exciting things that are happening around the world. The amount of news attention an item receives is in fact in inverse proportion to the probability of its occurrence, because things are more exciting if they are rare and unusual. Which means that if we are estimating how likely something is based on how many times we heard about it on the news, our estimates are going to be almost exactly reversed from reality. Ironically it is the very fact that we have more information that makes our estimates less accurate, because of the way that information is presented.

It would be a pretty boring news channel that spent all day saying things like this: “82 people died in car accidents today, and 1657 people had fatal heart attacks, 11.8 million had migraines, and 127 million played the lottery and lost; in world news, 214 countries did not go to war, and 6,147 children starved to death in Africa…” This would, however, be vastly more informative.

In the meantime, here are a couple of counter-heuristics I recommend to you: Don’t think about losses and gains, think about where you are and where you might be. Don’t say, “I’ll gain $1,000”; say “I’ll raise my income this year to $41,000.” Definitely do not think in terms of the percentage price of things; think in terms of absolute amounts of money. Cheap expensive things, expensive cheap things is a motto of mine; go ahead and buy the $5 toothbrush instead of the $1, because that’s only $4. But be very hesitant to buy the $22,000 car instead of the $21,000, because that’s $1,000. If you need to estimate the probability of something, actually look it up; don’t try to guess based on what it feels like the probability should be. Make this unprecedented access to information work for you instead of against you. If you want to know how many people die in car accidents each year, you can literally ask Google and it will tell you that (I tried it—it’s 1.3 million worldwide). The fatality rate of a given disease versus the risk of its vaccine, the safety rating of a particular brand of car, the number of airplane crash deaths last month, the total number of terrorist attacks, the probability of becoming a university professor, the average functional lifespan of a new television—all these things and more await you at the click of a button. Even if you think you’re pretty sure, why not look it up anyway?

Perhaps then we can make prospect theory wrong by making ourselves more rational.