Gambling survey section 2

1. Gambling, Addiction and Behavioral Economics

EXECUTIVE SUMMARY

The principal reason that gambling policy is both controversial and difficult to normatively frame is that gambling is irrational, in a specific sense. Gamblers both want to win each bet and to participate in an activity that is stimulating because there’s a serious probability of losing. Thus when the state licenses commercial gambling operations and extracts significant revenue from it, many think this amounts to exploiting the weakness of some citizens. This is of moral import particularly because non-trivial numbers of people gamble to the point of causing severe distress, and sometimes even catastrophe, to themselves and their families. Some extreme problem gamblers – so-called pathological gamblers – are popularly taken to suffer from a clinical disorder in the same sense that drug addicts and alcoholics are taken to suffer from such disorders. It troubles many people that commercial gaming institutions, and the state, may be earning profits and revenues by to some extent preying on sick people.

There is only one general way to try to make progress with respect to confused and conflicted intuitions on the proper goals and norms of policy around problem gambling. This is to conduct scientific research that attempts to understand what motivates people to gamble, and why some people find it difficult or impossible to gamble within controlled and reasonable limits.

One immediate problem in scientific study of gambling is uncertainty over basic concepts. This uncertainty makes it difficult to coordinate on instruments for recruiting and screening research subjects and interpreting research results. The psychiatric profession has operationalized the concept of ‘pathological gambling’, but not milder forms of problem gambling that are sometimes thought to be stepping stones to pathological gambling. ‘Disordered gambling’ (DG) is used to refer to problem and pathological gambling together. DG cannot be operationalized unless problem gambling is operationalized.

Yet another important issue, related to these, is whether some, many or most pathological gamblers (PGs) are genuinely addicted, in a medical sense, or whether common talk of ‘gambling addiction’ should be understood as metaphorical. This matters because it is important to know whether what has been learned about the effective treatment of substance addicts can be applied directly to treatment of some, many or most PGs.

A recently emerged scientific discipline, Behavioral Economics (BE), studies the relationships between motivation and reward. BE researchers are especially interested in addiction and addiction-like compulsive behavior because it is the clearest instance, and perhaps the most important instance for policy, of behavior in which people repeatedly act against what they say are their preferences, and then regret their actions. This is not cheap talk: people spend significant resources trying, often unsuccessfully, to restrain their own behavior with respect to drugs, gambling, and some other objects of frequent preference reversal.

BE researchers have developed rigorous and well-confirmed theories of this sort of behaviour. These theories have been tested on both humans and other animals and found to predict and explain the data well. A principal finding of BE is that the normal behavior of people and other animals when they are choosing between immediate and delayed rewards in which the later reward is larger than the sooner one is to reverse preferences twice over time: once as the smaller reward comes close, and then again after they have taken that reward. Thus the behaviour of addicts and DGs is, in this central respect, normal. On the basis of this discovery, our main question turns around. Instead of asking ‘Why do some people become addicts or DGs?’, we must ask ‘How do most people avoid becoming addicts and DGs?’

BE has supplied a general answer. People who are not addicts or DGs bundle their rewards by means of personal rules. A personal rule is imposed on a person by herself to restrict her access to smaller, more immediate rewards. The reason people don’t just ignore their personal rules under temptation is that breaking a personal rule ruins it, and does so at the time the choice is made. This is an immediate cost, not a delayed one. It implies loss of all the later, larger rewards that depended on the personal rule. Thus these rewards are ‘bundled’ together in the present, instead of being scattered separately through the future.

Based on the BE model of choice over time, we propose an operationalization of problem and disordered gambling. We cannot state this without anticipating something for which we present evidence in Chapter 3, that some, many or most PGs are genuinely addicted in a neuroscientifically specifiable and diagnosable way. Then our operational classification of DGs is as follows. A ‘problem gambler’ is anyone who (i) gambles often enough to be consciously aware of trying to develop personal rules to limit their frequency of gambling or their average amount gambled per session or both, but who (ii) regards these rules as inadequately effective because they regret their own gambling behavior on a regular basis, and who (iii) does not meet the psychiatrically accepted (DSM-IV) criteria for pathological gambling. Then a non-addicted disordered gambler (NADG) is any problem gambler or pathological gambler who is not an addictive gambler (AG). A disordered gambler (DG) is anyone who is either a NADG or an AG.

Evidence indicates that there is a significant difference between AGs and NADGs that is immediately relevant to treatment strategies and policies. AGs have distinctive personal (neurochemical) properties and are best helped by direct therapy aimed at these properties. NADGs, by contrast, are not picked out by distinctive personal properties. They are neuroscientifically normal people who are unusually vulnerable to bundling failures in environments typical of gambling venues. A principal focus of research that aims at reduction of the prevalence of NADG behaviour should be the circumstances under which casino and other standard gambling environments interfere with bundling. Once these circumstances are identified, changing them will have costs, to industry, to the state, and to non-DG consumers of gambling services. Determination of policy will require a combination of normative reflection and cost-benefit analysis. But this can only be usefully done once behavioral science learns more about the benefits that can be achieved by policies that facilitate improved bundling, and the costs of implementing these policies.

1.1. Gambling in scientific focus

Except where cultures specifically set up institutional barriers to gambling, the majority of human beings are disposed to regularly stake wagers against each other on the outcomes of natural and contrived processes.[1] No one seriously doubts that one motivation of most people, most of the time they behave in this way, is enjoyment of the periods of uncertainty that fall between placing bets and seeing how they come out. When people win bets they typically get a further shot of satisfaction, since almost everyone enjoys becoming a bit richer, especially when this results from something they themselves chose to do and they can thus congratulate themselves.

To scientists, stating the obvious, as we just did, is typically the first step in exposing what we don’t know and should take steps to find out. What proportion, if any, of gambles are mainly motivated by the lure of excitement, and what proportion are mainly efforts to get money? Are there other rewards that significant numbers of people pursue by gambling, such as feelings of control or escape from unpleasant realities? One cannot answer these questions just by reflecting on one’s own case or that of one’s acquaintances, since these samples aren’t likely to be representative. Furthermore, psychologists have recognized for many years that people’s reflective judgments about their own behavior are highly unreliable, because they more resemble story creation than objective reporting. The moment we recognize this, and make systematic efforts to collect and test samples of gambling behavior that are representative and objective, we’re embarked on scientific study of gambling.

There are two main grounds for thinking such study worthwhile. One is the fundamental engine of science, basic curiosity. Many people want to know more about the general patterns of human behavior just because they find people (in general) interesting. It isn’t obvious that nature would have designed people to like to gamble, yet the majority of them clearly do. Why? To the extent that we make progress on answering that, we automatically make progress on answering further interesting questions, because the explanation of why people enjoy gambling is bound to imply explanations of other things they like and dislike doing.

The other main motivation for scientifically studying gambling is the hope of furnishing a basis for accurate prediction of behavior. There are benefits in improving our ability to forecast when people will gamble, and how much they’ll gamble. These benefits are obvious for someone who’s in, or considering entering, the business of supplying people with gambling opportunities by bringing prospective gamblers together with things to gamble on. People who enjoy gambling themselves have an interest in predicting the behaviour of fellow gamblers. And it is a commonplace of both folk wisdom and scientific psychology that people usually need some knowledge of human behaviour in general in order to best predict what they themselves will do in different kinds of circumstances. A very inefficient way of finding out what sort of things you’ll do under different conditions is to actually put yourself under all conditions and find out directly; a more sensible approach, at least at a broad level, is to suppose that you’re typical of some reference class (i.e., women in general or South Africans in general or South African women or South African women in their forties, etc.) except where you have specific evidence to the contrary, and then find out by means of surveys what typical people in that reference class do.

Let us give an example of this. The gambling industry, as noted above, makes money by supplying people with opportunities to gamble. Partly this just involves providing geographical focal points where people who want to bet on things can find other people to bet against. But it also involves supplying games on which to gamble. What should these games be? It is just ‘obvious’ to most people that they should be slot machines and card games and so on, because that is what they’re used to. When they ‘look inside’ themselves they ‘find’ preferences for these sorts of activities. But this seems like an explanation only because we haven’t really confronted the full force of the question. Why do bookies and casinos get paid to supply people with things to gamble on in the first place? After all, every environment always has elements of uncertainty. People could bet on the weather, or on the colour of the next car that will pass, or on whether a squirrel will run up a tree or down it. Indeed, some people do enjoy betting on these sorts of things. But evidently most people like having a pre-established, structured common focus for their betting; otherwise the existence of expensive casinos, for which all gamblers pay a share of the costs, would be mysterious. And clearly people also like gambling on some kinds of uncertainties more than others, since casinos don’t concentrate on finding games that cost as little as possible to set up, as they would if people were naturally indifferent and could be trained like laboratory rats to prefer whatever was provided. Finally, almost all non-industrial cultures have converged on culturally preferred games with conventional rules for placing bets that require gathering and maintenance of some specific infrastructure (fighting cocks, cowrie shells, specially shaped pieces of bark, etc.).[2] Notice that once we seriously ask the question ‘Why are people more attracted to certain sorts of gambling scenarios than others?’ we’re approaching the question of why they gamble at all; but as we saw, we can’t make progress on this by each of us introspecting. We have to study lots of human behavior, and we have to do so by designing experiments that force nature to answer specific queries we put to it.

These considerations emphasize the value of better behavioral prediction for personal policies. Gambling also has a public policy aspect. Increasingly, many countries, provinces and municipalities get important revenue from supplying or licensing and taxing gambling opportunities such as lotteries and casinos. Thus we have an interest in being able to predict what different levels of public revenue we should expect from different possible institutional arrangements. Aiming at public revenue from gambling proceeds implies that we think it appropriate that there could be some transfer of a bit of the public finance burden, in which to some extent people who like to gamble pay higher per capita taxes than others, all else being equal. But then who are these people that our gambling policy will lead us to tax more highly? Are they richer than average, randomly distributed in wealth or poorer than average? Are they older than average or younger than average? Are they more likely to be men or women? Better educated or less educated? In a diverse society like South Africa we should want to know if they’ll be disproportionately drawn from one cultural community by comparison to another.

In asking these questions about who gambles more and less, we so far haven’t distinguished gambling, in kind, from any other sort of behavior that’s relevant to distribution and justice issues in public policy. We’d have reason to pose and to try to answer similar questions if we were setting out to design a new tax on bread. But gambling raises additional complications because it differs from bread-buying in two important respects: (i) no one regards the (normal) consumption of bread as morally worrying; and (ii) we have no reason to think that many people are importantly irrational in the way they consume bread.[3]

Many people think that gambling is morally improper. Others think that light gambling is morally harmless but that heavy gambling – that is, gambling beyond some level of frequency or with stakes that exceed some approximate threshold of someone’s resources – is morally improper. Still others think that gamblers themselves commit no moral offense, but that the state does something morally improper if it takes advantage of people’s enjoyment of gambling to get extra taxes from them. This belief might in turn have one or both of two different bases: one might think that the state shouldn’t exploit gamblers because gamblers are too likely to be economically vulnerable in other respects; or one might think that exploiting gamblers involves profiting from people’s irrationality, and that doing that is a form of preying on weakness, thereby resembling laying an extra tax burden on the physically or (in this case) mentally disabled.

Questions about whether certain kinds and patterns of gambling behavior, or certain public policies around gambling, are in fact morally bad, are philosophical questions. Questions about why people have the moral opinions they do, and about why and under what circumstances people do things they themselves think are immoral – thus opening themselves to immoral exploitation by their own lights – are anthropological and sociological questions. In this survey we will set both of these kinds of questions aside, though we don’t deny their importance. Readers who want to follow up on them in a careful way are referred to a recent discussion by Peter Collins.[4] Here we will take one of Collins’s main conclusions on moral attitudes to gambling as our starting point. This is that, regardless of what one thinks is the right answer to the philosophical question, there is a clear majority opinion on this question that prevails at the moment in most democratic jurisdictions, including South Africa. This (several-part) opinion is that

1. gambling at moderate frequency for moderate stakes (as referenced to people’s own income levels) is something that people should be allowed to do if they wish;

2. the state should tax this activity at a rate comparable to the rate at which it taxes other forms of adult-focused entertainment that doesn’t enhance health; but

3. the state should regulate and keep within normative bounds the visibility and extent of gambling, should not allow exploitation by either itself or by private parties of people who can’t control their gambling, and should carefully monitor, with a view to managing, the extent to which gambling disproportionately draws resources from economically vulnerable people, especially children whose caregivers gamble.

Again, let us be clear that we neither endorse nor criticize this consensus that Collins identifies. Some (especially religious) people are inclined to doubt its moral soundness and therefore work to change it. One of the foremost scientific authorities on gambling behavior, Howard Rachlin,[5] has argued that, in light of the psychological factors that cause people to gamble, the current consensus doesn’t lead to stable policy. That is, Rachlin thinks that commitment to norms (1) and (2) above almost inevitably leads governments to have problems maintaining coherent policies in line with norm (3), for two reasons. First, policies in keeping with norms (1) and (2) cause more people, according to Rachlin, to lose control of their gambling behavior. Let us call this ‘the pessimistic behavioral hypothesis’. Second, Rachlin thinks that governments that depend on gambling for important revenues will be unable to avoid the temptation to exploit people who have trouble controlling their gambling because there is so much revenue in doing so, and these people don’t have a strong lobby working on their behalf. Let us call this ‘the pessimistic political hypothesis’.

The work we review in this survey will not bear on the pessimistic political hypothesis. That is a subject for political economy. However, our survey certainly will bear on the pessimistic behavioral hypothesis. We cannot claim to know whether liberal gambling policies contribute to the extent or severity of gambling problems except insofar as we have some idea why most people gamble, why some people gamble more than others, and why some people try to reduce or cease their gambling and find that they cannot (or, at least, that they cannot without special help).

It is the call for explanation of this last fact that confronts us with the supposed irrationality of some gambling behavior. We need to consider this idea carefully here, for two reasons. First, as we will see, the view that some or all gambling is irrational is crucial to the public policy issues around the legal status of gambling, the state’s role in sponsoring it and profiting from its revenues, and the social obligations of the gambling industry. Second, it is because of the apparent irrationality of some gambling behaviour, in a particular sense we define shortly, that we can’t study it using only the tools we bring to bear on other consumption behaviour, and must in addition resort to a distinctive scientific approach called ‘behavioral economics’ that we describe in section 1.2.

Let us begin with the role of irrationality in the public policy issues around gambling. Remember that we have put aside the philosophical question about whether gambling is in fact morally permissible, in favour of asking what kind of policy is best given the approximate liberal-democratic consensus on this question identified by Collins. The third part of that consensus says that the state should act to soften the harm done by gambling to society in general, trading this off against our aim to specially tax more rational gamblers. But no part of the consensus as we stated it here says anything about whether or to what extent the public authority should be obligated to help people who think that their gambling is a problem to themselves.

Modern democracies are, in general, deeply ambivalent and unsure about this sort of issue. On the one hand, most people shy away from too much paternalism – the view that some eccentric preferences people have are crazy in themselves and should be interfered with over and above the implications they have for the welfare of other people. For instance, most of us might think that handling cobras is mad. Yet a few people enjoy doing it. South African law tells them not to do it in circumstances where snakes might escape their control and menace others. However, a South African man recently made headlines by setting records for time spent living in rooms full of cobras and mambas, until his unsurprising demise at the fangs of one of his reptiles. The law did not try to interfere with him; and most South Africans probably think it was right not to have. If we say that this person was irrational, what we mean is that he wanted things out of life that we find strange, but which were dangerous only to himself – and most of us seem to think that intervention against that kind of irrationality is not the business of the state.

But now suppose that the snake enthusiast had told friends and the news media that, fearing for his life, he was desperate to stop consorting with cobras, but found himself unable to follow through on his resolutions to do so. No matter how sincerely and fervently he swore off snakes, he reported, when the moment to leave them safely on the other side of some glass arose, he could not resist the impulse to get himself on the dangerous side of the partition. Many or most people of liberal-democratic views might in these circumstances think they had an obligation, or at least a legitimate reason, to help the unfortunate ‘cobraholic’ achieve his own stated goal in life by making it impossible for him to come within range of his siren’s call.

The different judgments in these two cases – the real one and the hypothetical one – stem from recognition of the fact that the cobraholic is irrational in a more profound way than was the late real man. The latter, as we saw, just had peculiar preferences. In the case of the imaginary cobraholic, it hard to say that he even has clear preferences. If he merely said he wanted nothing more to do with snakes, but never took any actions consistent with this declaration, we might think he was just being disingenuous to keep his audience on edge. But suppose he took time and trouble and spent money to keep snakes out of his life, and then spent yet more time and trouble and money getting around these very barriers he’d set up for himself. Then we might judge either that ‘he doesn’t know what he wants’ or that ‘he really wants to stay away from snakes but can’t pull this off on his own’. In either case, we might think we could, and/or should, help him.

The case of the imaginary cobraholic is fanciful; the real snake man did his thing while observers shook their heads in bemusement and then met his predictable end. But it is easy for anyone to understand the made up case because it has close parallels in the situations of the many people who take actions to try to stop drinking or taking drugs or gambling but fail in their attempts.

Economists use the ideas of rationality and irrationality in a narrower sense than other people. To economists, only the more profound kind of irrationality exhibited by the cobraholic really counts as true irrationality. This isn’t because economists don’t recognize strange preferences, like the real snake man’s, as being strange. Instead, they begin by taking note of the same distinction everyone else does – between cases of strange but clear preferences and cases where people don’t seem to even have straightforward or consistent preferences – and then, for the sake of avoiding confusion in their professional discourse, agree to call only the second kind of phenomenon ‘irrationality’. This reflects more than just an arbitrary taste in the use of words. It stems from some deep facts about the relationship between the subject matters of economics and that of the neighboring science of behavior, psychology. The next section of the survey describes this relationship.

Before we go on to that, however, let us draw attention to another aspect of the imaginary case of the cobraholic that the reader will have noted. Irrationality as economists understand it corresponds to a core part of what we get at in everyday contexts by the concept of ‘addiction’. The historical origins of this concept are closely related to the concept of ‘enslavement’ which at one time was called by the same word. Gradually, ‘addiction’ came to mean ‘enslavement to an object or to a kind of behaviour’ (to what psychologists generalize as ‘a reinforcer’), as opposed to enslavement to a person. Modern democracies have outlawed literal enslavement to persons.[6] But there is plenty of ‘addiction’ around, and modern societies are less sure about how they should handle it. There may be no actual cobraholics (or maybe the real snake man was actually a cobraholic – we’ll come back to this question later, in the final chapter of this report). But there are certainly people who try to drink less or nothing but end of drinking a lot and severely harming themselves. And there are also significant numbers of people who try to gamble less, or with smaller stakes, or not at all, but end up gambling to ruin.

We will come back, briefly in section 1.8 and then at length in Chapter 3, to the question of how helpful (or not) it is to lump people who are irrational (in the economist’s sense) about gambling with people who are irrational about alcohol and drugs. For the moment, we put this aside in order to stipulate a more precise way to talk specifically about different sorts of gambling behavior. We’ll need a finer set of distinctions than just the bare one between people who have an unusually strong but not irrational (in the economist’s sense) taste for gambling and people who would like to change their gambling behavior but have trouble doing so. Psychologists have developed some conventions for separating kinds of cases here, and we need to follow these conventions, both to avoid confusions and so that we can measure the extent of different social phenomena with consistency from one survey or experiment to the next.

One of the most crucial steps in setting up a problem for scientific study is the definition of fundamental concepts. Carelessness here can, all by itself, predetermine or overwhelmingly influence the later interpretation of results and so make whole investigations useless. Let us give an illustrative example for the case of gambling problems. Suppose on the one hand that one adopted a very liberal definition, according to which anyone who had ever regretted any particular participation in a game or any particular bet was said to have a “problem” caused by gambling. In that case, everyone who has ever gambled more than two or three times would be declared to have a gambling problem, since almost everyone who has bet that often has lost at least once, and everyone who has lost a bet at least regrets that bet in at least the mild sense that if they could have the decision to make it back, knowing the outcome, they’d decline it. In that case ‘problem gambling’ would be made almost synonymous with ‘gambling’ and whatever facts we later said we’d found about problem gambling would really just be facts about gambling in general. On the other hand, suppose we adopted a very stringent definition. For example, we might say that someone only has a clear, serious gambling problem if it leads them to bankruptcy. We know that if we use that definition, we’re very likely to “discover” that only a tiny proportion of gamblers have problems, so the costs of problem gambling aren’t likely to come out as large by comparison with the benefits derived from less regulated gambling. In both the cases of the overly liberal and the overly stringent definitions, subsequent empirical study doesn’t get allowed to do any real work; we will get out of it as conclusions just what we put into it by definition.

This potential problem caused by careless definition will be obvious enough, but the solution to it is far from obvious. What definition of problem gambling is the right one? We’re not going to find this written on a rock somewhere; so it looks as if any definition we try will be arbitrary, a pure decision we make prior to doing any investigative work, which will then drive those very findings.

Scientists partly overcome this problem by distinguishing between definitions – or, speaking more properly, analyses – and operationalizations. When you operationalize an idea you don’t think of yourself as trying to state the truth about its deep nature. Rather, you just take yourself to be stipulating what you’ll mean by it for the duration of some enterprise in which you’ll use it. The only test of a good operationalization is usefulness: does it divide the phenomena we’re studying into two or more piles that are both interesting? Notice that if we used either the overly liberal or overly stringent definitions of problem gambling as operationalizations they’d fail this test: the overly liberal operationalization would put almost all gambling in one pile and in the other pile leave only gambling by people who have hardly ever gambled, and the second pile isn’t interesting; the overly stringent operationalization would leave in one pile only gambling by people who went bankrupt as a result. That pile isn’t completely uninteresting, but it’s not nearly as interesting as the pile we really want, which is the one containing all and only the gambling by people who, if asked “Is your life made significantly worse by your unsuccessful struggles to stop or cut down on gambling?” would answer “Yes”.

One way to increase the probability of useful operationalizations is to deploy more than one of them, so as to divide our phenomena into more than just two piles. In the case of patterns in gambling behavior, this is what psychologists have done. Recent common practice among researchers is to attempt to operationalize three terms:

Pathological gambling: A chronic inability to refrain from gambling to an extent that causes serious disruption to core life aspects such as career, health and family. The diagnostic criterion established by the Diagnostic and Statistical Manual of Mental Disorders (DSM-IV) is that a person is a pathological gambler if they agree with five or more of the following statements:

1. You have often gambled longer than you had planned.

2. You have often gambled until your last dollar was gone.

3. Thoughts of gambling have caused you to lose sleep.

4. You have used your income or savings to gamble while letting bills go unpaid.

5. You have made repeated, unsuccessful attempts to stop gambling.

6. You have broken the law or considered breaking the law to finance your gambling.

7. You have borrowed money to finance your gambling.

8. You have felt depressed or suicidal because of your gambling losses.

9. You have been remorseful after gambling.

10. You have gambled to get money to meet your financial obligations.

Problem gambling: Researchers want this idea to capture people whose gambling behavior is at least a nuisance to them, and is so along the same dimensions as are used to operationalize pathological gambling. We want these operationalizations to line up in this way so that we can use the concept of problem gambling to clearly pick out people who are at risk of becoming pathological gamblers. US President Clinton’s Committee on the Social and Economic Impact of Pathological Gambling[7] operationalized problem gambling as “gambling behavior that results in any harmful effects to the gambler, his or her family, significant others, friends, co-workers etc.” In this they simply reflected what they found in the research and treatment literature. But the operationalization is too liberal to be truly useful; losing R5 on a football bet or being late for lunch with a co-worker because of a queue at the betting window constitute ‘harmful effects’, however trivial. It would be natural, and maximally helpful, if we could operationalize problem gambling as ‘pathological gambling lite’ (which might indicate incipient pathological gambling) by describing the problem gambler as someone who agrees with some number of the statements in the operationalization of pathological gambling less than five. However, this confronts us with the problem that the statements don’t all seem to be equally diagnostic. Anyone who agrees with (5), (6) or (10) probably has or had a relatively serious problem associated with gambling, and likewise for (8) if ‘depressed’ is interpreted clinically. But this cannot be said of the other statements unless their interpretations are specially restricted. It cannot be said of (1) under any plausible restriction.

Disordered gambling: This denotes the inclusive disjunction of the two ideas above, i.e., gambling that is either pathological gambling or problem gambling.

We said that researchers “attempt to” operationalize these three terms, and it will now be evident why we used this cautious formulation. Pathological gambling is indeed operationalized by DSM-IV. But problem gambling is not; nor can it readily be operationalized by reference to the operationalization of pathological gambling. And then the idea of disordered gambling inherits all the problems attaching to the lack of clarity around problem gambling.

The lack of a settled operationalization for any concept other than pathological gambling raises a problem for the writing of a survey like the present one. The point of a survey is to try to consolidate and compare many studies. This is precisely the activity in which one most depends on confidence that different groups of researchers mean exactly the same thing by every term they use to describe the phenomena they take themselves to be measuring.

We could get around this problem, in principle, by strictly confining our discussion to pathological gambling, remaining absolutely silent on anything else. This, however, would be deeply unsatisfactory procedure from the point of view of readers concerned with policy. Suppose, as seems likely, that there are substantially more problem gamblers, on some reasonable but necessarily (at this point) vague understanding of that concept, than there are pathological gamblers. In that case, the lion’s share of the social costs associated with a given gambling policy might be accounted for by the problem gamblers, and no major policy recommendations could be based on studies that ignored them. That would be a situation we’d just have to live with for now if it could honestly be said that the study of problem gambling hasn’t yet progressed far enough to license any policy recommendations. But we don’t think that can honestly be said.

Fortunately, we believe that the particular disciplinary perspective from which this survey is written, that of behavioral economics (henceforth ‘BE’, to be described in the next section) affords a way around the difficulty. We do not claim it is a complete or all-purpose solution, but we think it is good enough to let us get on with business. In section 1.6 below we will introduce the BE-inspired concept of a ‘personal rule’. We’ll operationalize ‘problem gambling’ in terms of this rule. We will use ‘pathological gambling’ as per the DSM-IV operationalization. Since we’ll thus have operationalizations of both ‘problem gambling’ and ‘pathological gambling’ we’ll be able to use ‘disordered gambling’ in the way indicated above.

We explained that an operationalization is not yet an analysis. This is important: whereas an operationalization is a tool for doing scientific study and therefore must precede it, analysis is one of the things one wants to get from scientific study. That is, scientific investigation aims at telling us, among other things, what problem gambling and pathological gambling ‘really are’. This is most usefully put in terms of policy-relevance. Policy-makers and industry participants need to know whether there’s a qualitative ‘jump’ between problem gamblers, as just typical consumers of gambling services who sometimes have bad days at the casino – as people have bad days in all their activities – and pathological gamblers as people for whom special policies and practices need to be adopted. (Most carefully regulated and administered casinos do, at present, implement an assumption like this as a prudential matter and aim to identify and then exclude pathological gamblers. They don’t, obviously, exclude or have special policies for those who merely occasionally gamble a bit more than they intended to.) If there is not such a qualitative jump and all people who gamble are on a smooth continuum without ‘bright lines’, then policy problems are going to be relatively more challenging. Louise Sharpe points out that we get divergent interpretations of some research data depending on which of these hypotheses we consider most probable.[8] Addressing this question about continuity is one of the basic underlying aims of our survey.

The perspective we’ll defend in this survey with respect to problem, but not pathological, gamblers is that there are probably no systematic personal differences between them and people who like to gamble from time to time but haven’t experienced problems. (People who don’t like to or for some other reason never gamble probably do have distinctive properties, but obviously don’t matter to the issues with which we are concerned.) We’ll maintain, that is, that problem gambling isn’t a property of a kind of person, but of a typical person in a kind of situation. Where pathological gamblers are concerned, we’ll maintain, matters are different. Surveys from a number of countries and regions of countries tell us that between 1% and 3% of the population of any country or province –sized jurisdiction will tend to be incapable of gambling nearly as moderately as they themselves wish to unless they’re assisted with clinical intervention. In Chapter 3 we’ll present reasons for thinking that (i) a substantial part of this population, perhaps the majority, are addictive gamblers, in that they have brains that physically malfunction in the same way that drug addicts’ brains do; and that (ii) for identified biological reasons (and not merely as a matter of statistical regularity) people who satisfy the DSM-IV operationalization of pathological gambling are at direct risk of becoming addictive gamblers. We’ll show why we think it will very soon be possible to reliably diagnose people in the addictive group by (non-invasively) examining their brains (or perhaps even their genes), and why there are also grounds for optimism that the day isn’t far off when they’ll be able to control their problem with medication. We’ll suggest that this will make policy decisions relatively straightforward where this group is concerned.

There is a good deal of careful ground to cover before we’ll be able to get to this. In advance of arriving at a basis for categorization that is supported by a marriage of theory, systematic observation and experiment, behavioral researchers must gather subjects for study and sort them according to some operationalization or other. For this they use what are called ‘screens’, questionnaires that are administered to people in order to classify them for data-gathering and experimental purposes. Here we encounter another problem related to definitions of phenomena. With respect to conditions that cause serious harm, such as disordered gambling, the clinical imperative – the need to help suffering people – has higher social priority than scientific research.[9] In screening people to determine who are pathological or problem gamblers for clinical purposes, we want to err on the side of avoiding ‘false negatives’. That is, we’d rather diagnose risk in some people who aren’t really in trouble than miss people who need help. Clinical screens build in this bias, and are criticized if they don’t. But then we face a dilemma in choosing a screen for research purposes. If we use the clinical screen, we will find in most populations we survey a higher prevalence of the target condition than is really there. If we instead develop a research screen that differs from clinical screens, then we’ll get mis-matches between clinical and research samples. This is problematic for two reasons. First, subject recruitment for studies of conditions that affect only small proportions of the population, such as pathological gambling, is difficult and expensive. Practical purposes require that we usually rely on samples already assembled (often by self-selection) for treatment purposes (so-called ‘convenience samples’). This is especially true, for obvious reasons, when an aspect of a research project is evaluation of a therapeutic approach. Second, there are serious ethical problems in sorting clinical populations into those we scientifically classify as ‘really’ manifesting the condition for which they are being treated and those we admit we have merely diagnosed for cautionary reasons. Since scientific research on subjects cannot be concealed from those subjects, doing this would send signals to the subjects as patients, and thereby undermine the point of having designed our clinical screens to minimize false negatives in the first place.

These objections carry the day in practice. In the study of disordered gambling, research subjects are generally recruited and pooled using clinical screens. This has several implications:

We should be cautious about estimating prevalence. One can do so only by application of a ‘double filter’. That is, if one has run statistical tests to determine the extent to which a screen gathers false positives, one can then use this to systematically down-weight prevalence estimates in a population obtained using that screen.[10] Notice that this requires that we have already made some progress in the direction of analysis; if you have nothing beyond a bare operationalization of your phenomenon with which to work, then you have no distinction on which you can base the statistical testing. To a considerable extent, this is where we still stand with respect to research on disordered gambling. This is why, for example, we had to report prevalence for pathological gambling as ‘between 1% and 3%’ - not a very exact figure, since the upper bound is three times the size of the lower bound. The problem isn’t lack of data; it’s ineliminable error when all one has to work with are operationalizations instead of analyses.

When considering any research report, especially of research into the efficacy of a therapy or policy, one needs to know where the screen used to recruit and pool subjects stands comparatively on the question of the average severity of pathology found in these subjects. Consider two hypothetical screens A and B. Suppose that screen A consistently produces samples in which the median subject agrees with 8 statements in the DSM-IV operationalization of pathological gambling, while the median subject recruited by screen B agrees with 6 of them. If one knows that, then one also has reason to suspect that research using screen B is more likely to find that a given policy or therapy is effective than is research on that same policy or therapy that recruits and pools subjects by means of screen A – because the A-screened samples will contain higher proportions of ‘hard cases’. (One can’t be quite sure of this, however, because of the fact that, as noted above, not all the DSM-IV criteria seem to be of equal diagnostic weight. Suppose, for example, that more of the B-screened subjects agree with statements (5) and (6).[11])

One of the ‘big’ questions about disordered gambling we identified above as central at the moment is whether there is a qualitative ‘jump’ – in scientific language, a discontinuity – between sub-groups of disordered gamblers. As we have also explained, if there is such a discontinuity then it is highly unlikely that our current operationalizations exactly line up with it; bringing our operationalizations closer into line with it, so that they approach analyses, is precisely one of our scientific goals. This means that we can’t fully trust results derived from work using just one screen, because each screen will have its own bias with respect to the population proportions it selects on either side of the discontinuity – but we can’t know in advance where these biases are, at least until we’ve clearly identified the basis of the discontinuity. To complicate matters further, one team of researchers has given reasons for thinking that there is more than one kind of pathological gambler, thus suggesting the possibility of two scales with discontinuities at different measurement points and of different magnitudes.[12] We should therefore ideally run every study we think is important on several groups of subjects recruited using different screens.

This set of circumstances confronts us with a trade-off. On one hand, it is very helpful for cross-comparability of studies that one particular screen has so far dominated disordered gambling research. On the other hand, this also limits our ability to bootstrap our way around the inadequacies of our initial operationalizations. In an ideal world we’d have a matrix: we would use all available screens on sub-sample populations in all studies, thus having cross-comparability and controlling for screen biases.

However, we don’t live in the ideal world. Here in the actual one, the dominant instrument mentioned above is the South Oaks Gambling Screen (SOGS) introduced in 1987.[13] The SOGS is closely based on the DSM-III-R[14] criteria for pathological gambling. (It has not been updated to reflect changes introduced with DSM-IV.) As of a 2000 survey, these diagnostic criteria had been used in over 90% of published research on disordered gambling.[15] The SOGS can be self-administered by prospective subjects, which is convenient and cost-efficient. Petry[16] recommends that for research purposes SOGS be used as an initial screen, and that a more discriminating instrument then be employed as a further filter. She reviews four such existing instruments,[17] among which the comparatively new Diagnostic Interview for Gambling Severity (DIGS)[18] seems especially promising. Another recently developed more intensive instrument is the Canadian Problem Gambling Index (CGI),[19] which facilitates distinguishing between pathological and problem gamblers. Its psychometric properties as measured to date are encouraging.[20]

We want to make it clear, however, that such double-filtering hasn’t thus far been the norm in disordered gambling research, and that establishing it as the norm would add significantly to the average cost of studies. Instruments like the DIGS and the CGI are time-consuming to administer. Their use thus multiplies research staff time, laboratory usage time and the amount of subjects’ time that must be requested. It thus also multiplies all of the main items in the cost budget of a typical behavioral study.

In many cases even double-filtering of the sort just discussed is inadequate. A number of studies have aimed at identifying the extent to which pathological gamblers tend to have additional psychological / behavioral problems (‘comorbidities’), such as substance addiction,[21] depression,[22] schizophrenia,[23] and anti-social personality disorder.[24] Determination of cormorbidity rates that are stable across populations of pathological gamblers, if there are any such rates, would be helpful for a number of reasons. First, it would assist in predicting the likelihood that a given person is at risk of becoming a pathological gambler, since establishing the comorbidity factors in question would carry at least some information about this. Second, it would be highly relevant to design of treatments and interventions, possibly helping to explain patterns of success and failure. (For example, interventions that work for pathological gamblers who lack certain specific comorbidities might fail for others who do.[25]) Third, efforts to explain such stable cormorbidity patterns as we find might lead us to the explanation of pathological or addictive gambling itself, to the extent that pathological or addictive gambling and cormorbidities are sometimes or often consequences of a common causal factor. Comorbidity data are what mainly lead Blaszczynski and Nower[26] to the view that there are three different kinds of pathological gamblers. They report that across studies with larger numbers of subjects, stable proportions of pathological gamblers show cormorbidity with, respectively (i) nothing, (ii) depression and other mood disorders, and (iii) anti-social personality disorder. Unsurprisingly perhaps, subjects in these groups respond differentially to therapy. They then argue, more controversially, that the different comorbid factors causally contribute to pathological gambling in different ways (hence explaining the varying susceptibility to different interventions).

Obviously, the more we discover about the significance of cormorbidity for predicting, explaining and treating pathological and addictive gambling, the more important it will be to screen research subjects for the identified correlated conditions. This does not imply anything especially onerous for research methodology where substance abuse is the possible factor in question, since it can be effectively screened for by comparatively basic instruments. However, reliable screening for psychiatric disorders requires use of sophisticated instruments and qualified expert administration and analysis. This would have major implications for the cost and complexity of research.

Note that an approach that focuses on searching for comorbidity factors is almost inevitable to the extent that pathological gambling is conceptualized as a kind of disease – that is, as a specific breakdown in the functional constitution of the person. Thus the reason Blaszczynski and Nower conclude from three comorbidity patterns that there are three different kinds of pathological gambling is that they implicitly identify any case of pathological gambling with its molecular etiological pathway.[27] This assumption sets researchers off in search of the site of damage. They often suppose it to be unlikely that something could go wrong with a person’s mind or brain that would produce only pathological gambling, in which case such searches are bound to be, at least in part, searches for comorbidity factors. This has tended to be the perspective of much recent psychiatry. Its promised implications for policy and intervention are straightforward. If we could identify the site of damage in the pathological gambler, then we could either seek to repair the damage in question or, if that is not possible, use diagnosis of the damage as a basis for knowing whose access to legal gambling opportunities should be restricted or blocked. And perhaps we could set up mechanisms to identify those people who are at risk of becoming pathological gamblers and effectively intervene to stop onset. In Chapter 3 we’ll present evidence that this approach looks like it going to bear fruit where addictive gamblers are concerned.

Where policy is concerned, however, what if there are pathological gamblers who aren’t addicted? Or suppose we found a reliable diagnostic but not a cure for addictive gamblers? We could probably prevent many of them from going to casinos or racetracks. Preventing them from buying lottery tickets or playing Internet poker would be more difficult. But would it even be appropriate to try to prevent such people from ‘playing’ the stock market? If so, what about real estate markets? Should pathological or addictive gamblers be prevented from re-selling houses or other assets?[28]

This question helps to remind us that gambling per se is a normal, not a deviant, kind of human behavior. By this we don’t just mean that it is widespread; we mean that a disposition to gamble is adaptive and a disposition to never gamble is severely maladaptive. Few people aspire to have the kind of personality that tries to avoid all risks. More to the point, few people aspire to not enjoy taking some risks. Aversion to risk, or mere inability to appreciate the pleasure of moderate risks, amounts to aversion to novelty and absence of both curiosity and ambition. This suggests that, except where addictive gamblers are concerned, it may be mistaken to take for granted that our main question must be: What, specifically, is wrong with pathological gamblers? Is it possible that gambling problems arise, at least in the first place, when people fail to learn to effectively manage some natural dispositions shared by almost everyone? There is a discipline, closely entangled and increasingly integrated with traditional psychology but importantly distinct from it, which begins from this alternative opening perspective. This discipline is called ‘behavioral economics’. In our opinion it has for some time now been the most fecund source of insights and productive research procedures for studying the broad class of behavioral dispositions popularly associated as ‘compulsions’ and ‘bad habits’ specifically including (at least) non-addictive disordered gambling. Where addictive gambling is concerned, it has been thought by some of its founders to be in tension with neuroscience, a matter to which we will return – in application to gambling – in section 1.7. Our view is that this tension, however natural, must be overcome, since both BE and neuroscience offer crucial insights into the phenomena of interest to us here, and many others. But before we get to all that, we must digress from our direct focus on gambling to explain to the reader what BE is in the first place.

1.2. Behavioral economics

One of the respects in which pathological gambling (in general) has often been grouped with stereotypical substance dependencies such as alcoholism is by reference to symptoms, including sensed loss of control and preoccupation in consciousness. But they have also been associated by appeal to the way in which these behavioral patterns both supposedly manifest impulsiveness. The report of President Clinton’s Committee on the Social and Economic Impact of Pathological Gambling says that “there is considerable consensus that gambling involves impulsiveness.”[29] However, the report is ambiguous with respect to the interpretation of this idea. Discussion of this ambiguity is a useful way of introducing the distinction between traditional psychology as a conceptual framework and the framework of behavioral economics.

According to DSM-IV, an impulse control disorder is manifest in a person to the extent that that person shows recurrent and significant “failure to resist an impulse, drive or temptation to perform an act that is harmful to the person or to others.”[30] On this interpretation impulsiveness is an ‘inner-referenced’ pathology, a tendency of some sort of behavioral ‘organ’ to misfire. Again, this is the perspective that the great psychologist Richard Herrnstein dubbed ‘molecular’. Presupposed here by way of contrast is a notion of ‘normally non-impulsive molecular behavioral causation’ and it is by implicit reference to this that we are to understand the idea of the subject’s doing harm to herself. Traditional psychology often asks us to contrast ‘healthy’ with ‘unhealthy’ behaviour in this way. The definition just given more or less identifies impulsiveness with unhealthy behaviour.[31]

This is in tension with the fact noted at the end of the preceding section that almost everyone gambles, and that a person who absolutely refused to gamble would not likely be regarded as ‘healthy’. This implies that ‘health’ must be compatible with harm caused by gambles, since the idea of a gamble that is guaranteed not to produce harmful consequences makes little sense. Reflecting the different usages found in the literature,[32] the Committee’s report also understands impulsiveness in a different sense: as a tendency to choose smaller earlier rewards over larger later ones. The alcoholic could have sobriety, accomplishment and social respect today and tomorrow, with all the possible streams of subsequent reward these can produce; but she chooses instead the vivid but transient pleasure of a high now, from which she must subtract the costs of hangover and social disapprobrium tomorrow. The disordered gambler could have money to pay his bills, and so secure his welfare for the month, but instead chooses to experience the satisfaction of a brief thrill, from which he must subtract the value of his expected loss.

Describing these scenarios in terms of choices immediately introduces into the picture the possible relevance of economic analysis. Economics is, in its broadest sense, the study of the way in which people (and other organisms and groups) try to promote their welfare in the face of scarcity. ‘Scarcity’ has two complementary connotations. First, an economic agent’s wants exceed what the environment furnishes, especially given other agents with wants of their own who share that environment. Second, acting so as to capture one stream of benefits typically rules out capturing another possible stream. The value of what is foregone as a result of choice is referred to as the ‘opportunity cost’ of the choice in question. Thus conceiving of impulsiveness in the way suggested immediately above amounts to seeing the impulsive person as being prepared to forego larger, later rewards as the opportunity cost of smaller, earlier ones. As soon as we frame things this way, we can introduce all of the technical apparatus of economic analysis. If indulging in a drink or taking a bet is a trade-off between benefits and costs, then we can ask questions about how much change would have to be made to the goods on either side of the ledger to produce the opposite decision.

A very large quantity of empirical evidence shows that the behaviour of people popularly thought of as ‘addicts’ is indeed systematically sensitive to relative opportunity costs.[33] In section 1.8, we will return to and explain the significance of this for the way in which we conceptualize the family of disorders in which disordered gambling behaviour has usually been included by psychologists, by the clinical community, and by policy-makers. We will then ask what this perspective suggests to be distinctive about disordered gambling as a member of this family.

First, however, we must say more to introduce and explain the nature of behavioral economics itself. A reader new to the literature in this area may find herself a bit confused at first because the term is used in two different ways. Sometimes the phrase ‘behavioral economics’ (BE) is used to denote the whole sub-discipline of economics that empirically, often experimentally, studies patterns in people’s responses to shifts in incentive structures and to changes in the schedules and prices of rewards and costs that they face.[34] This can be contrasted with more traditional economics as follows. Economic theories and models in the traditional vein are typically interested in what an ideally rational agent would do in a given incentive situation, given a particular profile of desires (a ‘utility function’) and a budget constraint. In addition, traditional economics typically searches for generalizations that hold good in any set of institutional conditions – in democracies or dictatorships or states of anarchy; in cases where government successfully stabilizes the value of money and in cases where it doesn’t, in cases, for that matter, where there is no money at all but people nevertheless exchange. (We could add here: in cases where there is legal gambling and in cases where there isn’t.) Against this background, BE is often used to refer to any approach that aims at theories and models of what actual people do in particular institutional settings. This question needs separate study because actual people are not ideally rational agents, and because real institutions cannot be maintained for free; they impose so-called ‘transaction costs’, and they will survive only to the extent that some agents deem themselves to be best off if they pay these costs.

When people encounter this contrast, they often respond by thinking that behavioral economics should supplant traditional economics. Shouldn’t we always care more about what actual people in concrete conditions do than about what ideal – but non-existent – agents in completely abstract situations do? (After all, actual people are never in completely abstract situations.) But this way of putting things is misleading. The concepts we use to understand actual people (or actual firms, or actual governments, or actual agents of other kinds) are given their meaning by their definitions in the general theory of ideal rational agency. Without this theory, we wouldn’t be able to state clear economic hypotheses about actual people in actual circumstances in the first place. In the present survey, however, all of this will be kept in the background; the reader will not encounter any instances of traditional, purely abstract, economic reasoning in what follows.

We have now identified one of the two things that ‘BE’ sometimes means. To avoid confusion, we will henceforth refer to this as ‘experimental / institutional economics’ (EIE), reserving the term ‘BE’ for something more specific, and more immediately relevant to disordered gambling and to the traditional ‘core addictions’. Whereas EIE involves no adjustments in standard economic theory, but merely restricts it for application to specific scenarios, BE involves a substantive theoretical innovation. The innovation in question is due mainly to the late Richard Herrnstein.

In standard economic theory, agents are modeled as maximizing an abstract function referred to as ‘utility’. This is a highly technical and widely misunderstood idea, which cannot be fully grasped or explained independently of the mathematics economists use to write their theories and models.[35] The crucial empirical implication of the idea that will be relevant here is the following. A utility maximizer will choose consistently over time, in the sense that she will vary her choice patterns between a given stream of benefits B and all possible opportunity costs of B (i.e., other benefit streams) only on the basis of changes in two conditions: (i) her stock of B in proportion to other goods, or (ii) the externally determined price of B relative to prices of other goods. Utility maximization restricts the directions of these relationships: the maximizing agent will accept higher opportunity costs for B the less of it she has, and she’ll choose a greater quantity of B the lower its price. The reader who isn’t used to economic reasoning is urged not to puzzle too long over these formulations, because all that will matter here is one of their consequences. This is that if an agent knows what consumption opportunities her environment is going to provide, she’ll allocate her resources over time in such a way that she’ll get the highest possible welfare overall. This will typically mean that she’ll often be giving up a possible temporary welfare increase at a given time for the sake of this general maximization. If the agent is uncertain about future consumption opportunities, this complicates matters technically, but as long as she calculates her probabilities rationally, the main point remains that she’ll choose in such a way that her overall welfare will be maximized so long as she doesn’t suffer from improbable bad luck. (The following contrast may help here. An uncertain agent who allocates resources as wisely as possible and then encounters bad luck will be disappointed but has nothing to regret, unlike the agent who contributes to her own misfortune by suboptimal allocations or failure to use information that was available to her.)

In this framework, addiction must be interpreted just as a sign of preferences that happen to make life hard. If we see a person waste away from alcoholism, we must simply suppose that this person considers drunkenness to be worth very high opportunity costs. Like any other utility maximizer, she gets as much welfare overall as is possible for her. That she becomes less and less happy over time is because alcohol consumption causes tolerance to increase; the opportunity cost of the same high rises steadily as the history of drinking plays out. So far as the alcoholic’s harm to herself is concerned, the only policy response by others that isn’t cruel and oppressive is to try to avoid unnecessary increases in the price of drinks, and helping her minimize her opportunity costs by assisting her to avoid situations that are potentially dangerous or embarrassing for drunk people. It would make no moral sense in this framework for others to help the agent avoid alcohol since (once potential harm to second and third parties is minimized) this would simply amount to lowering her welfare. All these implications of the traditional economic model of utility maximization to addiction were originally worked out in technical form by the Nobel Prize winning economist Gary Becker and his collaborators.[36]

No one can prove that people don’t maximize so-called ‘expected’ utility in this way. Expected-utility maximization is a mathematical modeling technology, not a psychological hypothesis, and we should evaluate it like any other tool: can we do the job we want it for – explaining and predicting behaviour – with no more sweat and strain than we’d face if we used an alternative and actually available tool? So-called addiction is one of the leading behavioral patterns – perhaps the leading behavioral pattern – for which the answer to this question is ‘no’. The problem is that most addicts don’t just consume their substance of abuse until they’ve run out of all resources, as the straightforward application of standard economic theory with expected-utility maximization predicts that they should. Instead, most of them spend resources trying to stop consuming their substance of abuse. It isn’t impossible for economists like Becker to explain this. For example, addicts might want to periodically go dry so they can re-calibrate their tolerance levels and then enjoy their highs less expensively for awhile. The real problem for expected-utility maximizing accounts of addiction is that most addicts successfully quit eventually. Becker-style theories must infer from this that addicts did not initially know that addiction was going to be bad, and only learned this through experiencing lower welfare levels than they expected. Again, this is possible. But now it’s relapse that becomes puzzling. We have to suppose that each time an alcoholic stops drinking but then relapses she’s aiming to get drunk more cheaply for awhile, but then, when she at last stops once and for all, this was because she finally realized that she was doing herself more harm than good. If she never does stop once and for all we have to suppose that she was not doing herself more harm than good, in her own terms.

To summarize all this: on the traditional economic accounts, we can either tell the story in a way that makes reasonable sense of addicts’ tendencies to relapse but not of the fact that most eventually recover; or vice versa. What we cannot do is explain both of these phenomena in a way that doesn’t involve completely ad hoc psychology for which there is absolutely no evidence.

Almost all people with experience of addicts and addiction find this to be too much sweat and strain to make the traditional tool of expected-utility maximization worth using, especially because Herrnstein provided us with an alternative one.[37] He spent his career showing that we get accurate predictions of most human and animal choice behavior if we instead model them as maximizing local rather than overall expected utility. He called the pattern of behavior we observe on this model melioration. The family of models that technically represent melioration share commitment to what Herrnstein dubbed the ‘matching law’. This says that if a person or animal can distribute their resources to a range of activities over what they account as an interval of time, they’ll do so in exact proportion to the magnitudes of rewards they experience for each activity in that time interval. BE, in the specific sense for which we will use that label here, is the intersection of economics and psychology that models behavior by applications of the matching law.

Herrnstein did not merely show that this way of modeling addicts’ behavior captures the data; he demonstrated that it applies to people and other animals in general. People popularly thought of as addicts are merely ideally vivid cases for illustration. All agents who match and meliorate will tend to behave, from the traditional economist’s point of view, short-sightedly, failing to take full account of their long-run welfare. We must say ‘tend to’ rather than the more straightforward ‘will’ for two reasons. First, the matching law and expected-utility maximization will predict the same behaviour whenever we focus on an agent choosing between two streams of benefits A and B in which A is more rewarding than B over all intervals. For example, an alcoholic will display consistent choice as between good whiskey and bad whiskey. Second, many human institutions are set up so as to incentivize people who meliorate to act as if they were maximizers; financial institutions are the outstanding examples here, finding devices like interest payments and derivatives to ensure that all failures to respect the logic of maximization can be efficiently exploited for someone else’s profit. (This simultaneously ensures that resources don’t go to waste and that investors have incentives to adjust their accounting of time intervals until they’re acting like maximizers.) But for the phenomena of concern to us here – first, addiction-like behaviour patterns in general, and then disordered gambling specifically – BE makes distinctive predictions and offers distinctive explanations. As we will see, these predictions and explanations have been very strongly supported by empirical behavioral research.

Most importantly, BE explains behavioral patterns in a so-called molar, rather than a molecular, way. It doesn’t aim to trace out causal sequences whereby some particular event in a person’s life, or in their brain, or in their consciousness, leads to this particular drink or to that particular refusal of a drink, to this specific ante at the blackjack table or that specific decision to go to a movie tonight instead of to the casino. What BE explanations aim to account for are patterns of behavior, the relative opportunity costs that people pay over a given stretch of time for different rewards. Pathological gamblers are, in the first instance, people who pay what seem to the rest of us to be – and what they typically eventually agree themselves to be – extravagant opportunity costs for whatever it is they find rewarding about gambling. They don’t offer this payment on every single possible occasion, but that isn’t what BE aims to explain. What is of interest in BE is that they offer the payment significantly more often than problem gamblers, who in turn offer it significantly more often than non-disordered gamblers. Note that on this perspective it makes no sense to think of the returns to gambling – either the excitement or the winnings, whichever turn out to matter more motivationally – as bad things for the gambler. They are, after all, what the gambler likes – what she gambles for. Indeed, almost everyone gambles for these rewards to some extent, regardless of whether they ever play structured games of chance and pay third parties to set them up. What is bad, for the gamblers, about disordered gambling is that they pay more for these rewards, over time, than they themselves want to.

BE is a general framework for thinking about behaviour. As will be discussed in the next few sections, melioration and matching are compatible with several specific models of the phenomena traditionally grouped together as ‘addictions’, including pathological gambling. It also implies a particular set of experimental methods for trying to choose among these different models. We describe these experimental approaches, and what has been learned about disordered gambling from their application, in Chapter 2.

1.3. Behavioral economics of substance use and addiction

The scientific community has not reached a consensus regarding whether or not disordered gambling and substance abuse are fundamentally the same type or are fundamentally different types of behavioral disorders. (We will give reasons for thinking that such a consensus is now becoming more likely in Chapter 3.) To the extent they are similar, then knowledge gained regarding substance abuse would also apply to gambling. To date, BE has been much more extensively applied to substance abuse than to gambling, so we will summarize some of the BE work on substance abuse as an illustration of what can be gained from the BE perspective. Because BE is a choice framework, the value of substance use, and therefore the extent to which it is chosen, is viewed as being determined by the benefits/costs of substance consumption in relation to the benefits/costs of other available activities. This research on substance abuse has included approaches based on EIE as well as on BE innovations derived from the matching law. We will summarize the former first, which has primarily consisted of investigations of the effects on substance consumption of (1) direct constraints on access to the substance and (2) constraints on access to other valuable activities.

Effects of constraints on access to the substance. The vast majority of this research has focused on the prices of substances that people abuse, and has employed a research strategy known as “demand curve analysis.” Demand is the amount of a commodity that is purchased, and a demand curve displays how purchases of a commodity change as its price changes. Figure 1 shows a hypothetical demand curve. In Figure 1, the price increases from left to right along the x-axis, and demand increases from bottom to top along the y-axis. The law of demand states that there will be an inverse relation between price and demand, such that higher prices lead to lower demand (and vice versa). As is typical of EIE, precise quantitative concepts and methods have been developed to study these relations. One such concept is elasticity of demand, which is a quantitative description of the degree of changes in demand in the face of price changes. Elasticity in this context is a metaphor for how much “curve” or “bend” exists in the demand curve, with inelastic curves having little or no bend and elastic curves having much bend. In Figure 1, the dotted line labeled unit elasticity goes straight from the upper left to the lower right of the graph. This indicates that demand is reduced in exact proportion to any price increase. But most demand curves are more like the curved line in Figure 1, with demand changing very little (inelastic demand) at the low end of the price range but changing very much (elastic demand) at the high end of the price range.

Prior to the work on demand for substances of abuse, it was not known if such commodities obeyed the demand law. That is, before this research, one possibility for what was meant in regarding these substances as “addictive” was that the demand law didn’t apply to them, and so consumption would continue at high rates regardless of price increases. That turned out not to be the case, and this research on substance use has demonstrated the value of such demand curve analyses. Bickel and colleagues[38] re-analyzed data from numerous drug consumption experiments (including cocaine, codeine, d‑amphetemine, ethanol, ketamine, methohexital, morphine, pentobarbital, phencyclidine, and procaine) by using demand curves. They found that the demand curves revealed the typical mixed elasticity for all drugs studied, with demand being inelastic at lower prices and elastic at higher prices (like the curved line in Figure 1). Similar quantitative relations were found when data from 17 studies of human cigarette smoking were reanalyzed[39], and in an experiment with human cigarette smokers that explicitly employed a demand curve analysis[40]. This same inverse relation between substance consumption and price has also been observed in the natural environment for alcoholic beverages[41], cigarettes[42], and illicit drugs[43].

In summary, this research on substance use has consistently found a quantifiable relation between consumption and price such that consumption decreases as price increases (and vice versa), and this relation has consistently been found in different species, for different abused substances, in normal and clinical (“addicted”) populations, and in laboratory and natural environments. It is particularly important to note that the relations between consumption and price also hold for individuals “addicted” to the substances. Such findings indicate that the consumption of abused substances can be usefully described with the same analytic tools that apply to all commodities.

Policy note. The quantitatification provided by drug demand curves in laboratory experiments has provided valuable information for assessing abuse liability (the extent to which a drug may become the object of excessive consumption) and in developing pharmacological interventions for drug abuse. For example, using a demand curve analysis, Bickel and colleagues[44] showed that demand for pentobarbital was over 12 times more responsive to price increases than was cocaine, which means that for every unit of price increase, demand for pentobarbital was reduced 12 times more than the demand for cocaine. Thus, demand for pentobarbital was found to be much more elastic than demand for cocaine, which implies that pentobarbitol has less abuse potential than cocaine. Of course, this finding matches casual observation regarding the abuse potential of the two substances. Given that different games of chance also vary in their potential to generate disordered gambling, demand curve studies that compared demand for gambling opportunities across games could yield valuable information regarding the games’ “abuse liability.” This is work that could usefully be commissioned from behavioral economists by industry participants or regulators.

Effects of constraints on access to alternative activities. This choice perspective implies that the relative value of substance use will also depend on what other activities are available to potential abusers and on the benefits/costs of those activities. Understanding the contribution of this more general context is a key issue in substance use research, because strong preferences for substances arise in natural environments that presumably contain opportunities to engage in a variety of other activities. In general, substance use should be determined in part by the benefits/costs of other activity opportunities, with substance use increasing as the value of other activities decreases. A number of laboratory studies with both animal and human subjects have demonstrated this relationship with a variety of substances.[45]

The same relation found in the laboratory between substance abuse and alternative activities should also be found in the natural environment. Vuchinich and Tucker[46] applied this reasoning to an analysis of relapse episodes in a clinical sample of alcohol abusers after treatment. They found that relapses occurred more often when the rewards (e.g., intimate, family, social relations) in the recovering individuals’ environments changed from more to fewer alternative rewards being available. Also, decreasing substance abuse and maintaining recovery can be thought of as involving the reverse process wherein the reward structure shifts from fewer to more alternative rewards being available. Such relationships between the maintenance of recovery and increased access to valued rewards other than drinking have been found in community samples of alcohol abusers[47] and in heavy-drinking college students[48]

Policy note. A simple yet potentially powerful implication of the BE perspective for clinical interventions is that more attention should be paid to the availability of valuable nonsubstance-related activities and the extent to which access to them depends on abstinence or reduced substance use.[49] The research summarized above clearly shows that the variety and availability of valuable activities other than substance use can have a critical influence on substance consumption, and these relations have important relevance for the development of clinical assessment and intervention procedures and policies. It also is quite possible that similar relations exist between changes in disordered gambling and alternative activities, and that more attention should be paid to these relations in clinical interventions for gamblers. It would be interesting to have, for example, household surveys and subsequent microeconomic analyses that tested the effect of new sports facilities or entertainment complexes in communities where there were already organized gambling venues. Note that these effects would probably vary from one culture or country to another.

Summary. This very conventional BE work on substance use and abuse has been important because it demonstrated that abused substances obey the same economic laws as other, “nonaddicting” substances. That is, even in “addicts,” consumption goes down as price goes up (and vice versa), and consumption goes up if valuable alternative activities become more difficult to obtain (and vice versa). This in turn establishes that BE provides an appropriate framework for the study of addiction: addicts are economic agents in the same sense as non-addicts, even though there are important differences between their patterns of choice.

1.4. The matching law and temporal discounting

The outcomes of our choices sometimes occur immediately, but more often the outcomes do not occur right away. Most often, we choose to engage in activities now that produce positive and/or negative outcomes sooner or later in the future. The value of delayed outcomes is an especially important issue for understanding substance use and addiction, and also gambling, because individuals are repeatedly choosing between an immediately available but relatively small reward (substance consumption or the fun of betting) and engaging in alternative activities that will produce more delayed but more valuable rewards in the future (e.g., positive intimate, family, or social relations; vocational or academic success; purchase of something wanted and saved for). In these types of choices, gaining access to the more valuable but delayed rewards is facilitated by foregoing the less valuable but sooner reward; and, conversely, repeatedly enjoying the smaller sooner rewards risks reduced access later to the larger later reward. When someone’s choices in these life areas go awry (e.g., excessive drinking or gambling at the expense of marriage and family), it is almost irresistible for us to introduce irrationality into the analysis. Most such individuals are in fact irrational, in the strict sense of the economist, as defined earlier in this chapter. Like the fanciful cobraholic, their preferences are inconsistent over time; at Time A they (truthfully) report a strong desire to drink less or not at all in order to maintain a happy marriage and family life, but at Time B they drink to excess and thereby further threaten the very things they desired at Time A. Research derived from the matching law has produced tremendous strides in our understanding of such choices and their disorders. We will now describe the theoretical and empirical basis of this significant progress. In order to do so, we need to delve into some details of the analysis. This gets a little complicated, but the mathematics we’ll introduce here are no more complex than high school arithmetic, and we believe a firm grasp of the basics will reap benefits for the reader in later chapters.

Experiments on the matching law[50] typically give subjects opportunities to respond on option 1 and option 2, manipulate some aspect of reward (e.g., frequency, amount, or delay) derived from option 1 and option 2 choices, then relate the pattern of choices to the pattern of rewards received. In these experiments, both animals and humans make a very large number of choices across multiple sessions, each of which lasts about one hour, and the individual rewards typically are a small amount of food for animals and a small amount of money for people. The first aspect of reward studied was frequency, with the same size of reward occurring more or less frequently from option 1 and option 2 choices. Those studies found, for example, that if reward occurred twice as frequently from option 1 as from option 2, then option 1 was chosen twice as often as option 2. Because the proportion of choices for each option matched the proportion of rewards received from each option, the relation was termed the matching law. Later studies investigated varying the amount and varying the delay aspects of reward, and produced results similar to those that studied frequency of reward. For amount, if rewards occurred equally often from option 1 and option 2, but the option 1 rewards were, for example, twice as big, then option 1 was chosen twice as often as option 2. For delay, when the same size rewards occurred equally often on option 1 and option 2, but receipt of the option 2 rewards were, for example, delayed twice as long as receipt of the option 1 rewards, then option 1 was chosen twice as often.

The huge leap forward in our understanding of impulsiveness and self-control occurred when these scientists began studying choice between rewards that varied in both amount and delay, with rewards from one option being smaller but sooner and rewards from the other option being larger but later. Equation 1[51] is a version of the matching law (see section 1.2), and is a very accurate description of the relation between choice and rewards that vary in both amount and delay. In Equation 1, B₁ and B₂ represent the number of choices of option 1 and option 2, A₁ and A₂ represent the amount (size) of each reward received from option 1 and option 2, and D₁ and D₂ represent the delay between the choice and receipt of the reward from option 1 and option 2. This equation shows that relative choice of the two options is directly proportional to the relative amount of reward and inversely proportional to the relative delay of reward.

(1)

This relation shows that there is an unsurprising tendency in both animals and people to prefer larger rewards to smaller rewards, and to prefer sooner rewards to later rewards. But these two tendencies are put in conflict when the smaller reward is sooner and the larger reward is later. Choice of the smaller, sooner reward (SSR) is labeled impulsive, and choice of the larger later reward (LLR) is labeled as self-controlled[52]. The critical importance of this form of the matching law for understanding impulsiveness and self-control is the manner in which the value of rewards is predicted to vary with their delay. This prediction is that preference between the two rewards will depend on the temporal distance from them at which the choice occurs. At some choice points the individual will be impulsive (choose the SSR) and at other choice points the individual will be self-controlled (choose the LLR), even though nothing differs between the choice points except for the temporal distance from reward. Moreover, this predicted preference reversal is in sharp contrast to more traditional views of changes in reward value over time, as we discuss shortly.

Temporal discounting. These choice dynamics are best described with a hypothetical but concrete example[53]. Say choice of option one (the LLR) leads to a 6-unit reward (A₁ = 6) and choice of option two (the SSR) leads to a 2-unit reward (A₂ = 2), and that the LLR is available 8 time units after the SSR. (The reward units are measures of value – you can think of them in terms of how much a subject would pay for the reward. The time units can be any regular unit of time – one hour, three days, five seconds, etc.) These relations are shown in Figure 2, with time along the horizontal axis, reward value along the vertical axis, and the rewards represented as vertical boxes at the times they are available. In Figure 2, the 2-unit reward (A₂) is available at time 14, and the 6-unit reward (A₁) is available at time 22. As derived from Equation 1, the value of any delayed reward at any point in time along the horizontal axis is given by dividing the reward amount by the reward delay at that point (A/D), as shown in Equation 2.

(2)

In Equation 2, v_i is the present value of a reward of amount A_i that is available after a delay of D_i. In Figure 2, the curves to the left of the rewards were drawn using Equation 2; they trace the respective A/D fractions for the SSR and LLR as delay changes. Thus, these curves represent the value of the rewards during the times before they are available. Importantly, the reward with the highest value curve at the time the choice is made will be preferred. This change in reward value with delay is termed temporal discounting, because the present value of a delayed reward is discounted below what its value will be at the time it is received in the future. Equation 2 is one form of a temporal discounting function, because it describes how reward value changes with delay.

In Equation 1, the B₁/B₂ ratio measures whether the subject prefers option 1 or option 2: a behavior ratio greater than 1.0 indicates a preference for alternative A₁, while a behavior ratio less than 1.0 indicates a preference for alternative A₂. At Choice Point X at time 12 in Figure 2, the LLR is delayed by 10 time units (A₁ = 6, D₁ = 10), and the SSR is delayed by 2 time units (A₂ = 2, D₂ = 2). Inserting these amount and delay values into Equation 1 yields a B₁/B₂ ratio of .60, indicating a preference for option 2, the SSR (impulsiveness). At Choice Point X, the absolute values (from Equation 2) of option 1 and option 2 are .600 and 1.000, respectively. But preference is predicted to change if we further delay both rewards by 10 time units, so that choice occurs at a greater temporal distance from both rewards. This is shown as Choice Point Y at time 2 in Figure 2. Now option one is 6 units of reward delayed by 20 time units (A₁ = 6, D₁ = 20), and option two is 2 units of reward delayed by 12 time units (A₂ = 2, D₂ = 12). Inserting these modified amount and delay values into Equation 1 yields a B₁/B₂ ratio of 1.80, indicating a preference for option one, the LLR (self-control). At Choice Point Y, the absolute values of alternative one and alternative two (from Equation 2) would be .300 and .167, respectively. Thus, the values of both rewards are less at Choice Point Y than at Choice Point X, because of temporal discounting, and the shape of the discount function inherent in the matching law predicts that preference between the rewards will reverse simply with the passage of time. For example, someone offered a choice between 1 pizza at 1:00 PM 7 days from now and 2 pizzas at 1:00 PM 8 days from now probably would prefer the 2 pizzas delayed by 8 days. However, at 12:55 PM on the 7^th day, if the same person were offered a choice between 1 pizza in 5 minutes and two pizzas in 24 hours and 5 minutes, their preference may have reversed and, if so, they would choose the more immediate single pizza.

This research derived from the matching law has revealed that irrationality, in the strict sense of the economist, is in fact normal! This is critically important and worth repeating: Matching law research revealed that both animals and people are naturally disposed to have inconsistent preferences for future rewards as time passes towards the availability of those rewards. At time A they prefer a larger but temporally more distant reward, but at time B when a smaller but sooner reward is available, their preference reverses and they choose the smaller reward. In the hypothetical example in Figure 2, our subject preferred the LLR at choice point Y but later this preference reversed and our subject preferred the SSR at choice point X. This is just like the problem drinker or disordered gambler who says she wants to indulge less so as to have a happy marriage, but then later changes her mind and drinks or gambles. According to this work, having inconsistent preferences over time is the normal state of affairs for both animals and people. Thus, the question of why we sometimes behave irrationally perhaps should be replaced with the question of how we ever can ever resist imminent temptations and behave in ways that are consistent with our long-term interests.

Subsequent research[54] revealed inadequacies with the simple A/D fraction in Equation 2 as a generally useful temporal discounting function, and led to the slightly modified version in Equation 3.

(3)

In Equation 3, v_i, A_i, and D_i represent the present value of a delayed reward, the amount of a delayed reward, and the delay of the reward, respectively. The “1” in the denominator of Equation 3 prevents the infinite rise in reward value when delay is zero, as in Figure 2. The k parameter is a constant for any individual that is proportional to the degree of temporal discounting, with higher and lower k values describing greater and lesser degrees of discounting, respectively. Thus, an individual with a higher k value would discount delayed rewards more than an individual with a lower k value; the former individual therefore would be more impulsive than the latter individual. As discussed below, such individual differences in temporal discounting are critical to understanding substance abuse and addiction and disordered gambling.

The discovery that hyperbolic temporal discounting was natural meant a significant departure from previous notions regarding how the value of delayed outcomes changed over time. This isn’t surprising: previous theory had supposed that people were basically rational (in the economist’s sense) and as we’ve seen hyperbolic disconting is irrational. Dating back to Samuelson’s 1937[55] discounted utility model, both economics and psychology had assumed that discounting would conform to an exponential function, as in Equation 4.

(4)

In Equation 4, v_i, A_i, k, and D_i are the same as in Equation 3, and e is the base of the natural logarithms. The key difference between exponential and hyperbolic temporal discounting is that the rate of discounting is constant at all delays in the former but varies with delay in the latter. That is, with an exponential discounting function, equal increments in delay produce constant proportional decrements in reward value. With a hyperbolic discounting function, in contrast, equal increments in delay produce larger decrements in reward value at short delays than at long delays. This critical difference is illustrated in Figure 3. Again, in Figure 3, time is along the horizontal axis, reward value is along the vertical axis, and the rewards are the vertical bars at the times they are available. The top and bottom panels show hyperbolic discounting and exponential discounting, respectively. The curves to the left of the rewards represent their value during times before they are available. The hyperbolic and exponential curves were drawn using Equations 3 and 4, respectively. The key difference is in the shape of the curves. The exponential curves show a constant change in slope, but the hyperbolic curves are relatively flat at the left of the figure then rapidly rise as the rewards become closer in time. It was quite surprising and initially disconcerting to learn that normal human choice is irrational, as in the hyperbolic curves, and not rational, as in the exponential curves. But now we are armed with that knowledge and so are in a much better position to discover means to get people (including ourselves) to behave in ways that are consistent with their long-term interests and not their short-term interests.

1.5. Temporal discounting, substance use, and addiction

The evidence is now clear that temporal discounting of reward value is better described by the hyperbolic discounting function than by the exponential function.[56] These choice dynamics obviously are important for our understanding of addiction. The SSR and LLR are analogous to an alcohol or drug consumption episode and a more valuable but delayed non-drinking or non-drug activity, respectively. The choice dynamics that result from hyperbolic discounting are consistent with the ambivalence that is a critical feature of substance abuse and addiction patterns. That is, even when reward availability does not change, individuals will show ambivalence over time, and their preference for substance use will vary depending on the temporal distance to the availability of the substance and the alternative activity. The LLRs will be preferred before the point where the reward value curves cross, and substance use will be preferred after that point. But, once the person exits the situation involving imminent substance use, preference will revert back to the LLRs, and the individual may regret the substance use episode. Such ambivalence is a key feature of addiction.

Significant advances have occurred over the last 10 years in our ability to measure temporal discounting in humans. The basic procedure involves having subjects repeatedly choose between hypothetical money amounts, with a smaller amount available immediately and a larger amount available only after a delay. For example, we start by giving the subject a choice between R10 now (SSR) and R1000 in a week (LLR). Most, but probably not all, subjects would choose R1000 in a week. Then we give them a choice between R50 now and R1000 in a week, and again they probably would choose the LLR. We continue gradually increasing the amount of the SSR until at some point the subject’s preference will change and they choose the SSR. Say the subject chose R1000 in a week over R780 now, but then chose R800 now over R1000 in a week. Given this preference switch, we can reasonably infer that, for this subject, R790 now is equal in value to R1000 in a week. For this subject, R790 is termed the subjective equivalence point, or sometimes the ‘indifference point’, for R1000 in a week, since the subject apparently is indifferent between R790 now and R1000 in a week. Then the procedure is repeated with the R1000 available at several increasing delays, such as one month, six months, one year, five years, etc. A specific subjective equivalence point is thereby found for each of the delays. Figure 4 shows the equivalence points for two actual subjects in a discounting study done by one of us.[57] In Figure 4, the delay of the LLR is along the horizontal axis, and the money amount of the subjective equivalence points is along the vertical axis. These two subjects were chosen because they have very different degrees of temporal discounting, with the filled points showing little decrease in value even at long delays, and the open points showing dramatic decreases in value at even short delays. The equivalence points are subjected to a routine statistical procedure that ‘fits’ the data to Equation 3, the discounting function, which results in an estimate of the k parameter for each subject. (Fitting a function to data involves finding values in a function so that values produced by the function are close to those in the actual data. In this case, assuming equation 3 to describe the data, fitting is a matter of finding a k value such that equation 3 describes a line passing close to the set of indifference points for a given subject.) The k values for these two subjects are shown in the figure.

BE makes a straightforward prediction regarding differences in temporal discounting between problem substance users and normal populations. This prediction is shown in Figure 5, which again shows an SSR that is available at time 6 and an LLR that is available at time 10. The reward value curves in both panels of Figure 5 were drawn using the hyperbolic function in Equation 3, with the top and bottom panels showing the curves for relatively high and relatively low values of k, respectively. The higher k used in the top panel produces steeper value curves than the lower k used in the bottom panel. Preference between the SSR and LLR reverses with the passage of time in both panels, with higher and lower discounting producing sooner and later preference shifts, respectively. As suggested by Figure 5, an individual with a high degree of temporal discounting (top panel) would spend more time preferring the SSR (i.e., substance use) than an individual with a low degree of discounting (bottom panel). This leads to the prediction that k values would be larger among substance abusers compared to normal individuals. This prediction has been supported in several studies of temporal discounting that compared normal individuals with problem drinkers, heroin addicts, smokers, cocaine addicts, and disordered gamblers.[58]

Summary. BE research on substance abuse and addiction has produced significant advances in our knowledge about the nature of these disorders. We now know that demand for addictive substances can be analyzed with the same economic concepts and methods as other commodities. We now know from the matching law research that inconsistent preferences over time between and SSR and an LLR is the normal state of affairs for humans. And we now know that one aspect of individuals with substance abuse or addiction problems is that they discount the value of future rewards much more so than do individuals without those problems. These basic empirical facts have provided a solid foundation for the development of a number of behavioral theories of addiction, which are summarized in the next section.

1.6. Three BE theories of addiction

The three theories briefly described in this and the next section all take the phenomenon of matching behavior, and especially hyperbolic temporal discounting, as their starting point. They differ in details of how broadly they take the scope of matching to be, what other considerations are important for addiction, and over whether there is anything particular to be said in the case of disordered gambling.

Ainslie’s Theory of Addiction Based on Hyperbolic Discounting. Far more than any other author, behavioral economist and psychiatrist George Ainslie[59] has vigorously and extensively pursued the implications of the preference reversals caused by hyperbolic temporal discounting for understanding impulsiveness and impulse control in addiction, as well as in obsessive-compulsive disorders, dissociative disorders, bulimia, itches, physical pain, and flaws of character.

Hyperbolic discounting can explain impulsive choices that are later regretted, and thus can explain the addict’s repeated surrender to temptation, despite earlier resolutions to quit or curtail substance use. That is excellent progress as far as it goes, but so far it does not explain how such impulsiveness can ever be avoided. How do addicts and the rest of us at various times resist temptation and exercise self-control? More precisely, why is the LLR ever preferred to the SSR, especially when the SSR is immediately available? Ainslie has described four processes that can be used to subvert preference reversals and thus can lead to choices that are more consistent with our long-term interests.

External Commitment. The most obvious self-control tactic is precommitment, meaning the deliberate engineering of the physical or social environment so that the impulsive choice becomes impossible or much more difficult to make. In Figure 2, for example, the LLR is preferred at time 2 but the SSR is preferred at time 12. An individual with this preference pattern could exercise self-control by doing something at time 2, when the LLR is preferred, that makes it impossible or very difficult to reverse course and choose the SSR when it becomes preferred at time 12. This device is exemplified by the myth of Ulysses and the Sirens, in which Ulysses chained himself to the mast of his boat before sailing past the deadly temptation of the sirens, whose charms he knew he’d be unable to resist if he left himself free to succumb. Notice that Ulysses thereby showed awareness of his own disposition for irrational preference reversal, and used this foreknowledge to block its dangers. An obvious example from the addictions field is the voluntary ingestion of the drug disulfiram, which a dependent problem drinker knows will make her feel very ill if she consumes alcohol. Interestingly, even pigeons will make precommitments of this sort under carefully set up conditions[60].

Control of attention. This refers to restricting the processing of information about the availability of the SSR. It is somewhat similar to the Freudian concept of repression, although originating obviously from a very different theoretical base. From Ainslie’s viewpoint, the avoidance of information, unlike in repression, can be conscious as well as unconscious. For example, a recovering alcoholic may avoid walking past her usual pub. In Chapter 3, we will see reasons for thinking that this is probably an essential aspect in the control of any full-blown addiction once it has developed.

Preparation of emotion. This describes attempts to inhibit emotional responses associated with the SSR or to augment emotions that are incompatible with the appetite for it. In the case of addictions, it mainly applies to the cognitive control of so-called ‘craving’ responses. For example, people trying to quit smoking may absorb themselves in pictures of diseased lungs, or put the money they had been spending on cigarettes in a kitty used to reward themselves with periodic indulgences, so as to make vivid to themselves the financial costs of smoking in addition to its health costs.

Personal rules. This process is by far the most important category of methods of self-control described by Ainslie, and the one that is used to create a major extension to theory. This part of Ainslie’s work provides a BE account of willpower, that mysterious force that is widely believed to be responsible for how and when we resist temptation and behave according to our long-term interests. We will explain personal rules at greater length than the other devices, both because this idea will feature most prominently in our policy discussion in Chapter 4, and because it involves more complexities.

The basic case of temporary reversals of preference due to hyperbolic discounting involves a single SSR-LLR pair, as in several of the previous figures. But, of course, in life outside the laboratory many identical or very similar SSR-LLR choices will occur repeatedly in the future. Ainslie’s personal rules concern whether and when an individual will perceive each SSR-LLR pair as a separate choice, or whether she will perceive the choice as being between a whole series of SSRs and a whole series of LLRs. If the individual perceives the choice as being between whole series of SSRs and LLRs, then she is engaging in reward bundling, a potentially powerful self-control process. Reward bundling is illustrated in Figure 6. In the figure, time is along the x-axis and reward value is along the y-axis. The figure shows four SSR-LLR reward pairs spread out at the times they will be available. The solid curves to the left of the rewards were drawn from Equation 3 and represent the value of the rewards during the times before they are available.

If the person facing this series of choices takes the SSR-LLR pairs one at a time, then we would see repeated preference reversals. At the far left of the figure, she prefers the Pair 1 LLR, but when the Pair 1 SSR is imminent, its value rises above that of the Pair 1 LLR, so her preference reverses and she chooses the SSR. After Pair 1 is past she faces Pair 2. Initially the Pair 2 LLR is preferred, but again as the Pair 2 SSR becomes imminent its value rises about that of the LLR and her preference once again reverses. The same thing then would happen for SSR-LLR Pairs 3 and 4. This is a common pattern for individuals with impulse control problems. Many people wake up each morning preferring not to do what they did yesterday, not to drink or smoke or gamble or indulge in whatever other self-control problem they have, but then later in the day as the object becomes available they succumb to temptation.

On the other hand, if the person facing this series of choices did not take them one at a time, but instead viewed their situation as one single choice between the whole series of SSRs and the whole series of LLRs, then we could see much more self-control. In Figure 6, the reward value curves for all four SSRs and all four LLRs are summed during the times before the Pair 1 SSR is available. These summed value curves are the dotted lines at the left. The key point is that the LLR sum is greater than both the SSR sum and the value of the Pair 1 SSR at the time the Pair 1 SSR is available. So, an individual who viewed this as one choice between series of rewards would prefer the LLR series even when the SSRs are immediately available. This also is a common pattern for individuals without (major) self-control problems. Many people wake up each morning not even thinking about drinking, smoking, or gambling, because they didn’t do it yesterday, or the day before yesterday, or the day before that; some time ago they made the choice not to engage in the SSRs and so now are reaping the greater benefits of the LLRs.

To avoid a reversal of preference, a person can bundle a whole series of choices together in anticipation of the higher aggregate reward that would be obtained from preferring the LLRs. She does this by adopting personal rules that dictate the choice to be made in a whole class of conflict situations involving the need to delay gratification. These personal rules correspond to the idea of ‘principles’ or ‘universal rules’ of behavior that have been recommended as a means of achieving will-power by philosophers down the ages. The person is seen as committing themselves to a course of action by making private side-bets, a form of personal rule-making in which the current choice serves as the best predictor of what future choices will be. If the current SSR is successfully resisted, the bet is won, and the expectation of future reward is proportionately strengthened and the ability to overcome similar temptations in the future is enhanced. On the other hand, if the SSR is not resisted, the bet is lost, and also lost is whatever elements of her view of herself had been staked, with a decreased expectation of being able to resist the SSRs of the same kind in future.

Personal rules and private side bets can be illustrated with Figure 6. Say the individual has a strict rule of “only go to the bar for a few drinks on Fridays and legal holidays.” If followed, this rule would allow enjoyment of drinking on the designated days but would also produce choices of the LLR over the SSR (drinking) the rest of the time. But when the Pair 1 SSR is immediately available on a regular weekday, there is nothing in the physical or social environment that prevents her from choosing the SSR or that even makes choosing the SSR much more difficult. What is at stake in the private side-bet is the difference between the summed LLR and SSR value curves at the time the Pair 1 SSR is available, because the reward that is chosen from Pair 1 strongly predicts the reward that will be chosen from subsequent pairs. If the first SSR is resisted, she gets the first LLR plus the enhanced expectation of also getting subsequent LLRs reflected in the LLR sum. On the other hand, if the first SSR is not resisted, she gets that SSR but also gets a reduced expectation of getting any of the subsequent LLRs. Because of the predictive value of current choices for later choices, which is a reward the chooser gets right now as she makes the choice, personal rules can bring into the present the value of rewards in the future, and thus allow people to overcome the effects of their natural hyperbolic discounting.

Take, for example, an overweight person who wants to lose some weight primarily for health reasons. If he thinks truly seriously about his situation, he can notice that when he chooses to fix or not to fix himself another sandwich, he isn’t just choosing between satiation now and sticking with his diet today. If he prefers to be trim in the distant future, his choosing a sandwich now provides him with evidence that he won’t choose his diet then either; for hyperbolic discounters, present choices predict future choices. In light of this, perhaps he can frame his present choice to reflect the fact that it isn’t just between immediate streams of benefits and costs, but has implications all the way along the flatter part of his discount curve into the future. If all those preferred states of fitness in the future are at stake now, then their value, added together, might be enough to trump the present value of the sandwich. It won’t typically be enough for the person just to tell himself this; if it were, self-control would be easy and addiction and impulsiveness would both be mysterious. The would-be dieter needs to arrange his patterns of action so as to make it impossible for him to avoid keeping the future implications of his action in mind while he makes his present choice. Suppose he therefore frames a personal rule to the effect that an extra sandwich is allowed only on a day after his visits the gym. Then, if today isn’t a day after a gym session, his choice is between a sandwich now and now facing the long-term ruin of his personal rule. The key thing to see here is that because present choices predict future choices the personal rule is a presently valuable asset. It bundles a whole series of future benefits and puts them at stake in the present moment. If he throws his rule away by breaking it, he will suffer from his knowledge of the loss of it here, where discounting is at a minimum.

Ainslie goes on to propose that, rather than being characterized by rational and consistent preferences as conventionally assumed, the self is in fact made up of conflicting preferences for a variety of rewards that, because of hyperbolic discounting, achieve dominance over each other at different points in time. Each appetite for and motivation to acquire each reward is described as an ‘interest’ in that the reward and the interaction between these competing interests is portrayed as an internal economic market-place. In particular, according to this model, the interests of a present self compete strategically with those of an anticipated future self that may attempt to subvert them. This battle may be complicated by the existence of mid-range interests against which short- and long-range interests may combine forces. This strategic interaction of interests can be illustrated by further complicating the example earlier of the overweight person. In this example, the short- and long-range interests are the extra sandwich now and being slim and healthy in the future, respectively. But there are other interests at intermediate ranges between short and long that may interact with the short- and long-range interests and impact the critical choice to have or not to have that extra sandwich. Say, for example, he has a mid-range interest in following his favorite football team, and another mid-range interest in getting dates with attractive women. Say on a day he didn’t go to the gym, his team happens to be playing a televised game. In this situation, his extra sandwich interest may recruit his sports interest and both together then overcome the future-slim-self interest, and he chooses to watch the game and eat the extra sandwich. Alternatively, say on a day he didn’t go to the gym, he meets an attractive woman but makes no overtures because he is overweight. Later when he gets home he will be tempted to make the extra sandwich, but the future-slim-and-healthy-self interest could recruit the attractive date interest, and both together overcome the extra sandwich interest, so he resists the extra-sandwich temptation and gets closer to his intermediate and long range goals of dating attractive women and being slim and healthy.

So how are relatively stable choices ever made in this intrapsychic free-for-all of successive motivational states attached to short-range, mid-range, and long-range interests? Ainslie here borrows concepts used to analyze interpersonal or international conflict[61] and applies them to the intrapersonal bargaining situation. [62] In particular, he uses the model of repeated instances of the ‘prisoner’s dilemma’ game, in which players must choose between cooperative and non-cooperative options, to represent a bargaining process involving successive motivational states.

Here is a version of the standard illustration of the prisoner’s dilemma: Two men commit a burglary and are arrested while in possession of burglar’s tools. Possession of the tools is a minor crime. The police have no evidence against the men, other than the tools, so they offer each man the following incentives. If neither man confesses, they won’t be convicted for the burglary, but both will get a 90-day sentence for the tools. If both confess to the burglary, they’ll both be convicted and both will get 2-year sentences. If one confesses and the other doesn’t, the confessor will go free and the non-confessor will get a 5-year sentence. The dilemma is: should each man do what is best for himself alone, or should he do what is best for both men as a pair? For each thinking only of himself, the best thing to do is confess, since he’ll then get his best possible outcome - going free if the other doesn’t confess or a 2-year sentence (which beats 5) if the other does confess – regardless of what the other does. The other prisoner reasons in exactly the same way, since his situation is identical. So they both confess and get 2-year sentences. But now notice that when we think of the pair as a whole, they’re both worse off than they would have been had neither confessed – then they’d have each got just 90 days. Ainslie has extended such interpersonal (i.e., between people) strategizing to consider the intrapersonal (i.e., within the person, between interests) bargaining we all engage in over time. Rather than two people at the same time considering what is best for themselves alone or what is best for the pair, we now have one person at different times considering what is best for the present self alone or what is best for both the present self and the future self as a pair. When the present self confronts an SSR-LLR pair of rewards, the best thing for the present self alone is to choose the SSR, because only the SSR is available at the present time. On the other hand, the best thing for the present self and the future self together as a pair is to choose the LLR, because then over time they’ll receive the greater benefits of the series of LLRs.[63] This intricate and subtle arena of competing personal rules and complex intrapersonal bargaining strategies leaves plenty of room for the evasions, distortions and other forms of self-deception well-recognized in the addict’s behavior, as well as in ordinary human conduct, and these are deduced by Ainslie from his theory. In Chapter 4 we will discuss the implications of these ideas, especially ways to promote reward bundling, for dealing with gamblers.

Two molar BE theories of addiction. The temporal inconsistencies in preference caused by hyperbolic temporal discounting are a critical element in explaining addiction from the BE perspective. Hyperbolic discounting alone, however, is widely regarded[64] as an insufficient explanation. Thus, choice dynamics other than hyperbolic discounting have been proposed to explain the apparently self-defeating behavior pattern of addiction. In this section, we sketch two molar theories (see Section 1.2 above) of addiction that have been developed within the BE perspective: (1) Richard Herrnstein’s melioration theory of addiction^[65], and (2) Howard Rachlin’s relative theory of addiction[66]. Both these theories accept hyperbolic discounting as fact, so both are entirely consistent with Ainslie’s work. However, these theories focus on an aspect of consumption other than hyperbolic discounting. That aspect is that consumption of addictive commodities during one time period reduces the utility derived in later time periods both from the addictive substance and from engaging in other activities. In general, utility is defined as the benefit derived from engaging in an activity. Because of tolerance, consumption of the addictive commodity in one time period reduces utility of the same level of consumption in future periods. For example, if an individual consumed two drinks every night over several weeks, she would get a pleasant high from the two drinks early in the period but the two-drink high would gradually diminish until by the end of the period the two drinks produced little effect. For this reason, regular drinkers must either be satisfied with diminishing benefits from drinking, or, as more often happens, increase their drinking over time to obtain the same high. Moreover, because consuming addictive substances can disrupt one’s ability to engage in and enjoy alternative activities, consumption of the former in one time period reduces the utility derived from the latter in later time periods. Thus, these theories propose a sort of intertemporal dependency between consumption and utility in different time periods that is critical in accounting for addiction. One final preliminary note: a key aspect of both theories is the difference between local utility and global utility, a distinction made earlier in this chapter (Section 1.2). Local utility refers to the utility one gets immediately from an action, while global utility refers to the overall utility one gets from a given sequence of actions.

Herrnstein’s melioration theory of addiction. Because of the process of melioration (see Section 1.2), in addition to hyperbolic discounting, an individual’s choices are relatively insensitive to global utility and much more sensitive to the comparison of local utilities. Herrnstein’s theory shows how this process can lead to addiction. As noted above, a key element of the melioration theory is that substance use lowers the utility derived from such use and also from non-substance use activities. This process is depicted graphically in Figure 7, in which utility is along the vertical axis, and the relative choices of substance use and of non-substance use activities is along the horizontal axis. The farther to the right along the horizontal axis, the more the individual is choosing substance use and not choosing other activities – the process of addiction – and conversely for movement to the left along the horizontal axis. So, at the far right, the individual is consuming the substance and doing little else, and at the far left, she is doing a lot of other things and not consuming the substance. The local utility of substance use, the local utility of other activities, and global utility are signified by lines 2-3, 1-4, and 1-3, respectively, at the various distributions of choices along the x-axis. The melioration process dictates that choices are made according to which choice option provides the highest local utility at the time. As seen in Figure 7, even though both local utilities and global utility are being reduced by substance use, the individual continues to use the substance. Increasing levels of addictive commodity consumption, movement to the right along the horizontal axis of Figure 7, reduce both local and global utility, but the melioration process continues to drive addictive consumption because the local utility of that consumption remains higher than the local utility of not consuming.

Theoretically, the process of recovery from addiction may involve a restructuring of how the choice alternatives are perceived. The above discussion assumed that only events during the substance use episode are included in the computation of local utility derived from substance use. So, the utility derived from an evening of heavy drinking is kept separate from the next morning’s hangover. If, however, the temporal boundary of what counts as local utility is expanded, this may alter the computation of local utility so that the value of a substance use episode is changed. In this example, if local utility is expanded to include the time spent drinking plus the hangover, and not just the drinking, this may reduce the utility derived from drinking and therefore lower its value. Notice that this is bundling again. From one BE model to another, successful bundling keeps emerging as the crucial defense against addiction.

Rachlin’s relative theory of addiction. Rachlin’s model is similar to and entirely consistent with Herrnstein’s. The key difference is that, whereas Herrnstein grouped all activities other than substance consumption into one large category, Rachlin focuses on a particular class of activities involving social interaction. Rachlin’s theory is based on two important aspects of social interaction:

First, social interaction is a close economic substitute for various addictive substances (i.e., smoking, drinking, drug use). When two commodities are substitutes, their consumption varies inversely; the more of one that is consumed, the less of the other that is consumed (and vice versa).

Second, the utility derived from substance consumption and from social interaction changes in opposite ways as these activities are engaged in. We noted earlier that, because of tolerance, the utility derived from a given level of substance consumption decreases over time. On the other hand, the utility derived from a given amount of social interaction increases over time. So, if you have two drinks tonight, those two drinks would provide much more utility if you haven’t had any alcohol for several weeks than if you’ve been having two drinks every night for several weeks. Conversely, if you encounter a stranger at a bus stop and converse for five minutes, that brief interaction would provide more utility if you’d recently been socially interacting regularly than if you’d recently been socially isolated.

Rachlin termed those changes in the utility derived from substances and from social interaction the price habituation of substance consumption and the price sensitization of social interaction. He framed this in terms of price changes because a given level of utility costs more or less over time. The more a substance is consumed, the higher the price for a given unit of utility (and vice versa). Conversely, the more one interacts socially, the lower the price for a given unit of utility (and vice versa).

How the price habituation of addictive commodities and the price sensitization of social interaction can lead to addiction is illustrated in Figure 8. Like Figure 7, in Figure 8 movement to the left along the horizontal axis signifies less drug use and more social interaction, and movement to the right along the horizontal axis signifies more drug use and less social interaction. The utility derived from a given choice of substance use and social interaction is signified by lines 1-3 and 2-4, respectively, at the various distributions of choices along the horizontal axis. The price habituation and price sensitization of substance use and social interaction, respectively, are shown with lines sloping down to the right. Movement to the right raises the price of both substance consumption and social interaction (a given unit of utility of either costs more because the substance use has been increasing and social interaction has been decreasing. Conversely, movement to the left lowers the price of both substance consumption and social interaction (a given unit of utility of either costs less because the substance use has been decreasing and social interaction has been increasing.) Given the melioration process, the person will choose the option that provides the highest utility at the time the choice is made.

As an example, assume that over a period of time an individual’s choices are stable with a pattern of drinking and social interaction at the point of intersection of lines 1-3 and 2-4 in Figure 8. This choice pattern involves relatively more social interaction and less substance use. However, say an event (e.g., marital separation) then occurs that makes social interaction more expensive in the sense that it now is more difficult to obtain, which would reduce the number of choices of social interaction. Because drinking and social interaction are mutually substitutable, after the event the individual begins to drink more to make up for less social interaction (movement to the right along the horizontal axis in Figure 8). Because of the price habituation and price sensitization of drinking and social interaction, respectively, drinking more and socializing less will decrease the utility derived from both drinking and socializing. But even though both utilities are decreasing as the individual moves to the right, the choice dynamic of relative addiction theory continues to drive addictive consumption because the utility from drinking exceeds that of the utility from social interaction.

Given its focus on the utilities and costs of substance consumption and social interaction, relative addiction theory would look to these variables as being important in the recovery from addiction. Events that produced a large enough increase in the price of addictive substance consumption and/or a large enough decrease in the price of social interaction could tip the balance of the cost/benefit ratios so that an addicted individual would curtail or quit their consumption.

1.7. Extensions of BE research to gambling

We hope we have now persuaded the reader that BE theory and research has significantly advanced our knowledge of substance abuse and addiction. Because the behavioral manifestations of substance abuse and addiction and of disordered gambling are so similar, if not the same, there is every reason to be confident that BE could make significant contributions to our understanding of disordered gambling as well. (Note that we are careful here to say that the behavioral structure of disordered gambling and addiction are similar. We will later argue that only a subset of those with pathological gambling have a genuine addiction. But BE’s basic model applies to many behavioral patterns – from overeating to frittering away too much time playing internet solitaire – that are also not addictions.)

As applied to disordered gambling, the larger, later rewards (LLRs) would appear to be the same as those contrasted with substance abuse and addiction: adaptive functioning in key life-health areas such as intimate, family, and social relations, and vocational success. The smaller, sooner rewards (SSRs) are different, however, because the gambler takes nothing into her body as does the substance abuser or the addict. Neuroscience is now telling us about this difference in detail, and we will describe these advances in Chapter 3. In any case, the apparent irrationality of the disordered gambler and of the substance abuser or addict are the same: At time A they truthfully report that they want to gamble less or not at all so as to improve their lives in more important ways, but at some future time B that stated preference reverses and they gamble. Then later at time C, when regret sets in, they are confronted again with their inconsistent preferences for a better life and excess gambling.

In this section we will discuss Ainslie’s and Rachlin’s specific perspectives as applied to disordered gambling. Both models are entirely consistent with each other, but they focus on different aspects of impulsiveness and self-control. Because Rachlin’s model of addiction in general is consistent with Ainslie’s, and then adds to it a particular account of disordered gambling , all policy and intervention implications of Ainslie’s model will carry over to Rachlin’s and then be supplemented by it. We should make clear that all that is being attempted at this point is logical extrapolation from theoretical perspectives. We will turn to the implications of direct empirical studies in the succeeding chapters of this report. Also note that in this section we will only briefly mention practical implications, as they will be spelled out in considerably more detail in Chapter 4.

We’ll first discuss how Ainslie’s model can be used to conceptualize the choices faced by the disordered gambler. As noted earlier, in this case the SSRs are opportunities to gamble and the LLRs are quality of life concerns in other areas. As in Figure 6, the gambler faces a long series of such choices. Any of Ainslie’s self-control tactics discussed earlier (external commitment, control of attention, preparation of emotion, and personal rules) could be brought to bear to make current choices more consistent with long-term interests. We will focus, however, on personal rules (reward bundling), because this is by far the most important self-control tactic and the one with the most extensive implications.

An individual could develop personal rules that are in force before she even enters the casino. As with the earlier discussion of rules regarding drinking, a gambler could, for example, generate a rule that states she can only go to the casino on Fridays and holidays. If followed, this rule would allow her to enjoy gambling (the SSR) on a relatively small number of designated days, and also enjoy the greater benefits of the LLRs on days she is not gambling. The more times the rule is followed, the stronger it will become as she develops heightened expectations of continuing to be governed by it in the future and in reaping the benefits of the LLRs. As with all personal rules, however, it will always be vulnerable to being broken if a coalition of interests gets strong enough to subvert it.. If given an opportunity, the short-term interest in gambling could recruit a mid-range interest to its side of the conflict and subvert the rule. Suppose, for example, that on a Friday in the casino the gambler meets an attractive man who asks her if she will be back on Saturday night. Rigid adherence to the rule would require her to say “no,” which she may well do. But as Saturday night approaches, her mid-range interest in dating attractive men could combine with the short-term interest in gambling again on Saturday, which may be enough to overwhelm the long-term interest, and the rule. If she gambles on Saturday, she would have broken the rule, and in so doing has now reduced her present expectation of rule following in the future and of receiving future LLRs.

Gamblers can also develop personal rules that are in force while they are in the casino. As an example, suppose there are two gamblers, A and B, and that A has no personal rules regarding the extent of his gambling while in the casino. For A, with no rule, there is nothing at the moment that can bring possible future LLRs to bear on his current decisions. With nothing to connect his present self, which strongly prefers the SSRs of gambling, to his future self, which will prefer the LLRs, the SSRs will have their way with his decisions and he will gamble until he runs out of money or the casino closes. On the other hand, suppose B has a rule that states: “I’ll count my money every hour. If I’m losing money at the count I’ll leave the casino. If I’m winning money at the count I’ll stay another hour, but under no conditions will I stay longer than three hours.” At each count B faces the choice of doing what the rule dictates, or continuing to gamble. If he follows the rule, then it becomes stronger and he has increased his present expectation of continuing to reap LLRs in the future; if he breaks it he will bring on the reverse present expectation. If the rule collapses and he fails to generate another, effective, one then he will be in the same situation as gambler A. Of course, like rules in force outside the casino, rules for choices inside the casino can be subverted by a coalition of interests. Suppose at the hour 2 count B is losing money, and the rule enjoins him to leave the casino. But, it happens that 15 minutes ago a valued business contact arrived at the table and had been looking at him expectantly. His short-term interest in continuing to gamble could coalesce with the mid-range interest in networking and together they may subvert the rule.

These examples of applications of bundling in gambling situations have been simple and fanciful. We will discuss more serious measures that could be adopted as public policies to make it easier for people to bundle while gambling, or that could limit casinos’ capacities to inhibit their bundling, in Chapter 4.

Rachlin’s relative addiction theory supplements Ainslie’s emphasis on intrapersonal bargaining with additional, specific implications for gambling behavior. Rachlin assumes that gambling behaves like the consumption of addictive substances in two ways. First, the more one gambles, the less utility one derives from a given amount of gambling. In other words, gambling is price habituated like addictive substances, so there is an inverse relation between the extent of gambling and the utility derived from gambling. Second, gambling is mutually substitutable with social interaction, so the less one gambles the more one interacts socially (and vice versa). If Rachlin is on the right track, then measures should be sought that increase the price of entry to gambling situations for disordered gamblers, and/or decrease their price of entry to situations of social interaction. We will discuss these possibilities more fully in Chapter 4.

As mentioned earlier, Rachlin is the only BE theorist to date who has an explicit and specific model of problem gambling. According to this theory[67], a person at risk for problem gambling is prey to an insidious perceptual process or accounting scheme. We will introduce the idea by referring the reader back to our earlier discussion of Herrnstein’s point that, the size of the behavioral unit over which utility is calculated is not carved in stone. With drinking, for example, is the utility of drinking calculated by adding up each individual sip, each drink, or each drinking episode? In that same discussion, we also noted that recovery from addiction may involve a process of restructuring the alternatives so that the behavioral unit over which utility is calculated is made larger. An individual whose drinking behavioral unit also included the next day’s hangover would be more likely to reduce their drinking than an individual whose behavioral unit included the night of drinking alone.

In the case of gambling, the question of the behavioral unit over which utility is calculated concerns how bets are aggregated into larger units. At one extreme end of a continuum, each bet could be seen as a single, independent event. Alternatively, at the other end of the continuum, each visit to the casino could be seen as a single, independent event (where an ‘event’ just names the unit used for cost-benefit analysis by the person). Rachlin argues that the disordered gambler sorts her gambles into strings in which each string terminates with a win. That is, she thinks of each win as redeeming the sequence of losses that preceded it, regardless of how long that sequence was in any given case. Her losses redeemed, the DG is ready to start a new sequence, with a clean record, following every nod from lady luck. This perceptual or bookkeeping scheme interacts with hyperbolic discounting in an insidious way. Each string the gambler perceives will have the form of a win preceded by n losses where n ≥ 0. It is the sequences in which n is largest that account for most of the financial distress – this is what makes problem gambling a problem. But now: the overall outcomes of long sequences will have their values hyperbolically discounted relative to the elements of short sequences. Long sequences, as noted, are where the losses are disproportionately concentrated. Thus, the gambler weights wins after short strings at closer to their full value than wins after long strings, with the average discrepancy being greater the worse the ratio of wins to losses!

Let us show an example. Ls and Ws below represent sequences of losses and wins, respectively, in a simple game of tossing a coin:

WLWLLWWLLLWLLLWWWWLLWW

There are 11 wins and 11 losses in the entire string, so as a whole it is neither positive nor negative in value. In the same W-L sequence somewhat modified below, the outcomes are organized as in Rachlin’s theory, with each sub-string terminating with a win. The numbers below each win represent the overall outcome for each substring:

W	LW	LLW	W	LLLW	LLLW	W	W	W	LLW	W
+1	0	-1	+1	-2	-2	+1	+1	+1	-1	+1

Summing the outcomes of the substrings again results in zero, so the overall outcome is neither positive nor negative. Importantly, however, because the W that terminates each sub-string is some distance into the future when the string begins, the present value of the overall outcome of the sub-string will be discounted. Using the hyperbolic discount function in Equation 3, with k equal to 1 and the delay equal to the number of coin tosses in the substring, the actual overall value and the discounted overall value of each sub-string would be:

	W	LW	LLW	W	LLLW	LLLW	W	W	W	LLW	W
Undiscounted	+1	0	-1	+1	-2	-2	+1	+1	+1	-1	+1
discounted	+1	0	-.33	+1	-.67	-.67	+1	+1	+1	-.33	+1

Summing the discounted values of each sub-string yields a value of 4, a decidedly positive overall outcome. Because of this perceptual organization of behavioral units into strings terminating in a win, the disordered gambler perceives positive value when in reality, in this example, the overall outcome is neutral. Because of the perceptual organization and hyperbolic discounting, the disordered gambler fools himself into perceiving positive value where in fact there is none.

This looks like a disastrous way to think, so it is initially surprising that, as Rachlin documents, animal behavior suggests it to be common in nature. Why would natural selection have built so perverse a perceptual quirk? The answer presumably is that in normal environments rewards are distributed in a patchy way. Thus when an animal finds any reward in an area, it should persist through some disappointments in searching the region further. But modern gambling environments, unlike any in nature, are built to exploit the bias for following up reward. Thus a disposition that was helpful to our ancestors is harmful to many of us. Nevertheless, this notion suggests that we consider possible policy interventions that encourage disordered gamblers to employ alternative accounting schemes.

1.8. How many kinds of disordered gambling are there?

Disordered gambling (DG) is simultaneously a social problem, a clinical challenge facing individual people and their counselors and physicians, and a scientific puzzle. It is not easy to keep inquiry motivated by all of these considerations from being, on the one side, monopolized by one of these considerations to the detriment of the others, or, on the other side, falling into incoherence. As Howard Schaffer[68] notes, DSM-IV equivocates on whether it regards pathological gambling (PG) as a ‘real’ phenomenon when it cautions that “[I]nclusion here, for clinical and research purposes, of a diagnostic category such as Pathological Gambling … does not imply that the condition meets legal or other non-medical criteria for what constitutes mental disease, mental disorder, or mental disability.” Is the motive for this equivocation mainly derived from social policy considerations or scientific ones? There is probably no straightforward answer to this question. But to the extent that there was scientific consensus, it would be bizarre for DSM to continue saying what it does. If there were agreement that pathological gamblers are merely people who are irresolute in the face of bad habits, then PG should be removed from the DSM. If, on the other hand, there were consensus that some people showing PG symptoms were genuinely afflicted by a systematically measurable and sui generis disorder they could not control, then the current equivocation would be a betrayal of patients.

The problem here is not just that we remain somewhat ignorant about PG (and DG). In science we are always somewhat ignorant; but we act, at any given time, as if our most carefully evaluated beliefs are the true ones. The deeper problem, as Shaffer also notes, is that attempts to settle on an operationalization of PG are pulled in three directions. First, there is the fact that “the clinical issue … is not now, nor has it ever been, whether pathological gambling is an addiction or the result of a biobehavioral vulnerability. From a clinical perspective, this issue involves establishing a working formulation that clinicians and patients can share … From the treatment side, there is little or no value to understanding any individual as addicted or mentally disordered unless it permits clinicians to choose a treatment plan that will maximize the well-being of the patient.” Second, there is the fact that “[f]or science, the value of improving our understanding of pathological gambling and addiction rests in the development of better theory. Improved theory can guide better research.” Third, and finally, “[f]rom a community perspective, as our understanding of addiction and pathological gambling improve, then the vehicle for more effective social policy emerges.”

Shaffer unintentionally (we presume) exaggerates this tension into being irresolvable by implying that social policy concerns trump the others. “The relationship between pathological gambling and addiction,” he announces, “ultimately rests upon socio-cultural acceptability. After all, “It is best to think of any affliction – a disease, a disability … – as a text, and of ‘society’ as its author.”[69] The reason this makes the issue intractable is that ‘society’ would only arrive at a policy consensus if either (i) clinicians developed a single set of diagnostic and treatment practices that were overwhelmingly and demonstratively effective, or (ii) scientific researchers arrived at a clear consensus among themselves on the facts about what PG is – to which ‘society’ could then simply be advised that it had better pay attention, on grounds that preferring falsehoods is just self-defeating.

Now, since PG is clearly a very complex phenomenon, it is hard to imagine any process by which (i) might come about other than by (ii) coming about. Achieving (ii) would not guarantee (i), as the case of cancer shows.[70] Still, the point remains: either we must hope we can make progress on (ii), and thereby make progress on treatment and policy clarity, or we should give up and admit we’re stuck.

For reasons we’ll present in detail in Chapter 3, we believe that science has recently made considerable progress indeed towards (ii). We stress ‘recently’; the majority of the groundbreaking work we will review has occurred over the past two years. Note that progress towards (ii) does not have to – and in fact has not – consisted in boiling disordered gambling phenomena down to just one thing. Progress can and has consisted in sorting the phenomena clearly into distinct piles. These piles are not importantly sorted, at this point in time, by reference to developmental etiology; we still don’t know what causes disordered gambling or any one of its subspecies distinguished in section 1.1, in the sense of knowing why they afflict some people rather than others and in some circumstances rather than others.[71] Instead, the piles are sorted by reference to models of how different disordered gambling syndromes maintain themselves. This is at once useful for both treatment and policy purposes: to the extent that one knows how a syndrome is maintained, one knows, at least in principle, something about what would be required to disrupt that maintenance.

Most of the progress to which we just alluded has been in neuroscience. This is not so surprising: across a wide range of applications, the neurosciences show every sign of being the great scientific story of the 21^st century, as physics and information science were in the 20^th. Shaffer, notwithstanding his probably unintended undermining of optimism we noted earlier, anticipates this. “An independent gold standard,” he wrote (six years ago), “will probably come from neurogenetic or biobehavioral attributes.”[72] We will argue in Chapter 3 that there is now good reason to believe that we have isolated some systematic neurochemical mechanisms that cause a substantial proportion of PGs, as operationalized by DSM-IV, to become incapable of controlling their gambling (at least in the short term) without specific clinical assistance; that we have a broadly correct explanatory model of these mechanisms that enables us to say why it is gambling, and what it is about gambling, that relates these mechanisms to the familiar behavioral patterns traditionally associated with chronic PG; and that at least the broad outline of a standard pharmacological treatment (and perhaps much more than the mere broad outline) is coming into practical view. If we are right about this then, for the first time, it is appropriate to say that there is one kind of disordered gambling – the most severe kind – that we’re beginning to truly understand. If we are right about this, then at least that part of the more general DG phenomena should at some point soon be defined in DSM without equivocation. Furthermore, as we will argue in Chapter 4, social policy concerning this part of the phenomenon becomes relatively straightforward and shouldn’t even be controversial.

However, we think it is equally clear that the majority of cases where people regret their own gambling behavior are not cases of people in the grip of neurochemically diagnosable pathological gambling.[73] In this majority of cases of DG, non-addicted disordered gambling (NADG, which includes the non-addictive cases of what DSM-IV would classify as pathological gambling), all current evidence is compatible with the following conception. All people, as Ainslie argues and as described in sections 1.2 through 1.6, are naturally inclined to discount the future hyperbolically. This means they are all at risk of choosing SSRs over LLRs and then subsequently regretting it. Most people avoid this, most of the time, by finding personal rules that bundle LLRs with choices amongst SSRs. But most people also violate their own personal rules some of the time, and are not usually regarded as therefore having a problem. Indeed, Ainslie argues that healthy people are healthy partly because they break their personal rules from time to time; people who don’t do this are tyrants over themselves, and usually viewed by both others and by themselves as rigid and obsessive. But some classes of personal rules tend to get consistently and systematically broken. These fall into two main classes.

First, some people fail to develop effective personal rules in certain whole areas of behavior. Everyone knows otherwise competent people who ‘make the same mistakes over and over again’ (as we say) in their love lives, or who launch one implausible diet after another. It would be misleading – and almost certainly scientifically incorrect also – to think of these people as addicted to something. But if there are people who are like this with respect to gambling, and if they frequent gambling venues – as they might if that’s where their friends go, or if it’s the main centre of entertainment in town – then they might well meet the DSM-IV operationalization of pathological gamblers. By the criteria we will present in Chapter 4, they would not be addicts. But they would tend to consistently regret their gambling behavior as their preferences kept reversing following choices of SSRs (in their cases, bets).

Second, there may be sets of environmental circumstances in which almost any personal rule any normal adult might make would tend to be subverted. Most people will blurt out the carefully protected secret, for the momentary pleasure of another’s response, when drunk enough – and then subsequently regret it. Most people successfully observe personal rules against adulterous affairs, but most also know that these rules should be applied well before they find themselves alone with temptation in the jacuzzi because at that point it will be too difficult to stop. Casinos are often accused of being places deliberately contrived to make bundling of LLRs with SSRs difficult for most people. Free drinks, jangling coins, burbling and flashing slot machines, interference with the body’s natural clock for tracking time of day – all of these are said to be devices that interfere with personal rules. Combined in cunning overall venue designs, they may be sufficient to cause most people to gamble to the point of subsequent regret.

Notice one important difference between these two kinds of case. In the first sort of instance, we still have a distinctive type of person, a person who, as a result of some specific property of theirs, is disposed to gamble problematically. In the second kind of case, we might have problem gambling without there being any problem gamblers, that is, individuals who by dint of special characteristics of their own are more likely to fall into traps. In reality, what is probably most relevant to issues around regulatory policy as applied to casinos is interaction of the two phenomena, in which there are substantial numbers of people who are disposed by personal properties of their own in such a way that their ability to bundle tends to fall apart in casino environments arranged to put pressure on them.

People who are disposed to bundling breakdowns in gambling environments, but who lack the distinctive neurochemical properties we will describe in Chapter 4, will be referred to through the remainder of this report as ‘non-addicted disordered gamblers’ (NADGs). We do not know of any research that tells us how they are distributed between the problem and pathological gambling classes as DSM-IV criteria would sort this. The reason for this is that the scientific criteria picking out the addicted gamblers (AGs) are still too newly discovered to have featured in any large-scale statistical surveys of DGs. Not, therefore, knowing what proportion of pathological gamblers are AGs, we cannot know how many NADGS (if any) slide all the way to being pathological gamblers. However, in Chapter 4 we will see some reasons for worrying that if a problem gambler becomes a pathological gambler, she may be at risk of being turned, by physical changes in her brain, into an AG.

In any case, we are now in the position we promised to arrive at, of operationalizing problem (and, therefore, disordered) gambling. We will regard as a ‘problem gambler’ anyone who (i) gambles often enough to be consciously aware of trying to develop personal rules to limit their frequency of gambling or their average amount gambled per session or both, but who (ii) regards these rules as inadequately effective because they regret their own gambling behavior on a regular basis, and who (iii) does not meet the DSM-IV criteria for pathological gambling. Then a NADG is any problem gambler or pathological gambler who is not an addictive gambler. A disordered gambler is anyone who is either a NADG or an AG.

Notice that these operationalizations are not made in terms of either raw frequency of gambling or amounts gambled. There are almost certainly people who gamble a great deal, who have no inclination to gamble less frequently or with smaller stakes, and whose brains show no neurochemical traces of addiction. Such people are not, by the criteria given here, DGs. This reflects our BE perspective: the preferences of such people show no reversal, we have no basis for regarding their behavior as irrational or self-defeating, and by hypothesis they have no medical pathology.

This completes the statement of the theoretical framework on which the report is constructed. We use BE analysis to distinguish the overall target of interest, disordered gambling. Later, we will turn to neuroscience – in particular, to a new branch of neuroscience that comports with BE, neuroeconomics, in conjunction with neurochemistry – to isolate a specific subset of disordered gamblers who, we will maintain, warrant a distinctive policy response.

In light of this combination of perspectives, we should mention to the non-scientific reader that it is unusual. Behavioral economists often regard neuroscientists as methodological rivals. This stems from something we described in section 1.2, BE’s emphasis on molar accounts in preference to molecular ones. There are some scientists (‘reductionists’) who think that molecular accounts are the only serious kinds of accounts of phenomena there are. This is dogmatism, not an opinion well supported by the history of science, and behavioral economists are right to resist it. But it would be another kind of (in this case, ‘behaviorist’) dogma to insist that where behavioral patterns are concerned, the best explanation is always and necessarily to be given in molar terms. We think we should let the facts lead methodology rather than the other way around, and that it has just turned out to be the case that most disordered gambling patterns are molar-level phenomena, but that there are AGs whose behavior, in each individual case, is best explained by a recurrent variety of molecular explanation. Even here, our BE perspective still lurks necessarily but invisibly in the background: it is only because the AGs show similar molar behavioral patterns to the NADGS that all are grouped, and usefully so, together as DGs.

We should also mention that we are by no means the first scientific writers on DG to argue that it has distinct varieties. As noted in section 1.2, Alex Blaszczynski and Lia Nower[74] argue that there are three types of DGs. They agree with us that there are AGs and NADGs, but then also argue for a third group whose DG behavior is caused by neurochemically and behaviorally identifiable clinical depression. There are indeed data showing correlations between depression and heavy gambling. We have not followed Blaszczynski and Nower’s lead in breaking them out as a distinct category, for methodological reasons related to both scientific and clinical concerns. We are taking our topic to be DG per se. Where AGs are concerned, though a non-trvial proportion of them are multiply addicted, in Chapter 3 we will review evidence that gambling addiction is a sui generis phenomenon onto itself, that produces distinctive patterns of gambling. By contrast, the evidence to this point suggests to us that depression is simply one of the factors that makes some people vulnerable to exploitation as NADGs; depressed people have difficulty bundling. We expect that the best therapy for most such people is to have their depression treated directly. But as gamblers their patterns seem to resemble those of NADGs rather than AGs, so they do not seem to motivate distinctive gambling-related treatment or policy interventions. We of course remain open to evidence showing that there is either a behaviorally or a neurochemically distinct category of depressed gamblers (distinct, that is, from other DGs and from other depressed people).[75] If such evidence is forthcoming this will be reflected in future editions of this report.

We conclude this chapter with a summary of the structure of the report to follow. We have now sketched general issues in the scientific study of DG, and have explained the particular conceptual and methodological framework in which we approach it here. Next, in Chapter 2, we will review empirical studies of the behavioral patterns of DGs. As we will see, this work so far leaves considerable gaps in our knowledge. This, it will emerge, amounts mainly to deficits in our knowledge of DG involving NADGs, since we’ll show in Chapter 3 that where AGs are concerned the past two years have seen breakthroughs in the achievement of understanding, which point towards development of effective neuropharmacological therapies. Finally, in Chapter Four, we will summarize suggestions for treatment and policy approaches indicated by current scientific knowledge of DG, and also suggest prospective kinds of studies that, in light of current knowledge, look promising as routes to improved treatments and policies.

2. Behavioral Economics, Impulsivity, Addiction and Disordered Gambling

EXECUTIVE SUMMARY

Addicts to and abusers of a wide range of substances are known to be more impulsive than otherwise comparable individuals, in the sense that they have stronger preferences for smaller rewards delivered soon, compared to larger ones that arrive after some delay. Behavioral economists mark this distinction by saying that substance abusers and addicts discount ‘more steeply’ for delayed rewards.

Another form of discounting concerns not delay, but probability. A person is impulsive with respect to probabilities if their valuation of a reward declines relatively slowly as the likelihood of the reward arriving is reduced. This kind of discounting is less extensively studied than delay discounting, but is clearly of interest for the study of disordered gambling. The empirical studies of delay and probability discounting undertaken to date and including subjects with varying gambling severity (including problem gambling) do not establish clearly whether disordered gamblers are primarily impulsive about delays (like substance abusers) or whether they have distinctive responses to probabilities as well. Further experiments could settle this question relatively simply, and thereby provide a more precise behavioral characterization of disordered gamblers.

Behavioral measures of impulsivity are not the only approaches. That said, behavioral measures, based on experimental measurement, do correlate with gambling severity more successfully than more qualitative and psychiatric measures based on self-report and clinical appraisal. This does suggest that they relate more directly to the phenomena of disordered gambling than psychiatric categories.

Cognitive psychological assessments of decision-making, especially the Iowa Gambling Task, also separate disordered gamblers, who do poorly at them, from normal control subjects. It is not, though, completely clear what the Iowa Gambling Task measures, or how it relates to a number of other important tools for assessing decision-making. Further behavioral experiments could shed light on these questions relatively simply, and thereby enable two relatively isolated bodies of work concerning disordered gambling to be connected. Work in neuroeconomics (surveyed in the following chapter) also suggests ways of connecting the two bodies of work.

Finally, we know that various factors can temporarily influence decision-making, some by changing discount rates. Some individuals may be more vulnerable to these influences, and some environments (such as casinos) might be unusually endowed with them. Both possibilities are relevant to debate over policy regarding gambling and disordered gambling.

2.1. Impulsivity and addiction

In this section we review a range of work showing that behavioral economics has something important to say about addiction in general, especially with reference to the paradigmatic substance addictions including alcohol and nicotine. We go about things this way because it will be useful to establish the credentials of behavioral economics before turning to problem gambling specifically, and also because doing this will establish a framework for discussing some of the peculiarities of gambling compared to substance abuse.

The first sort of work we’ll review concerns empirical findings regarding impulsivity and substance abuse. Impulsivity here is to be understood in the narrower behavioural sense, as relatively greater ‘steepness’ of discounting for delay (see section 1-4 above). There are links between this specific sense of impulsiveness and the more general sense used in psychiatry and personality psychology, but these approaches are discussed separately in connection with work specifically on gambling (section 2-3 below). Some of what is said in the first few paragraphs of this section repeats points made in earlier sections, including parts of section 1-4 and section 1-5. They are included here in the interests of making the present discussion as clear as possible.

When behavioral economists refer to the ‘steepness’ of discounting, they have in mind a graphical property of a discount function, itself constructed (for an individual agent) by fitting a curve to a series of indifference points determined by offering a series of choices between pairs of rewards of varying size and delay. The previous sentence contains a lot of jargon, and the next few paragraphs attempt to set out the key ideas referred to, and to outline the experimental procedures used, at least in the case of human subjects.

Anyone concerned to maximise her rewards would rather have R2000 now than R1000 now. Similarly, anyone concerned to maximise her rewards would rather have R1000 now than R1000 after a delay of 1000 days. How about the choice between R1000 now, and R2000 after 1000 days? Most readers of this report should realise that doubling in value in under three years is a very impressive return, one you’d be very unlikely to beat if you took the R1000 and invested it. But most readers also realise that there are some people who would take the R1000 now in this case, and some who would wait for the R2000. That choice splits people into two groups – some preferring the smaller sooner reward, and some the larger later one. What would you do, though, if you wanted a more precise way of distinguishing people with respect to their preferences of this sort? (That is, their preferences as relating to varying delays.)

Suppose that you took a single person, and offered them a whole series of choices, where one option was always R1000 now, and the other was always some amount after 1000 days. Assuming that our subject initially preferred to wait to receive R2000, we would reduce the delayed amount bit by bit, say by R25 every time. At some point we’d find that the subject switched to preferring the smaller sooner reward.[76] In fact, we’d expect this to happen some time before the later reward comes down to R1000. In between the range of amounts that our subject would not consider worth waiting 1000 days for in preference to R1000 now, and the range that she would consider worth waiting for, is a small range (in theory a single value) where she prefers neither option over the other. In these cases economists would call her ‘indifferent’ between the options, and the pair of options (amounts and delays) together determine an ‘indifference point’ that we can plot on a chart of delay versus amount.

That point, to repeat the key idea here, would (in this example) represent an amount of money at a delay of 1000 days that our subject neither preferred nor found inferior to R1000 now. Suppose that the delayed amount in question was R1200, then our point would look like this (with amount on the vertical axis, delay in days on the horizontal one):

We could repeat the process for a number of different delays (1 day, 10 days, 100 days…) and plot the indifference point in each case on the same graph, as long as in each case the ‘reference amount’ was R1000 now. With several such points, a pattern or definite line may start to get suggested. If the set of indifference points stand in a regular relationship of some sort with each other, then some mathematical equation could describe a line passing through (or very near) those points. Finding such a line is called ‘fitting a curve’ to a set of points and the resulting equation, in this sort of case, is called a discount function.

The chart below shows value on the vertical axis, and increasing time on the horizontal axis. The lines represent discount curves. What this means is that each point on a curve represents the value of a reward at a certain delay, and the collection of points represents the way in which the value of a reward declines as the delay before its delivery increases. The topmost two lines are drawn using an ‘exponential’ discount curve, the other two are drawn using ‘hyperbolic’ discount curves. The two hyperbolic curves are drawn using the same equation, but with a different value for the discount rate in each case. When we say that some person, or type of person, discounts ‘more steeply’ than another, we mean that their discount rate is higher, and hence that the value of a reward declines more rapidly as delay to delivery increases. The ‘steepness’ is the steepness of the downward slope from the point of immediate delivery.

In the curve fitting process, the vast majority of the experimental data concerning humans and animals finds that delay discounting fits a hyperbolic curve better than an exponential one. (See section 1-5 above) The steepness of an individual discount curve or function is represented by the value of a single parameter in the equation. An example of a simple hyperbolic formula fitting most of the data very well is equation (3) discussed in section 1-4 above:[77]

(3)

In this formula, recall, v_i, A_i, and D_i represent the present value of a delayed reward, the amount of a delayed reward, and the delay of the reward, respectively, and k is a constant representing degree of impatience. In the graph above the steeper of the two hyperbolic curves has a ‘k – value’ of 1.5, the shallower one a k – value of 0.75. Most claims about the relative steepness of discounting in a comparison of two individuals or groups are claims about different values or average values of the k parameter manipulated in the curve-fitting process.

The results we’ll report almost all take the form of associations, or correlations, to the effect that in some or other population of substance abusers or addicts, delay discounting was found to be steeper than for control subjects, or steeper than some norm or other contrast class. That is, that addicts and substance abusers were less tolerant of waiting, or were more impulsive, in preferring a smaller reward sooner than a larger later one to a greater degree, than non-addicts and abusers.

Such correlations do not by themselves establish causation. That is, they don’t tell us by themselves whether the individuals in question became abusers because they were disposed to be more impulsive beforehand, or whether use (or abuse) of a substance had made them more impatient. It could well be that both are true – that some people are prone to immoderate substance use because of being a bit less patient, and that this consumption makes them even more impulsive, so that they become ‘proper addicts’ with time. (For a more thorough discussion of the difference between real addicts and other sorts of disordered consumption see Chapter 3 below.)

To shed light on causation we’d need more ‘longitudinal’ data than we have.[78] That is, we’d need measurements of the discount rates of whole cohorts of people over a number of years from when they were relatively young, so that follow up study could establish the extent to which this or that degree of impulsiveness in youth was a predictor of a self destructive pattern of behavior, including disordered gambling, in later life. Fortunately there are some suggestive studies with a longitudinal component, indicating that the effort could be worth making. In one such study impulsivity at age 12-14 was found to be a significant predictor of involvement in problem gambling at age 17.[79] That is, younger boys who disproportionately (compared to their peers) preferred smaller sooner rewards to larger later ones, were more likely than their peers to be disordered gamblers a few years later. A related finding is that low impulsivity is a good predictor of academic success.[80] This isn’t surprising, since the rewards of academic study are delayed, so impulsivity can be expected to manifest in the costs of present effort not seeming worth the distant rewards.

These last two results are also of interest since it suggests that a ‘common cause’ may explain both some of the non-gambling problems in the lives of disordered gamblers, and help explain their disordered gambling, rather than most of the problems being the consequence of the gambling. That is, in at least some subjects it seems plausible to suppose that their failure to complete post-secondary qualifications, or keep up payments on investments, or otherwise follow through on longer-run plans might be the product of the same underlying impulsivity as their disordered gambling, rather than being casualties of the need to feed their condition. We return to these and related questions in our discussion of policy implications in chapter 4.

Other useful work compares the discount rates of current, former, and never consumers of some addictive substance, and finds, for example, that matched ex-smokers and never smokers both discount less steeply than current smokers, suggesting either that smokers with low discount rates are more likely to become ex-smokers or that stopping smoking is associated with a change in discounting behavior. Note that this work doesn’t directly tell us what causes what, since the data in this study is not truly longitudinal and suggests only that both smoking status and discounting change at more or less the same time. It does, though, point to the possibility of changes, including reversible changes, in discount rates over extended periods,[81] and suggests a search for the determinants of such changes.

Finally, in the laboratory, early impulsivity is a predictor of the extent of cocaine self-administration in rats.[82] Manipulation of reward-relevant chemical activity in the brain is followed by changes in the steepness of discounting in rats, providing further encouragement to those who seek a drug-based therapy for at least some problem gambling.[83] (See chapter 3 below for much more on this topic.)

Although associations shed little direct light on causation, the ones we shortly review are informative and useful. This is for several reasons. First, that discounting tasks can find discount functions for substance abusing subjects at all is good confirmation of the applicability of an economic approach to studying addiction. Unless such people were able to make reasonably regular comparisons of the relative values of rewards of different magnitudes and at different times, it would not be possible to measure their discount rates with any confidence. Second, the fact that addicts tend to discount more steeply is reason to suppose that the behavioral notion of impulsivity, as measured by relative discount rates, is a good tool for sorting disordered consumers from normal ones, and a good basis for trying to work out what it is that makes the difference. After reviewing some of the evidence that behavioral notions of impulsivity apply to addictions, we note some pieces of evidence bearing on the interpretation of the associations we report.

One study compared regular smokers with non-smokers on a variety of clinical and personality measures of impulsivity, as well as three distinct behavioral tasks, assessing impulsivity as a function of delay, as a function of reduced probability and as a function of effort required to obtain a reward. It found that smokers had higher impulsivity scores than most controls on the personality questionnaires, and that they were more impulsive when it came to delays, but not on the other behavioral tasks.[84]

A complementary study of current smokers, former smokers, and people who had never smoked at all found that current smokers discounted the value of delayed money more steeply than both never smokers and ex smokers. Smokers also discounted the value of delayed cigarettes more steeply than they did money. Never and former smokers were not significantly different from each other. This work suggests, but does not show, that smoking is associated with a temporary and reversible change in discounting.[85] The result that current smokers discount more steeply than non-smokers has been confirmed in other research.[86]

In one of the first studies of this sort, the discount rates of a mixed population of inpatient substance abusers was compared to the discount rates of those who worked at the unit, and it was found that the substance abusers discounted delayed rewards more steeply.[87]

Considering alcohol specifically, it has been found that college students who were heavy and problem drinkers discounted for delay more steeply than light drinkers.[88] A further study comparing currently drinking alcoholics, abstaining alcoholics, and nonalcoholic control subjects found that drinking alcoholics discounted more steeply than abstainers, who in turn discounted more steeply than non-alcoholics. Alcoholic subjects discounted alcohol more steeply than money.[89] Finally, it was found in a further study that discount rates were associated (in a further sample of college students) with a number of substance consumption variables, including age of first use of alcohol, tobacco and marijuana, number of times incapacitated by alcohol, and total number of illegal drugs used.[90]

Turning to drugs other than alcohol, in a pair of careful studies the discount rates of opioid dependent subjects recruited from a treatment programme were compared to those of control subjects matched for age, education, gender, IQ and income, and drawn from the same community. Opioid dependent subjects discounted the value of delayed hypothetical money substantially more steeply than did controls. In addition, they discounted the value of delayed hypothetical heroin even more steeply than they did money.[91] These studies, and the second of the studies on discounting and alcohol described above, both show that there are subject specific variations in the rate at which different sorts of reward are discounted, and that discount rates can be significantly higher for rewards of abuse.

A further study using real (i.e. non-hypothetical) money also found that heroin-using subjects discounted dramatically more steeply than non-heroin using controls, who had again been matched carefully in other respects,[92] and that the same went for cocaine-abusers.[93] A complementary study found that among a population of heroin users, all of whom discounted very steeply, those who discounted most steeply were also those most likely (on the basis of self-report) to share a needle in a case where sufficient sterile ones were not available.[94] It has also been found that mild deprivation from opioid drugs leads to steeper discounting of delayed rewards in already dependent subjects. [95]

This collection of results certainly encourages the view that steepness of delay discounting is an important factor distinguishing a variety of substance abusers from non-abusing subjects. At the same time this therefore suggests that the general behavioral economic framework for the study of decisions is appropriate for studying the decisions of addicts, or at least substance addicts. What about disordered gamblers?

2.2. Discounting: delay and probability

The great majority of the empirical work concerning discounting in human subjects, including work on addictions, focuses, as was explained in section (1.4) above, on discounting as a function of delay. Being removed in time, though, is not the only way in which a reward of a certain magnitude can be separated from consumption. Nor, for anyone interested in gambling, is it the most interesting. The other main way in which the value of a reward of a certain magnitude can be manipulated is by varying the probability that it will arrive.[96] That is, just as the present value of R100 after a delay of 1 month is, for most agents, lower than that of R100 right now, so the value of a 50/50 chance of getting R100 is, again for most, less than that of a guaranteed R100. Since gambling essentially involves risk and uncertainty, it is plausible to suppose that gamblers, and especially disordered gamblers, might be distinguished specifically by different responses to risk than control subjects. Disordered gamblers, rather than merely having different responses to delay in common with other addicts, might be specifically unusual in their responses to risk, and may even be no different to non-disordered gamblers in their responses to delay.

Whereas, in the case of delay, the discounting of an individual subject is assessed by asking a series of questions about which of two rewards, varying in size and delay, she prefers (e.g. would you rather have R100 now, or R120 in a week?), in the case of probability, the questions relate pairs of amounts and chances (e.g. would you rather have R100 now, or a 50/50 chance of R250?). As in the case of delay discounting, a discount function for an individual agent is constructed by fitting a curve to a series of indifference points determined by offering a series of choices between pairs of rewards of varying size and likelihood. And, as in the case of delay discounting, the curve that best fits most of the experimental data on the patterns of choices of real agents, animal or human, is a hyperbolic one. The steepness of this curve is a measure of risk sensitivity, just as the steepness of the delay discount curve is a measure of impulsiveness.

Part of the reason that we know so little is that delay discounting promises a general explanation of self-defeating behaviour, because (as explained in Chapter 1) the fact of hyperbolic discounting for delay predicts that agents’ priorities will not be stable over time, and hence that people will make plans only to break them, forswear temptations only to give into them, and regret these patterns of behaviour. Not only that, some theorists have held that probability discounting is a derivative form of delay discounting, and hence of limited independent interest.

Part of why they thought this was that it seemed likely to them that the two sorts of discounting distinguished here are manifestations of a single, fundamental, discounting process – either concerning chances, delays or something not quite either. Among the reasons for thinking this are that any delay to delivery of a reward involves an element of risk – especially risk that you won’t be there still to collect it, that some misadventure including the efforts of a competitor will take it away between now and scheduled delivery. Thus delay and uncertainty are connected in everyday contexts.[97] Not only that, some argue that the fact that the same (hyperbolic) form of mathematical function describes both sorts of discounting data should lead us to think that there is a single underlying process. There is also some evidence that in individual subjects the rate of discounting for delay and that for probability are correlated.[98] Unfortunately, though, there have been relatively few studies in which the delay and probability discounting of subjects have both been measured and compared, so we know too little about the extent to which they typically relate to one another. Given the importance of risk to gambling, considerably more study on probability discounting (in populations with mixed gambling severity) is required.

Against the suggestions that the two kinds of discounting might be manifestations of a single process, there are several ways in which the two apparently come apart (‘dissociate’, as scientists say). One study asked subjects about the relative preferability of pairs of immediate and delayed rewards in each of two different currencies. One of the currencies (the Polish Zloty) was the money of a country undergoing rapid inflation, while the other (the American Dollar) was not. The experimenters found that subjects treated the chances of immediate rewards of the same magnitude equivalently irrespective of currency. (That is, their discounting for reduced likelihood was the same irrespective of currency.) However, when considering guaranteed delayed rewards, they discounted the rapidly inflating currency more steeply.[99] This selective effect on one kind of discounting tells against the view that there is a single discounting process. In a complementary study it was found that subjects’ cultural background could make a selective difference to one or the other kind of discounting.[100] In other work it was found that while smokers discounted more steeply for delayed rewards than non-smokers, and were less sensitive to reduced probabilities, the difference between smokers and non-smokers was much more pronounced in the case of delay discounting.[101]

More importantly for present purposes, there is evidence that some gamblers differ from non-gamblers (understood as people with little or no experience of wagering or otherwise paying to participate in games of chance) in their probability discounting, but not their delay discounting. In one study at least, it has been found that subjects that gambled (and who scored in the ‘problem’ range on a standard gambling severity measure) were less sensitive to reduced probability than non-gamblers, but did not discount delayed rewards more steeply than non-gamblers.[102] This result is apparently in conflict with the finding of another study to the effect that pathological gamblers did discount more steeply for delay than non-problem controls (and, consonant with prior work on abusers of more than one substance, that pathological gamblers with substance abuse problems discounted yet more steeply for delay).[103] This doesn’t suggest that gambling is a substance, but rather that disordered gambling seems to be behaviorally similar in a key way to substance addictions – having more than one consumption problem, even if only one is to a substance, seems to be associated with less severe discounting than when you have two of them.

Direct comparison of the two studies is difficult given that the gamblers in the second study (finding an effect in the case of delay) had much more severe gambling problems (with SOGS scores of around 12-13, compared to 6-7) than those in the first study (finding an effect for probability but not delay). Subjects in the second study were also older (with a mean age of 44, compared to 19.6) and with an average of 8 years of gambling problems each. The explanation for this difference is clear enough: one study used students as subjects, the other recruited subjects from a problem gambling treatment facility. This contrast and the possible conflicting results of the two studies, illustrates the need for larger studies in which as far as possible matched subject populations nonetheless differ significantly in gambling severity. Further explanation for the discrepancy is given in Chapter 3 below, in the discussion of neuroeconomic findings about disordered gambling.

Not only are delay and probability dissociable, but most real choices, pretty obviously, involve elements of both kinds of separation. That is, most real choices concern rewards that won’t arrive immediately, and which are also not absolutely guaranteed to arrive. In particular, when people gamble, they explicitly make such choices, declaring by betting that they’d rather have the chance of having whatever the bet could win, when the outcome is decided or determined, than have some of the money they already have, and that they pay to place the bet. Since gambling involves uncertainty, it would be useful to know a lot more about probability discounting, and especially population differences in its expression, than we presently do.

Among the things it would be useful to know is how exactly non-disordered gamblers, and disordered gamblers of differing severity of problem, differ. Perhaps, for example, moderately disordered gamblers are more like substance abusers in being relatively more sensitive to delay, but don’t differ in their probability discounting, whereas severely disordered gamblers are both impatient and insensitive to risk. If there are indeed subjects (and severely disordered gamblers are a likely example of this) who are impulsive and risk-insensitive, then we need to conduct research on how particular situations, and particular series of choices and results of choices, might be especially destructive for such people. Casinos, for example, may turn out to be exquisitely tailored traps for such people. Or it may turn out that sufficiently extreme discounting makes the whole world a trap.

Before saying more about what we could find out about disordered gambling with more and better research on discounting, though, a look at some of the existing results concerning disordered gambling is in order.

2.3. Impulsivity and Disordered Gamblers

The discussion of behavioural economics, and its results with reference to addicts of various sorts, over the preceding pages might have given the impression that there is a single science of behaviour, where even though there is not full agreement about how to explain the data, all practitioners are working with one set of empirical instruments for gathering the data itself. This impression would be misleading. Alongside, and for historical reasons significantly independent of, the tradition of behavioural research are other streams of enquiry with their own methods and tools. Among the most important of these for the purposes of the study of disordered gambling are cognitive psychology and psychiatry.

Behavioural research, as we have seen, emphasises the economic dimension of choice, considering the patterns of costs and benefits of actions, the budget constraints of the agents that make them, and so forth. The tools of behavioural research are experimental protocols for extracting the data required to construct mathematical representations of aspects of the choice making of an individual agent. An example of this is the series of options presented so as to determine a set of ‘indifference points’ (see section 1-5 and 2-1 above), through which a curve representing a discount function can be drawn. To recap, an indifference point represents a pair of options (values and delays, or values and probabilities) of which a given agent finds neither preferable to the other. A set of such points together determine a ‘discount curve’ which is a profile of one aspect of an agent’s choice behaviour, for example sensitivity to delay (see the graph in section 2-1 above). Differences between individuals and groups can sometimes be described as differences in the properties of these curves. We saw examples of this above where substance addicts of various sorts were found to discount delays more steeply than non-addicts.

Cognitive psychology for its part understands decision-making as a process primarily focused on the acquisition and manipulation of knowledge and the control of action, largely independently of questions about what makes some outcomes desirable, or what motivates agents at all. The historical reasons for the relative neglect of motivation by cognitive psychology aren’t relevant to this survey, but the fact of it helps explain why cognitive psychologists have studied various kinds of decision-making, including pathological patterns of choosing such as addictions, in different ways. Cognitive psychologists tend to be more concerned than behaviourists about the architecture of the processing systems (the brain and its sub-systems) involved in decision, and in how the inputs to the decision process are represented and manipulated. The new field of neuroeconomics, combining tools and techniques from behavioural economics and cognitive neuroscience, is a particularly clear indication that long-overdue rapprochement is underway. This field is discussed in Chapter 3 below.

Psychiatry, finally, has considered decision, especially pathological decisions, in a way intended to be continuous with the rest of medical science. Psychiatry is often defined as a branch of medicine, and unlike behavioural economics and cognitive psychology, its subject matter is in turn often defined in terms of mental illness, and its treatment and prevention. Much psychiatry as relevant to disordered gambling draws on the field of personality psychology, attempting to develop frameworks for classifying all humans by analysing personality into some or other set of dimensions (such as ‘sociability’).[104] Most theories of personality attempt to classify subjects’ impulsivity, but there is considerable variation in whether impulsivity is a dimension itself (so a basic fact about a person is his or her level of impulsivity) or whether impulsivity is a name for a more complex trait depending on how a subject is ranked on two or more other dimensions. The lack of consensus within psychiatry and personality psychology about the measures of impulsivity suggests that all is not well with the measures and the theories that inform them. This is especially so given the precision and power of behavioural approaches, meaning that there is an attractive alternative. Some comparative research suggests that disordered gamblers are classified correctly with psychiatric impulsivity measures, but with less effective discrimination than behavioural ones.[105] Given this, it is not surprising that various studies report that disordered gamblers are more impulsive by the lights of personality psychology measures of impulsivity,[106] or that high scores on these measures at one age is a predictor of disordered gambling later on.[107]

All three of these main approaches (behavioural economics, cognitive psychology, psychiatry) have an established interest in disordered gambling, and all have something to say about how disordered gamblers differ from other people, and about what might make the difference. There is, though, no simple fact of the matter about how the findings of the three approaches introduced here relate to one another. In some cases, as we explain, they say complementary things, in others they seem to conflict, while in some it is hard to tell whether they are talking about the same thing at all. This is because the operational content of their different tools and notions of impulsivity differ in various ways, and there is very little experimental work aimed at determining how the different measures relate to one another in a single population.

In case this fragmented image of the sciences of human choice seems far-fetched, bear in mind two things. First, the phenomena are very rich and complex. They comprise, after all, human patterns of choice in varying environments, and the responses of other humans to the choices made by one another. Second, divisions within science can persist for extended periods if supported by divisions in how practitioners are trained. The position of psychiatry within the medical establishment, while other approaches are largely outside it, is one example of this. The further division within scientific psychology between cognitively oriented researchers and behaviourally minded ones is another. As we’ve mentioned before, and will soon demonstrate at length, the rising science of neuroeconomics is an illustration that a new generation of practitioners who bridge cognitive and behavioural approaches are on the rise.

One widely used experimental task in the cognitive science literature is the ‘Iowa Gambling Task’ (IGT).[108] Since there is a body of work finding that performance on this task separates addicts of various kinds, including gambling addicts, into one pile and normal subjects into another, there is reason both to want to know how and why it works, and also to know how it might relate to the discounting procedures that we’ve already seen sorting the subjects into the right piles.

In the Iowa Gambling Task[109] a subject is asked to choose a series of cards from four decks. The cards are face down, so subjects are uncertain about the consequences of their choices until each one is made, and the selected card turned over. The subject can chose a card from any deck at any time, and are told to try to maximise their earnings over the course of the task. Each card leads to some combination of winnings and penalties, for example winning R250, and paying R50. Unknown to the subjects, two of the four decks are ‘bad’ decks in which large wins are on average more than cancelled out by regular penalties, so that consistently choosing from those decks is guaranteed to lose money. The two other decks are ‘good’ ones where the balance of reward and penalties favours the subject who persists with those decks, even though the rewards associated with individual cards are generally smaller. Although subjects aren’t told this, a standard implementation of the test involves turning 100 cards. The performance of a subject is determined by counting the changing proportion of choices from good and bad decks made as the subject is exposed to more and more cases of the effects of choosing from each deck. Most subjects start allocating their choices roughly equally, and a ‘normal’ or ‘healthy’ subject should learn to make fewer and fewer selections from the bad decks as they proceed through the task.

Some famous early research using the task found that patients with damage to the ventro-medial prefrontal cortex performed much worse than undamaged controls.[110] In light of other results with the same subjects and certain other considerations, it was proposed that normal subjects successfully formed unconscious[111] ‘somatic markers’ or emotional associations with actions and the consequences of choices and that disruption of this process could lead to impaired decision making.

From the perspective of behavioural economics, the Iowa Gambling Task is, it must be said, a rather peculiar research instrument. There is no good reason to have four decks rather than two. Since complex preference rankings can be broken down into sets of preference relations between pairs of options, behavioral experiments have largely focussed on tasks and situations with only pairs, to keep things no more complicated than they need to be.

(Indeed a variation of the task developed for use with children dispenses with half of the decks for the very reason that the extra two aren’t required, or even distinguished in scoring the performance of subjects.[112]) There’s also enough that the subjects don’t know to confound interpretation of some of their behaviour. For a start they don’t know for sure what the gains from any deck will be, even though those are fixed for each deck. This means that, unlike a discounting task of the sort described in the previous section, choices don’t directly convey information about the individual subject’s response to risk, even though the risk of gains and punishment is clearly important in the task. As subjects proceed with the task they do get exposed to information about the frequency and magnitude of penalties in each deck, but don’t ever know those contingencies in advance of their choices either. Again, interpretation of the results cannot clearly say whether the behaviour of subjects is telling us something about their preferences, for example that they favour the bad decks just as they favour all risky options, or their ignorance, in case they have failed to come to understand that those decks are bad at all. (In fact, there is a variation on the Iowa task that makes the likelihoods known up front for that reason.[113]) No matter what they know, their choices involve a considerable mixture of delayed and probabilistic rewards and penalties. That is, if they assume (probably tacitly) that the remainder of each deck is pretty much like what they have experienced it to be, then their view of that deck will include a mixture of frequencies and probabilities of gains and losses of various sizes in a way that leaves it quite unclear how their choices relate to their individual discount curves for delay and probability. The number of choices made by subjects is also too small to disentangle these components. In a behavioural economics discounting task, a human subject will typically have to answer around 70-90 questions in order to locate enough indifference points to determine a discount curve with reasonable confidence. To measure both delay and probability discounting for a single subject accurately, then, would take nearly twice as many questions as one run of the Iowa Gambling Task. Yet discounting is surely a factor in performance, for reasons we return to shortly.

More recent work has suggested that some of what the Iowa Gambling Task has purportedly measured may not be what it was intended to measure. A team of researchers developed a variant of the task, called the Soochow Gambling Task, in which the ‘good’ and ‘bad’ decks had the same average net earnings as in the Iowa Task, but where the frequency of rewards was different. This research found that normal subjects (that is, subjects who were not brain-damaged in areas known to correlate with bad Iowa Task performance) and who did ‘well’ at the Iowa Task, made mostly bad choices in the Soochow Task when the frequency of reward from the bad decks was high. On the basis of this work some of the early research on the Iowa Task has been reinterpreted, suggesting that frequency of reward is a much more important determinant of how normal subjects behave.[114]

Having noted these concerns and difficulties, it should also be said that some significant results have been found with the Iowa Gambling Task as well. Although it has not been extensively used as a tool for the study of addiction, it has been found that subjects with substance dependence (the dependencies in the study being referred to here were of various sorts including alcohol, crack-cocaine and methamphetamine) have been found to perform poorly at the IGT compared to normal controls.[115] The same has been found for patients on a methadone maintenance programme attempting to treat heroin addiction,[116] and for long-term heavy users of marijuana.[117] This suggests that, at least some of the time, the Iowa Gambling Task sorts relatively impulsive people into one pile and the others into another pile in the same way, roughly, as measures of discounting.[118]

A few other facts about the performance of different types of subject on the Iowa Task can be noted fairly quickly. In contrast to the results described above (in section 2-1) showing that cigarette smoking was clearly associated with steeper delay discounting, it seems as though worsened (or in any way changed) IGT performance is not associated with smoking.[119] This result tentatively suggests that Iowa performance is to some extent independent of discounting. Nonetheless in a population of methadone-maintenance subjects, smoking was associated with further worsened performance at the IGT than was the case for non-smoking patients.[120] This result tentatively suggests that Iowa Task performance, like discount rates, is cumulatively affected by multiple consumption problems, even if one of them is to addiction to a substance, and the other disordered gambling.

A complication, though, is that the association between education and IGT performance seems to be roughly the opposite of the relationship in the case of delay discounting. That is, having spent more years in education is associated with less steep discounting for delay,[121] but with worse performance at the IGT,[122] even though in general IGT performance improves with age,[123] even if some older subjects undergo a decline.[124] We also know that administration of testosterone leads to worsened performance on the IGT.[125]

More significantly for our purposes, there is some work showing that disordered gamblers do poorly at the IGT compared to subjects from control groups. In one of these studies the disordered gamblers were found to be unimpaired at a different task intended to assess ‘executive cognition’ or the higher order cognitive functioning needed when the ‘rules of the game’ change and old responses have to be suppressed in favour of an alternative behaviour. (For example, remembering to call football ‘soccer’ while talking to a North American.) That the disordered gamblers were not impaired at executive cognition suggests that their poor performance on the Iowa Task was unrelated to executive cognition.[126] The other study found that the ‘profile’ of disordered gamblers (in a comparison of subjects with various sorts of decision making difficulties) was closer to that of alcoholics than sufferers of Tourette’s syndrome.[127] It has also been found that problem gamblers perform worse than normal subjects at a task structurally similar to the IGT in key respects (the ‘game of dice task’ developed in Germany).[128] In this last study, though, unlike the first one mentioned in this paragraph, the frequency of bad choices made by disordered gamblers was found to be correlated with weaknesses in executive cognition.

As with the case of whether disordered gamblers differ from non-disordered gamblers in their delay discounting or their probability discounting (see section 2-2 above) the results just described need not be in conflict, even though they do initially seem to that way. There could, for instance, be different sorts of disordered gambler, some of whose problems are more related to executive cognition, and some of whose problems are not so related. In that case, the apparent conflict in results of the two studies could be because of neglected differences between their populations of disordered gamblers. Real differences between the IGT and the Game of Dice Task, alternatively, may mean that the apparent conflict disappears when experimental work on disordered gamblers using both tasks takes place, and we find that, and in what precise ways, they don’t measure the same thing.

Clearly enough, as already noted, some fraction of the responses of subjects when engaged in the IGT will be related to their discount rates. We don’t know how great a fraction partly because of the near-total dearth of experimental work assessing the same subjects in both ways, but we do know, from separate studies, that as well as performing poorly on the IGT, disordered gamblers are distinguished by their discounting. Some of these results have been mentioned already (in section 2-2 above). Here is a quick recap, along with some further results:

· It has been found that problem gamblers discount delayed rewards more steeply than non-gamblers.[129]^,[130],[131]

· It has also been found that gamblers (not recruited as problem gamblers, but with SOGS scores in the problem gambling range) did not discount delayed rewards more steeply than non problem gambling control subjects, but were less sensitive to risk measured in a probability discounting task.[132]

· A measure of the attitude of risk, or perhaps the value of the future, and hence a measure of delay discounting, of a person is their willingness to engage in risky sexual activity. It has been found that substance abusers with gambling problems are more likely to engage in risky sexual behaviours than abusers without gambling problems.[133]

· Cigarette smoking disordered gamblers have been found to have more severe gambling problems than non-smoking disordered gamblers, further confirming the association between smoking and disordered decision-making.[134]

· It has also been found that disordered gambling substance abusers had dramatically higher discount rates (or steeper discount curves) than non-problem gambling substance abusers.[135] In connection with this, it is worth noting that the prevalence of disordered gambling is much higher among substance abusers than the population at large.[136]

Although there is some possible conflict in these results, they are generally encouraging for the view that behavioural economics provides a suitable framework for making sense of disordered gambling. The very same research tools that distinguish drug abusers from others seem to distinguish disordered gamblers from subjects who are not disordered gamblers. This in turn suggests that disordered gambling is an example of a general kind of impulsivity we already know something about.

Further behavioural study of disordered gamblers is needed, though. As we’ve emphasised more than once, hardly any studies include gamblers of significantly varying severity of disordered gambling, and hardly any assess their subjects for both delay and probability discounting. No study has ever included assessment on the IGT as well as at behavioural discounting procedures to begin shedding light on the relations between these instruments, both of which seem able to distinguish disordered gamblers from others.

Another crucial kind of research of which there is far too little concerns factors that temporarily influence discounting or choice behaviour. Almost all of the work above separates one group from another (heroin abusers from people who don’t use heroin, or disordered gamblers from people whose gambling is unproblematic, etc.) measuring their performance at the experimental task, whether a discounting task or the Iowa, once off for the purposes of the study. We know, though, that factors of various sorts can induce temporary changes in how people make decisions. These factors include priming effects from seeing or thinking about something before making a choice, chemical effects arising from consuming various kinds of substance including alcohol, as well as factors like pain.

Starting with alcohol, it has been found that acute administration of alcohol leads to increased impulsivity in rats.[137] A study of real electronic gaming machine (EGM) players in a real venue found that, according to the EGM players, alcohol consumption was a major contributor to diminished control over whether to continue playing or not.[138] Using the questionnaire based instruments of personality psychology to measure impulsivity alcoholics, as well as the children of alcoholics, are more impulsive than the general population.[139]

A different study in an experimental setting used real gaming machines, and measured the extent to which subjects persisted in the face of losses. Half of the subjects drank an alcoholic beverage before participating (equivalent to 30 grams of alcohol) while half drank a non-alcoholic placebo beverage. It was found that subjects who had consumed alcohol were twice as persistent against losses as members of the placebo group. Half of the subjects in the alcohol group lost their entire original cash stake, compared to only 15% of the placebo group subjects.[140]

One question raised by this striking result is whether alcohol consumption leads to modification in discounting. In one very careful study all subjects were first assessed for delay and probability discounting, and then administered a beverage which in half of the cases was alcoholic while in other half it was a placebo, before being assessed again for delay and probability discounting. This study found that alcohol consumption had no effect at all on discount rates in the case of either delay or probability.[141] This result is, at least initially, very surprising, perhaps especially so given that alcoholics are known to be more impulsive in the specifically behavioural sense (see section 2-1 above).

It is possible, though, that the absence of a change in discounting in this case is explicable by reference to peculiarities of the procedures used to assess discounting. In them, subjects are repeatedly asked which of two options they prefer, but do not experience wins and losses during the procedure. Sometimes, at the end of the procedure, one or more of the choices are randomly selected and the subject paid whatever they said they preferred in the selected case. (This is done to encourage subjects to make the choices seriously, since any given choice could turn out to be ‘real’.) In cases where the subject chooses a chancy option, they might well fail to earn anything, but their loss will take place only after the discounting procedure has been concluded, and hence have no effect on the procedure or its result.

More simply, if there are no losses in a standard behavioural discounting task, then one can hardly expect to see an effect of greater perseverance in the face of losses on the part of subjects who have consumed alcohol. This suggestion is supported by some of what is already known about the cognitive effects of alcohol consumption. Psychologists maintain that the effects of alcohol on social behaviours (at least) are more pronounced in situations containing both actual rewards and actual penalties,[142] and especially where more substantial rewards and penalties were incurred sometimes.[143] It is therefore clear that a comparative assessment of discounting behaviour under the influence of alcohol should be made in a situation with real consequences of choices, including choices leading to negative outcomes. Such an experiment would be easy enough to construct, either by determining the outcomes of at least some probabilistic choices in a discounting task right away, or by alternating choices in a discounting task with choices in some other task leading to losses and gains.

The findings that alcohol consumption leads to impaired decision making at gambling should be read in the light of additional work showing that gambling itself can lead to greater voluntary alcohol consumption (compared to a control activity of watching an action movie while alcohol was freely available), and that gamblers who were also drinking were more likely to undergo negative mood changes, even when gambling losses were factored in, than people who gambled but drank non-alcoholic beverages. That is, not only do people who drink and gamble make bad decisions, they are made less happy at the time they are gambling by the fact that they have been drinking.[144]

Leaving the case of alcohol aside, there are features of the external environment known to impact on decision. It has been known for some time that simply seeing a smiling face, even if so briefly that it is not consciously noticed or remembered, leads to greater reported approval of what is observed shortly afterwards.[145] Some very striking experiments in the early 1990s showed that the emotional content of sentences we consider can have appreciable effects on our behaviour. This work found that people who thought about sentences mentioning age walked more slowly for a period afterwards than those who didn’t, and that thinking about sentences relating in different ways to assertiveness and politeness also tended to make the test subjects measurably more assertive or polite. (In these experiments the subjects believed themselves to be engaged in a language test with the sentences in question, and their rate of movement, or propensity for assertive or polite behaviour was measured in the corridor after they believed the experiment to have been concluded.)[146] A more recent study found that subliminal presentations of smiling faces before subjects were given the opportunity to pour themselves a drink led to pouring and consuming more of the offered beverage, and to increased willingness to pay for more of the beverage. Subliminal presentations of frowning faces had the opposite effect.[147]

It is not known how, if at all, these manipulations and others like them impact on discounting, or on performance on tasks like the IGT. We do know, though, that some stimuli can make a difference to discounting. A few years ago two researchers hypothesised that attending to pictures of attractive members of the opposite sex would get male subjects into a ‘mating mindset’ and that one feature of this would be temporarily steeper discounting for delay. Their reasoning was that unless the present was valued disproportionately, opportunities would be sacrificed too often. They found just what they predicted, and also (although to a smaller extent) that looking at pictures of expensive cars had a similar effect on female subjects.[148]

Some of the specific details of what is found in these sorts of studies may be surprising, just as the magnitude of the effect of some of the manipulations may be. The general point, though, is surely not in the least surprising: what people see around them makes a difference to what they do and how they feel about it. The ways that all sorts of places including public buildings, shopping centres, banks, hospitals and, of course, casinos are designed reflects this. It is a matter for politics and policy, rather than science, to decide how much manipulation is acceptable advertising, and how much is problematically coercive. Science can help by trying to quantify the effects of different environmental factors, and the specific differences made to some kinds of subject (for example those at risk of being disordered gamblers) and other determinants in fluctuations in relative vulnerability to impulsivity. This knowledge should surely inform and influence both treatment and policy.

2.4. Conclusion

Abusers of substances and disordered gamblers both show up as relatively impulsive on behavioural and psychological measures of impulsivity, and also tend to perform poorly on some cognitive measures of decision. We need to know more than we do, which is presently rather little, about how the behavioural and cognitive approaches relate to one another.[149]

The extent to which early impulsivity is a predictor of later disordered gambling is not known, although one suggestive study indicates that it is a predictor, and hence that individuals at risk of developing into disordered gamblers could be identified early. Longitudinal research is clearly required.

Although we know quite a lot about the effect of environmental factors and other consumption behaviour on some kinds of decision making, we know relatively little about what difference any of these factors (except alcohol consumption) make to gambling behaviour, and the risks of disordered gambling.

3. The Neuroeconomics and General Neuroscience of Pathological Gambling

EXECUTIVE SUMMARY

Two branches of neuroscience, neurochemistry and the newly emerged discipline of neuroeconomics, have jointly achieved major breakthroughs over the past few years in the understanding of addiction.

Neuroeconomics studies the behaviour of the brain as a calculator of the relative values of different possible and actual rewards that a person or other animal could pursue. Neuroeconomists have developed a rigourous and well-confirmed model of the computation of reward values, known as the `reward learning / predictor valuation model’. Principal evidence for this model has been gathered by scanning subjects’ brains under functional magnetic resonance imaging (fMRI) while they perform behavioural tasks under experimental control. This method has allowed the model to be tested in application to specific brain systems.

The most important of these is the so-called reward system, implemented by the joint action of an old part of the brain, the limbic system, and a part that has expanded with human evolution, the prefrontal cortex. This system integrates four main computational activities: (i) learning environmental cues that predict reward, (ii) learning comparative values of rewards, (iii) focusing attention on cues that predict rewards, and (iv) motivating the system to act on the basis of these cues. A finding of neuroscience that is surprising to common sense is that pleasure is only indirectly connected to reward, and is a relatively weak motivator of behaviour compared to other things, especially prospects for positive surprise.

Neurochemical and neurodynamical study of the reward system has broadly revealed how it works. Key to its operation are relationships among three main brain chemicals, or neurotransmitters. These are dopamine, serotonin, and GABA. Each is favoured by dedicated receptors in different parts of the system, and the arrangement of these receptors creates natural pathways by which the four computational processes above are carried out and integrated.

Drug addiction has been identified as, at the neurochemical and neurodynamical level, a pathology of the reward system. Different drugs exploit different pathways, with that exploited by stimulant drugs (cocaine, amphetamines) being the simplest, and those exploited by alcohol and opiates being more complicated. But all share general properties. They create flooding of dopamine into a particular part of the reward system. This causes the system to focus attention on the environmental cues that predict this flood, which are those associated with self-administration of the drug. The system then learns to orient toward these cues and is motivated to act on them. At the same time, glutamate levels are increased in prefrontal cortex and modify its neural connections. The modifications in question interfere with normal prefrontal inhibition of older brain systems and thereby cause behavior to become more impulsive. This facilitates drug-taking, and so addiction amplifies and locks itself in. The reward system becomes less receptive to alternative sources of reward. As addiction solidifies, the system learns to respond not to the drug itself, effects of which are readily predictable, but to the cues that predict it.

The neuroeconomic model explains why this all happens. Because the reward system is a learning system, it responds mainly to surprising events. This aspect of the system is responsible for the familiar pattern whereby an addict becomes sensitized to her drug, consuming ever more of it and with increasing concentration on it to the exclusion of other rewards.

A good deal of recent evidence strongly suggests that gambling, in some people, triggers reward system response that mirrors its response to the most direct form of chemical hijacking, that of stimulant drugs. This evidence is reviewed in depth in the chapter.

In light of what we now know about the reward system, and in light of the neuroeconomic model, it is less surprising that gambling should be a target of addiction. Gambling activities are engineered to produce rewarding surprise – that is, to produce exactly what stimulates the reward system. The gambler is buying a maximally convenient, direct manipulation of her mesolimbic reward system. If the result is neuroadaptation of prefrontal circuits to reduce inhibition of impulse, her brain is changed into the characteristic brain of the addict. She will experience cravings if she stops gambling, because other rewards will be less able to attract the attention of her reward system or motivate action. Thus she will be prone to relapse if she adopts a policy of trying to abandon gambling as its costs to her mount.

Fortunately, knowledge of the neurochemistry of addiction brings with it the promise of chemical interventions to counteract it. The chapter reviews very recent pilot studies – there being nothing more extensive as yet – that strongly suggest that AG can be effectively treated by drugs that disrupt addictive learning. These drugs do this by interfering with the neurotransmitter interactions that produce dopamine response amplification and sensitization. Of particular interest and promise are so-called `atypical antipsychotics’ that inhibit reuptake of dopamine in the heart of the reward system. The chapter concludes by indicating grounds for hope that within a few years there will be standard neuropharmacological therapies that allow addictive gambling to be effectively managed in the majority of cases, or perhaps even in all cases.

3.1. Neuroeconomics

People are not identical with their brains. By this we don’t mean to allude to any possible non-physical reality; that is an issue for philosophers and not relevant to our scientific ambit here. Rather, we just mean that insofar as people are identified with the totality of forces that influence their behavior and self-understanding, some of these forces reside outside their brains – especially in the patterns of cultural expectations and interpretations in which people participate. Let us illustrate what we are talking about by referencing it immediately to disordered gambling phenomena. In some cultural circumstances, a person might fit our definition of a problem gambler not because they’re compulsively drawn to gambling per se, but because they’re conflicted between accepting and rejecting some prevailing social norms, and gambling venues are convenient settings where these conflicts play out. Think, for example, of princes from conservative Arab monarchies who frequent European casinos. Some of these people might struggle to reduce their use of these facilities because of conflicts between social expectations on them and their own perceptions of themselves. They would thereby fit the definition of a ‘problem gambler’. But we should not suppose that we could ever find a single kind of pattern in their brains that would be equivalent to the problem as experienced separately by each one of them. What they have in common with each other is a shared social problem that could express itself in different neural patternings from one individual to another. Of course, the brain plays an essential role in the control of every behavior by every human being. Our point here is merely that we should not suppose in advance that phenomena we’ve grouped together under social considerations necessarily have anything interesting in common from a neuroscientific point of view. All human institutions give rise to some distinctive social problems that are only problems at the level where we categorize patterns by reference to social properties. Organized gambling institutions are no exception. Thus our broad category of ‘disordered gambling’ almost certainly includes some problems of this kind. Neuroscience should not be expected to be of much help in addressing them.

However, in this chapter we will explain why neuroscience does look like it will be the overwhelmingly important contributor to effective individual and public policy responses to an important sub-variety – one that may constitute the overwhelming majority of – pathological gambling. We stress that this is an empirical scientific discovery. Even if everyone might have expected that pathological gambling would have turned out to have a standard neuroanatomical or neuropsychological component, it was a perfectly reasonable hypothesis until very lately that it might also have had an irreducible social aspect that wasn’t most usefully addressed by neuroscience alone. In fact, we think that as recently as two or three years ago this was still the best hypothesis. However, for reasons we will describe in this chapter, discoveries made and reported mainly since 2002 suggest very strongly that much or most pathological gambling is one expression (along with addiction to stimulant drugs such as cocaine and amphetamines) of a quite specific neural pathology. Because the pathology is relatively specific, it lends itself to direct pharmacological response. We will argue that, thanks to these recent neuroscientific breakthroughs, there is real hope for the first time in human history that forms of addictive / compulsive behavior that have plagued us since at least the dawn of civilization could soon become as manageable as tooth decay or the control of large wild predators.[150] In the subsequent chapter we will comment on the implications of this for the organized gambling industry and the regulatory and health policies that govern it.

This is the chapter of our report that will involve by far the greatest number of scientific ideas and terms with which most readers will be unfamiliar. The chapter is mainly about the brain, including its parts and the chemicals that make it work. None of these parts or chemicals have everyday names, so it is simply impossible to talk about neuroscience in plain English. Yet we are claiming that neuroscience is about to revolutionize our understanding of and capacity for dealing with pathological gambling. It would therefore undermine the whole point of our work in this report to pass over the details because they can’t be expressed non-technically. We are going to make an argument here about pathological gambling, and it is going to be based on evidence. Therefore, the reader must see what this evidence is so she can decide whether she thinks our argument is persuasive.

Faced with this challenge, of presenting evidence to non-technical readers about things and processes for which there are no words in non-technical language, we have adopted the following devices. First, at each stage where some technical terms are about to put in their initial appearance, we will introduce them explicitly. Terms being introduced in this way will be put in bold type. Second, in the parts of the chapter where we’re presenting technical evidence, we’ll follow each paragraph that states that evidence with a non-technical synopsis that states its main point in everyday English. We want to make clear that the reader trying to make up her mind about whether the evidence supports the conclusions we draw from it will not be able to do this just by reading these synopses. To reiterate: the evidence can’t be put in everyday English because there are no everyday words for the objects of neuroscience. The synopses are for the reader who just wants to know what we are claiming and not why we are claiming what we do. Where readers with a main interest in policy and treatment are concerned, this might be exactly what they’re looking for, and the synopses are for them. To readers who want to follow the argument and get a sense of why neuroscience portends such big changes in the shape of the pathological gambling problem, we suggest that they struggle with the unfamiliar words. For that is all that is truly technical here. We have tried to explain all deep complexities in terms of more straightforward ideas, and there are many subtleties we’ve left out as inappropriate in a document that isn’t intended as a scientific paper. What will make the argument outside the synopses difficult reading are just new words, and we’re going to introduce all of them. But once they start coming in thick batches, as is inevitable, anyone aiming to follow the details will find they need to go slowly.

Straight off the bat we need to introduce vocabulary for the various branches of the sciences that are relevant to the brain mechanisms and processes underlying pathological gambling. Neuroscience refers indiscriminately to all of them together. Neuroanatomy studies the physical components of the brain, their placement relative to one another, and the mechanics of their interactions. (It is the part of neuroscience most analogous to auto mechanics.) When the reader looks at Figure 1 in section 3.2, the areas she sees labeled are neuroanatomical parts. Next: neurochemistry and neurodynamics study, respectively, the molecular and the electrical / computational bases by which these parts of the brain handle and transform information and pass it from one part to another. This refers to the fact that the brain does its work in two main physical ways. The first is by sending chemical signals that communicate about overall conditions – e.g., “Things are good here”, “Things are getting worse here”, “I am excited” – and generic responses – e.g. “I want to fight”, “I want to get away from this”, “I want you [another brain area] to stop talking to me”. The second is by sending electrical signals that communicate about specific states of affairs – e.g., “That is a red object”, “That is Mom”, “The capital of Zambia is Lusaka”.[151] In this chapter, the details we’ll get into will mainly concern neurochemistry rather than neurodynamics. Finally, neuropsychology studies the ways in which neuroanatomy, neurochemistry and neurodynamics interactively influence perception, motor behavior, decision, thought, emotions, mood and consciousness. When you wonder about how your brain contributes to production of your thoughts and actions, identified in terms you can present to yourself in everyday language, you’re mulling on neuropsychology.

The disciplinary perspective from which we have written this chapter is that of a newcomer to the above set of neurosciences, one which will strike many readers as a bit bizarre on first encounter: neuroeconomics. We think that this is the most fruitful organizing platform from which to provide an integrated neuroscientific account of (much or most) pathological gambling. However, neuroeconomics is a sufficiently novel construct to most people’s ways of thinking that we can’t introduce it, as we did the neuroscience branches above, with just a few sentences. Instead, several pages of explanation, including explanation of the perspective’s relevance to pathological gambling, are in order.

The basic assumption behind neuroeconomics is that natural selection couldn’t leave all valuation that an intelligent organism[152] must perform up to social-psychological processes. By ‘valuation’ we refer to the relative ranking of possible or actual states of affairs in terms of ‘better than’ and ‘worse than’.[153] Without an evolved internal currency in the nervous system, a creature couldn’t assess the relative value of events relevant to expected supplies of different things it wants. For example, suppose an organism encounters a mating cue and a food cue at the same time, using different sensory modalities (i.e., information processed through different sense organs). To which shall it respond? Or, if it can respond to both, how shall it trade off possible relative allocations of attention to them? An organism that could only compare them by first working out a speculative theory of its own economic situation would likely fare very badly in any real environment. If it is to be guided by the evolutionary learning of its species – which is a very powerful source of information because it’s been going on for so long – then its biological brain, not just its learned social / cultural contingencies, will need to play a role in informing its judgments. An organism that failed to make any use in its economic decisions of what its species had learned in evolution would be a very badly adapted – indeed impossibly badly adapted – organism. This means that the brain must directly compute over units that can represent the value of mating or eating (or fleeing or fighting or sleeping etc.) at some more abstract level – roughly as a person can weigh her house and her breakfast and the value of her life to her spouse in one currency, namely Dollars or Rands.[154]

So, what is this internal currency? Just a few years ago, no one had a very clear idea. Now, thanks to functional Magnetic Resonance Imaging (fMRI)[155] and other new technologies[156] that let us non-invasively scan groups of neurons (brain cells) in operation while people behave and think, we know.[157] Experiments performed initially with monkeys, but so far confirmed in humans, overwhelmingly suggest that the internal currency consists in variations in firing rates in dedicated neurons in parts of the brain that collectively make up the so-called ‘reward system’ (that is, the system for processing information about actual or potential valuations). An increase in spike rates in these neurons means that a stimulus seems to the brain to be ‘better than predicted’, a decrease means it’s ‘worse than predicted,’ and no change means it’s ‘just as expected’ (so, this signal says to the rest of the brain, don’t pay more or less than the default level of attention to it). Just to reassure the reader that this is a properly worked out scientific hypothesis, and not just a vague qualitative association, we can state, on the basis of work including carefully controlled observations over simian and human subjects, a ‘predictor-valuation’ model due to P. Read Montague and Greg Berns.[158] (Mathematics is coming! Again: the reader will not need to try to follow the details for the sake of things we’ll say later. But we think it is important that she see the form of this model, so that she understands what kind of theory it is: it’s a set of exact principles, like Newton’s laws of motion. Thus we can exactly relate neuroeconomics to the behavioral economics discussed in Chapter 1. We think that if we didn’t show the reader the form of the model, she wouldn’t be able to understand why this is a kind of economics.) Suppose first that R(x,n) estimates the value of a reward distributed at various possible times x, y, z, …, n in the future, scaled according to the uncertainty attending to the intervals between the estimation point and each time as follows:

R(x,n; D) = ò_-_¥⁺^¥dy G(x – y, (x – n)D)r(y), where G(z,b) = (2pb)^-1/2 exp{-z²/2b} and D is a constant.

Then the following equation describes the value F(n) the brain attaches to getting a particular predictor signal at perceptual time n:

F(n) = ò_n⁺^¥dx e^-q(x-n) ò_-_¥⁺^{¥ dy} G(x – y₁(x – n)D)r(y)

= ò_n⁺^¥dx {e^-q^(x-n)} • {R(x,n; D)}

= ò_n⁺^¥dx {discount future time x relative to perceptual time n} • {diffused version of reward estimate r(x) for some x and n}.

The predictor-valuation model provides the molecular foundation for the whole-system-level (‘molar’; see Chapter 1) model of the kind of learning in which the reward system engages. This is so-called ‘Temporal Difference (TD) Learning’.[159] Once again, we will give the equation, though (again) the reader will not need to try to learn it for the sake of understanding anything that comes later in the report. TD learning is a rule – an algorithm – that describes the way in which the reward system estimates a value function V^* (that is, a principle for taking a stream of possible information as input and turning it into a stream of advance relative value estimates as output). As McClure, Daw and Montague (2003)[160] describe it, this function “relates the situation at a particular time s_t to the expected, time-discounted sum of rewards (idealized as numeric measures r of received utility) that can be earned into the infinite future.” Now suppose that t, t+1, t+2 etc. represent times on some arbitrary measurement scale. Then the TD equation is

V^*(s_t) = E[r_t + gr_t₊₁+ g²r_t₊₂ + g³r_t₊₃ + …]

which we close[161] by writing

V^*(s_t) = E[r_t + g V^*(s_t_{+ 1})]

where g is a discounting parameter, as discussed in previous chapters of this report, between 0 and 1.

What this equation describes is the procedure by which the reward-system learning algorithm continuously inputs new information to keep refining its estimate of V^* to get a particular stream of actual temporal valuations V. From this we can define a measure d of the extent to which the value estimates of two successive states and a reward experienced by the system are consistent with one another:

d(t) = r_t + g V(s_t_{+ 1}) – V(s_t).

d is an error signal that pushes V(s) towards better estimates as it gets more data. For example, if V(s_t_{+ 1}) turns out to be better than expected, then d(t) will be positive, thus indicating that V(s_t) needs to be adjusted upward.

Learning that these molar and molecular functions describe the value-learning processes of the brain amounts to discovering the brain’s economics, since we can use them to quantitatively model the opportunity costs the brain implicitly pays – in its own terms – for attending to one stimulus over another or issuing one motor response[162] rather than another.[163] This is why we refer to the study of the reward system using these parameters as ‘neuroeconomics’. fMRI work has suggested that there are levels of discrimination in value-coding structure in the reward system from which the predictor-valuation model abstracts. (That is, the model averages over some of the details of reward representation at the molecular level.) Some neurons encode expected value (magnitude X probability) of rewards, while others encode a reward’s subjective utility in context (i.e., its opportunity cost in terms of its cardinal ranking[164] relative to other possibilities). It is likely that as neuroeconomics develops further over the coming years, it will call forth application of the full set of technical resources of mathematical microeconomics in understanding the principles by which the brain relates perception to the allocation of attention and preparation for action.

We recommend neuroeconomics as the framework for integrating the neuroscience of (much or most) pathological gambling, and addiction-like behavioral phenomena in general, because these are pathologies of the reward system. As we discussed in Chapter 1, they are patterns wherein the system locally values rewards that cause the global welfare of the person the system regulates to drop over time, and to regret her earlier choices despite the fact that she could (at least from some past point) have foreseen her own regret. Under what circumstances does V(s_t) keep pushing the valuation of another gamble, or another hit of cocaine, higher than the available alternatives, even when some other aspect of the person – either some other system in her brain, or her whole self in aggregate – would prefer that she choose one of the alternatives? Just a few years ago the answers to this question were speculative. Now we know a good deal about them – though very far indeed from everything about them – on the basis of neuroscientific evidence. They are the subject of this chapter.

Before we go into the brain, however, let us pause to reiterate a few conceptual lessons driven home by the discussion in the previous chapters. If all people naturally observe the matching law (see chapter 1 section 2) and meliorate rather than globally maximize, then we don’t need to resort to a specific reward system pathology to explain mere problem gambling. All we need for that is the fact that the person enjoys the gambling experience at all (some people don’t), and that she receives that enjoyment in the short run. Then hyperbolic discounting alone will cause her to tend to regret the amounts she gambles except to the extent that she has learned to bundle her future choices with present ones, in the way described by Ainslie that we discussed in Chapter 1 sections 6 and 7. With respect to gambling (as with alcohol, marijuana and many potentially addictive targets) we know that most people successfully learn to bundle – we know it simply because most people manage to gamble non-pathologically and drink moderately. (Most also occasionally bundle less well than they wish they had awhile later, and many deliberately choose not to bundle from time to time, for reasons discussed in Chapter 1.) From the policy perspective, we address mere problem gambling by deciding to what extent we should institute measures that prevent gambling facilities from making it more difficult for customers to bundle, or that force them to help customers bundle a bit better. We will discuss this in Chapter 4.

What is shown by widespread participation in controlled gambling and drinking is that, at least with respect to these activities and some others, people can marshal resources stronger in effect than the brain processes we are about to review which dispose them toward addiction. Knowing this leaves many open questions. Are there opponent brain processes that, in most people, are sufficient by themselves to counteract addictive dispositions? Or is social reinforcement of moderation required? Must people make reference to socially constructed behavioral book-keeping strategies to maintain their personal (bundling) rules? (i.e., if Robinson Crusoe had a distillery on his island, and a limitless supply of fermentable material, could he avoid alcoholism?) Certainly, if social scaffolding is necessary to maintain personal rules, then there must be some brain processes that the scaffolding recruits and operates through; but this does not show that those processes could be adequate for the job on their own. It is relevant here that scientists in laboratories can bring every rat to addiction to any substance that is a target for addiction in any people; rat brains, at least, do not have adequate natural safeguards against their own addictive mechanisms. So perhaps our brains are like that too, and require social support to keep in order.

From an evolutionary point of view it is not surprising that brains on their own are vulnerable to runaway consumption of some rewards. Until our own ancestors developed agriculture, no animal could exercise enough control over its own supply schedule to need to control its addictive dispositions itself. When a group of elephants comes across fermented fruit, each one eats as much of it as he or she can and, if there’s enough to go around, they all get intoxicated. But however much they would enjoy repeating the experience the next day or the day after, they do not have this option because they cannot cause fermentation to happen, or collect and maintain a reliable supply of their drug. Thus their ecological circumstances are a barrier against addiction, and there was no reason why natural selection would have wasted resources equipping them with an inboard one.

What about people? Since at least the dawn of agriculture, we have been able to control our supply schedules so as to allow for addiction to substances. By all evidence substance addiction has been a problem for some people since at least that far back.[165] Has natural selection responded by evolving a homeostatic (i.e., self-correcting, like a home air temperature system) neural mechanism that, in most people, blocks consumption of addictive substances after a certain threshold? Or did social learning gradually fill our cultural environments with normative reference points, customs and rituals that allow most people to avoid being enslaved by their own brain mechanisms? Were our early agricultural ancestors more threatened by addiction, before the relevant social scaffolding and/or homeostatic mechanisms evolved, than modern people? These are all questions for physical, social and economic anthropology. At present, we know some things that are relevant to addressing them, but we don’t know the answers. In particular, we don’t know whether a genuinely asocial person who nevertheless (miraculously, all by herself) controlled her supply schedule could avoid addiction.

What about gambling? Pathological gambling, unlike substance addiction, does not require control over a scarce resource. So there is no obvious reason why pathological gambling, if some of it is addictive in the way that some substances are, had to wait for the development of agriculture. And, certainly, some contemporary hunter-gatherers gamble a great deal (by comparison with their devotion of time and energy to alternative pursuits, such as procuring food).[166] There is no evidence that non-human animals gamble, but this doesn’t tell us as much as one might suppose. Wild animals with a taste for gambling would be behaviorally reckless animals, and such animals would generally perish quickly. Natural selection might therefore, just as in the case of dispositions to substance addiction, encounter no pressure to build neural safeguards against hijacking of animal reward systems by habitual gambling, since, again, ecological contingencies would ensure against such habits becoming common. Perhaps similar evolutionary logic applied to early hominids. To find out whether any animals are susceptible to addictive pathological gambling we would have to rely not on observation but on carefully contrived experiments, in which we deliberately put animals into non-natural environments where their possible addictive tendencies could flourish. To our knowledge, no experiments of the relevant type have been attempted.

Finally, in trying to come to grips with the relationship between social and neural factors for addiction, one might ask: are there some specific addictive targets for which socially reinforced personal rules are generally not sufficient? Nicotine is an obvious candidate here; comparatively few people smoke non-addictively. However, this is a far-from-decisive case for the existence of neural addictors that are independent of social learning. First, until fairly recently people who liked smoking had little reason known to them for bothering to resist addiction; there might well now be growing numbers of non-addicted (e.g. ‘weekend’ smokers) whose prevalence has not been carefully studied.[167] Second, even if nicotine is unusually addictive, this may be not because it is an unusually effective neurochemical hijacker of the reward system. Rather, it might simply be that the marginal cost of each individual cigarette smoked is so low, and is known by smokers to be so low, that people must develop uncharacteristically sophisticated personal rules before they can learn to control smoking, by which point addiction will usually have taken effect.

We have now reviewed salient points in our current scientific ignorance about the relationship between problems due just to hyperbolic discounting, against which personal rules alone can be effective, and so-called addictions. This general ignorance made it possible to doubt, until very recently, whether any pathological gambling, in the sense of the DSM-IV operationalization, is addictive in a neurochemical sense.. That is, it was possible to suppose that pathological gamblers might simply be people who are, for social-psychological reasons, especially incompetent at constructing personal rules. Since we have known for some time that some substances make people addicted by neurochemical mechanisms, this would have been one basis on which it could have been reasonable to deny that there is really such a phenomenon as addiction to gambling. If this is how the facts looked like coming out, then this survey could have stopped with the discussion of the behavioral economics of disordered gambling, and not required a long chapter on neuroscience.

As we will now show, however, this is not how the facts have come out. In some people, gambling hijacks the reward system in the same way that stimulant drugs do. Neuroscientists have isolated many of the main details of the mechanisms and processes by which this happens. They are in the course of developing chemical therapies that show great promise for reversing such hijacking. And there are grounds for optimism that they will soon be able to predict who is at risk for addictive gambling on the basis of a genetic screen. While non-addictive disordered gambling is probably something at which policy should aim only for some socially variable level of control (because that’s the best that is possible except at outrageous costs), addictive gambling might well, in consequence of the recent breakthroughs in neuroscience, be something we can truly manage and aim to minimize. For reasons we will discuss in the final chapter, we think this has significant implications for industry participants and regulators.

One last point on which we must touch before narrowing our focus in this chapter to addictive gambling (AG) is: how much of pathological gambling is AG? The answer is that we do not know. The possibility that we might soon be able to diagnostically screen for AG has only opened up over the past couple of years, so it is still too soon to do the kinds of scientific surveys we’d need to do to estimate the prevalence of true AG among pathological gamblers as classified by DSM-IV. However, there are grounds for thinking that the proportion is very high – for thinking, indeed, that all pathological gamblers might be AGs. This is simply that, as we will document in section 3.5, empirical studies have found strong correlations between neurochemical properties of the reward system that have been identified with addiction and tendencies to pathological gambling. The pathological gamblers in these studies were of course identified using SOGS and similar screens. We know that these screens over-select for pathological gambling, as discussed in Chapter 1, section 1. Therefore, most samples probably contained people who, not being pathological gamblers, were surely not addicted gamblers (if there are any, which was the open question in these studies). But in that case, the proportion of non-addicted pathological gamblers in the samples that would have been consistent with finding strong correlations between addiction and pathological gambling has a low upper bound, because of the interference with statistical significance, the ‘noise’, already coming from the screening errors. Here, the tendency of SOGS to make errors in one direction as a screen for research purposes actually helps us make a useful inference. Most pathological gamblers, we will argue, are probably addicted, and it is possible that all are.

Because we assume that AG is a subset of pathological gambling – i.e., that no mere problem gamblers are AGs – we can without confusion use ‘PG’ here to refer to pathological gambling or gamblers (depending on grammatical context). Whenever there is a need to mention the distinction between pathological and problem gambling, both phrases will be spelled out. ‘DG’, as before, will refer to disordered gambling or gamblers (depending on grammatical context), and NADGs to non-addicted disordered gambling or gamblers.

So, now, into the brain.

3.2. The Reward System

The organization of the chapter is as follows. Because AG is a pathology of the reward system, we must first follow the recent breakthroughs in neuroeconomics that have brought about our new understanding of this system. This will occupy the present section. In Section 3.3 we will then focus on addictive phenomena in general, so that we can later go on to understand AG as a specific – and highly revealing – manifestation of a general kind of reward system malfunction. Sections 3.4 to 3.6 will zero in on our primary topic, AG itself. This will include discussion of AG as the simplest conceptual template (by which we mean, basic form for purposes of organizing our thought about more complicated forms) for addiction in general. In section 3.5 we will review the experimental neuroscience of AG, and in section 3.6 we will survey implications for drug-based treatment (that is, neuropharmacological treatment, i.e., based on drugs that work on neurochemicals).

Our first task, then, is to introduce the basic neuroanatomy, neurochemistry and neurodynamics of the reward system. A key general point is that the system integrates aspects of both an old, interior part of the brain, the limbic system, that we share with vertebrates in general, and the prefrontal cortex (PFC) whose dramatic growth (along with that of the frontal cortex) in hominids was a salient feature of our evolution and a necessary factor in our unique form of adaptive intelligence. Whenever evolved structures must co-opt the functional outputs of other structures that evolved much earlier, we should have our eye out for characteristic pathology patterns that result from the fact that the older structure was not selected for its capacity to serve the newer one. Natural selection cannot yoke together different systems evolved under different pressures and with different available building blocks, and avoid what engineers call ‘kludges’ – that is, incipient vulnerabilities that are patched around when the system is running under optimal conditions, but that expose predictable bugs when circumstances degrade in characteristic ways. The widespread and relatively standardized pathologies of addiction may well be kludges stemming from the imperfect adaptation of the limbic system to its role as a service department for newer cortical systems.

The figures below show the brain regions that play leading roles in the neuropsychology of reward. In the old interior part of the brain we have the striatum, which is itself part of the basal ganglia and includes the nucleus accumbens. It interacts intimately with the ‘emotional centre’, the limbic system that is composed of the hypothalamus, amygdala, and hippocampus. Below them are the ‘basecamp for reward’, the ventral tegmental area (VTA) and, flush beside VTA, the pars compacta of the substantia nigra (SNpc). Just exterior to these, and separately evolved, is the anterior cingulate cortex.[168] In front and outside we have the recently expanded prefrontal cortex, divided into ventromedial and dorsolateral regions. At the bottom of it is a dedicated site of cortical reward processing, the orbitofrontal cortex. Reward processing and response patterns in people are functions of the ways in which these structures interact.

To help fix the idea of the various circuits and directions of influence we’ll be describing, we begin with a series of images of the brain with areas of interest highlighted. The brain is structured, layered and folded in all directions, so no single image can do the job of showing what we need here.

In the first figure below the exterior of the brain is shown from the side, with the motor cortex, the dorsolateral prefrontal cortex (PFC) and the orbitofrontal cortex labeled. The ventromedial PFC borders the dorsolateral PFC, but lies on the insides of the two hemispheres of the brain, and so would not be visible from this orientation. The areas highlighed in this figure occur on both sides of the brain.

Figure 1

The main parts of the set of reward specialized brain systems we will be discussing are deeper within the brain. The second figure below is also a side view, but in it the one hemisphere has been removed, so that the middle of the brain (‘facing’ in the same direction as the image above, so that were the eyes included in either image they would be looking to the left) is being viewed. The substantia nigra is located close to the ventral tegmental area appearing on this figure:[169]

Figure 2

The third figure shows the brain in a more roughly three-dimensional view, as though it was facing in the direction of the thick black arrow. Note that the hypothalamus borders the thalamus, and does not appear on this figure because it is obscured by the dorsal striatum.[170]

Figure 3

Our final figure is schematic, sacrificing attempt at fidelity to anatomical proportions in order to represent a system of connections. Some of the terms appearing in it (‘dopamine’, ‘GABA’) are brain chemicals (neurotransmitters) discussed later in this chapter.[171]

Figure 4

It has been recognized for many years, mainly on the basis of work with animals, that the VTA and the SNpc are the two structures that send reward signals out through the rest of the brain. Their role was revealed in a tradition of experiments whereby rats who were supplied with levers they could press to directly stimulate these areas through implants would do so repeatedly and relentlessly, ignoring less direct reinforcers like food and water to the point of death.[172] In this pattern researchers immediately recognized the prototype for addiction. The discovery of its underlying neurochemistry and neurodynamics has constituted the main part of the history of reward-system research. Relevant facts based on animal work have been under assembly for decades; but development of the integrated neuroeconomic model had to await fMRI investigation since 2000.

We distinguish VTA and SNpc from one another mainly by reference to the different pathways through which they feed the reward signals forward. These signals consist in release of a neurotransmitter (see next paragraph) called dopamine. It is through variance in dopamine-releasing rates that VTA and SNpc implement the predictor-valuation model described in the previous section.[173] Above-baseline dopamine release is interpreted by the rest of the brain as meaning that a reward has been received that was better than expected, disappointment relative to expectations is indicated by below-baseline dopamine release, and baseline dopamine release is the default state meaning that reward levels are as predicted. It will be crucial to our understanding of addiction and PG that the brain in general is a system that responds to changes; it is not like a monitor of a film that explicitly registers everything going on.[174] It attends to surprisingly good and surprisingly poor rewards and tends to ignore conditions that don’t change the status quo. The large human brain is primarily an organ adapted for learning – to which status quo conditions make no contribution. Therefore, brains do not waste resources on them. The reward system is no exception to this principle. Though common sense is not used to thinking of reward-processing as a form of learning, that is what it is and that is how it is neurally implemented.[175]

We will have a great deal to say about dopamine throughout this chapter, so before we proceed further we should offer an explanation of neurotramission and the neurotranmitters that perform it. One of the features of the brain that most contributes to its complexity is that it uses a two-dimensional information transmission system. One dimension is electrical. Brain cells are wired together in circuits through which varying electrical pulses are passed along connecting cables called dendrites. Different brain cells participate in different neural pathways partly by means of their various specific dendritic connection patterns. However, neurons don’t transmit the signaling pulses from one to another by implementing an electrical switchboard. Instead, they influence one another’s probability of receiving signals by emitting chemicals at synapses. How likely it is that a given signal will cross the space between the synapse of one neuron and the synapse of another neuron is a function of the level of that synapse’s chemical of interest in the fluid of the brain at the site in question. Neurons are sorted into types on the basis of which neurochemicals their synapses release and respond to. These neurochemicals are called ‘neurotransmitters’, and so neurons are typed after the neurotransmitters in which they deal. For example, thanks to the recent popularity of anti-depression agents like Prozac that modulate serotonin levels in different parts of the brain, serotonin has become the best-known neurotransmitter among the general public. We then group all the parts of the brain that communicate with one another by releasing and taking up serotonin as together comprising the serotonergic system. Dopamine is the neurotransmitter we will have most to say about in this survey. Following the previous example, the pathways through the brain created by linked dopamine releasers and receivers are referred to as the dopaminergic system. We apologize to the reader for the fact that this talk of ‘systems’ – for which we aren’t responsible – wasn’t designed to help lay people keep easy track of things. We will refer to three neurotransmitter systems, the dopaminergic, the serotonergic and the noradrenergic, in this report. The reader will get confused if she thinks of the reward system as a separate system from these neurotransmitter systems. The latter are distinguished from one another neurochemically. By contrast, the reward system is distinguished neuropsychologically, that is, by reference to its role in relating the whole organism’s perceptual uptake and its behavior. Consider, by analogy, the training system and the player-ownership system in rugby: neither system is a part of the other, but they are related, and the same players participate in both systems. So, similarly, the neurotransmitter systems play distinct roles within the reward system, while also playing other, complementary, roles in other neuropsychological systems such as the language-processing system. One and the same neuron may be part of the dopaminergic system from the neurochemical point of view, part of the limbic system from the neuroanatomical point of the view, and part of the reward system from the neuropsychological point of view.

The reward system is the set of processes in the brain that computes the outputs broadcast as dopamine pulses by VTA and SNpc. We will begin our tour of it with the hypothalamus. This directly monitors basic bodily states, such as blood sugar and temperature levels, that are direct proprioperceptive (i.e., inwardly but unconsciously perceived) indicators of changes relevant to the overall system’s present and expected utility. Hypothalamic information serves as fundamental input to the reward system, and connects reward to representation of hedonic pleasure,[176] which (simplifying somewhat) is the business of a part of the anterior cingulate cortex. We have italicized ‘connects’ here to stress that although reward and pleasure are related, they are not the same thing. This may be the single most important point on which neuroscience calls for a revision in popular ‘common sense’ about addiction and compulsive behavior. When reward and pleasure are muddled together, as they are in everyday thought and talk we’re led to think that (e.g.) a heroin addict must pursue heroin ‘because she likes it’. Then her attempt to kick her habit seems to require us to say that she simultaneously doesn’t like it – and now we find that it’s hard to say anything coherent about the whole matter. Neuroscience forces us to separate not just two but three kinds of psychological state: (i) the extent to which a state is hedonically pleasurable; (ii) the extent to which a change of state is treated by the brain as a reward; and (iii) the extent to which a person judges a change of state as improving or reducing their immediate or longer-term welfare. It turns out from the study of the brain – i.e., this is not something one settles by appeal to philosophical opinion – that (ii) and (iii) are more intimately associated with one another than either is with (i). A landmark fMRI study suggesting this was performed and reported by Berns and colleagues in 2000.[177] They found that subjects’ subjective reports of pleasure discrimination between fruit juice and water were correlated only with activity in cortical areas associated with sensory processes, whereas responses to unpredictability were correlated with activity in nucleus accumbens (Nacc), thalamus and medial orbitofrontal cortex (OFC). These areas have in common that they are involved in dopamine-regulated input to attention allocation and (through connection with motor neurons) to preparation for action. Thus pleasure response seemed to be occurring outside the direct reward-response loop.

A wave of subsequent experiments building on this classic investigation[178] have established as a basic principle of neuroeconomics that people (behaviorally and neurally) ‘want’ many things they don’t (hedonically) ‘like’.[179] Furthermore, they can readily learn not to want things that they like, if the latter become predictors for other things they don’t want or like. For example, people who successfully lose weight often do so by learning to find fatty foods that they formerly enjoyed disgusting to consider. Of course, many targets of addiction induce hedonic pleasure while they’re being consumed, and this is often crucial to their initial attractiveness. However, just as many smokers actually find the sensations of at least many of the cigarettes they smoke (e.g., the 30^th of a long night) mainly unpleasant (while smoking them anyway), so we should not assume that a pathological gambler must experience a warm subjective thrill while she’s rolling the dice. Hedonic pleasure is so incidentally and unreliably connected to reward and to the motivation of behavior that we’ll have little further to say about it in this report.

Let us add just a bit more on this theme, since, as noted, it is probably the most important point on which scientific and everyday conceptions of addiction and compulsion diverge. There is an obvious reason why most people don’t clearly distinguish between pleasure and reward, even though they play such distinct roles in the brain. Consciously experienced pleasure is one (only somewhat reliable) proxy for, or predictor of, reward. Precisely because it is the consciously experienced signal of probable reward, it’s one we notice when we introspectively reflect on the motivations for our behavior. (We also notice some motivators programmed into us by sociliazation, such as injunctions to moral duty; these get coded in public language and we notice them by literally talking to ourselves as if we were other people, not by directly ‘looking inside’.) However, most of what goes on in our brains is not consciously accessible in itself, while nevertheless contributing to our interpretations of what we think we’re experiencing. Hedonic pleasure signals evolved in primitive ancestors of ours that had much less complicated reward systems than we do, and the pleasure system we’ve inherited from them probably responds unequivocally – that is, in a way not strongly mediated by reward signals – only to some very basic cues, such as air temperature, whether we are damp or dry, and sexual contact. The rest of what we loosely interpret as pleasure is probably built out of second-order associations between pleasure signals and reward signals (e.g., getting into bed predicts warmth and sex, so getting into bed comes to be experienced as directly pleasurable). As creatures whose main evolutionary adaptation has itself been adaptive intelligence, however, we have reward systems that were precisely selected so as not to give pleasure signals priority over our curiosity about changes in environmental reinforcers. This is why pleasure is not a very reliable proxy for reward in people. More to the point, as emphasized at the bottom of the preceding paragraph, it isn’t very important to behavioral motivation. People who suffer catastrophes to their anterior cingulates and report losing hedonic color from their subjective worlds still behave in broadly normal ways.[180] By contrast, lesions to the components of the reward system typically produce profound breakdowns, ranging from Parkinson’s disease to total motivational disintegration. Rats whose reward systems were trained to respond positively to terrible-tasting food ate the food while making facial expressions characteristic (in rats) of unpleasant experiences.[181] The fact that we are evolved to pursue what stimulates our reward system more diligently than what makes us feel good will turn out to be a fundamental part of the neuroscientific explanation of addiction and AG.

Returning from our digression on the pleasure/reward distinction to our tour of reward-system neurochemistry and neurodynamics, the hypothalamus dynamically represents (i.e., keeps updating with changing events over time) the visceral state of the body to limbic and cortical (‘cognitive’) systems. Each of these systems uses this basic data in its own distinctive way, and neuroscience is still some way from having an integrated account of the whole network. We know, of course, that at the molar level we get the appearance of hyperbolic discounting of utility, often controlled by development and maintenance of personal rules, as described in Chapter 1. Various molecular hypotheses are consistent with this molar account.

In a 2004 paper McClure and colleagues[182] report on an fMRI experiment that compared subjects offered choices that included offers of immediate rewards with subjects choosing only amongst more delayed prizes. They observed increased activity in limbic-system (ventral striatum and medial OFC) neurons when choices included earlier rewards, and increased PFC activity whenever subjects actually chose later, larger rewards. On the basis of this, they hypothesize that molar hyperbolic discounting is an abstract description of the aggregated composition of two separate discounting mechanisms lodged in different brain systems. Specifically, they “hypothesize that short-run impatience is driven by the limbic system, which responds preferentially to immediate rewards and is less sensitive to the value of future rewards, whereas long-run patience is mediated by the lateral prefrontal cortex and associated structures, which are able to evaluate trade-offs between abstract rewards, including rewards in the more distant future”. In earlier work, Laibson[183] shows how to technically recover some of the behavioral data explained by hyperbolic discounting by means of curves that combine two exponential curves of different slopes. He calls this ‘d-b discounting’, where ‘b’ (‘beta’) denotes a steep-sloped function corresponding to discounting of immediate prospects and ‘d’ (‘gamma’) denotes a less steep-sloped function corresponding to discounting of more delayed prospects. Then the hypothesis proposed by McClure and colleagues is that lateral prefrontal areas (which they called the “d system” after Laibson’s model) are active whenever subjects make any decisions. The limbic (“b system”) is active at a level comparable to that of the d system (“(and with a trend toward greater b-system activity)”) when subjects choose earlier options over later ones.

Non-technical synopsis: Brain systems don’t hyperbolically discount. The front of the brain discounts exponentially and the back of the brain does so too, but more steeply – it cares less about the future. Overall human (and other animal) behavior just seems hyperbolic because at that level our measurements average over cases where the front of the brain is in charge and cases where the back is in charge.

As Ainslie and Monterosso[184] point out, this hypothesis outruns the data. The evidence from increasingly many studies indeed tells us that elevated limbic activity and decreased or impaired prefrontal activity are independently correlated with impulsivity. But the key word here is ‘independently’. (This could not emerge given McClure and co.’ss experimental design; we will encounter the evidence in studies to be reviewed shortly.) We are not in a position to claim that we have rival centres of discounting that compete for control of molar choice. This would greatly complicate the existing model of the reward system, as Ainslie and Monterosso point out: “Although the increased activity of … lateral prefrontal cortex … in response to larger/later selections is an important finding, to accord these areas status as a separate decision-making mechanism would add a complicating factor that would have to be reintegrated with motivation”. Kable and Glimcher[185] have new fMRI data that we interpret as showing that, in fact, neurons in limbic and prefrontal areas implement similar (hyperbolically shaped) discount functions. If this is right, then bundling must, as Ainslie argues, involve incorporating future-indexed expected utility into immediately indexed utility. We could then say, if we like, that all DG episodes involve failure to so incorporate – that is, failure to bundle. Mere problem gambling involves (more or less often) occasional failure to bundle, for circumstantially contingent reasons that vary from occasion to occasion. PGs are people who cannot bundle where gambling is concerned. We should look directly to neurochemistry and/or neurodynamics to tell us what prevents them from doing so.

Non-technical synopsis: Even the smallest parts of the brain, and not just the whole person or other animal, seem to discount hyperbolically. Thus all avoidance of addictive-style behavior, even by parts of the brain, must involve bundling, and so there must be a neuroscientific account of it. Therefore, if PGs are people who can’t bundle no matter how hard they try, we should look in their brains to explain why their neural bundling mechanisms aren’t working.

Other hypotheses about specific contributions of different brain systems to overall behavioral valuation remain open. Glimcher and Rustichini[186] suggest that ventromedial prefrontal cortex attaches so-called ‘somatic markers’ – roughly, emotional biases – to outputs of the reward system that cause most people to fail to behave like expected-utility maximizers (see Chapter 1, section 3), even in the very short run, under certain kinds of situations. They discuss evidence, for example, that most people are averse to losses, in the sense that they’ll prefer an expected reward of a given magnitude to another of identical magnitude if the latter has a higher probability of inflicting a loss, and that they’ll pay (in expected utility) to avoid ambiguous choices. Glimcher and Rustichini cite observations suggesting that people with damage to prefrontal brain areas behave more like (short-run) expected-utility maximizers in these conditions. This speculative possibility is among several that merit further research, and is relevant to predicting the behavior of NADGs under different sorts of circumstances. There is no evidence, however, that the explanation of addiction or obsessive-compulsive behavior lies in these complications. They may account for respects in which healthy gamblers don’t behave like perfectly rational neoclassical economic agents, but they do not appear to be part of the basis of AG.

Non-technical synopsis: There’s evidence that emotional fear of loss might play a role in protecting people from NADG. But damage to emotional systems doesn’t seem to be a direct part of the basis for addiction (or, therefore for AG).

So let us focus more specifically on the reward system, within which, as we will see, the basis of AG seems to lie. A first key finding is that unpredictable stimuli produce stronger responses in VTA, Nacc and ventral striatum than do familiar ones.[187] A second finding is that the stronger Nacc response is a direct function of an elevated surge of dopamine to it from VTA. This response specifically signals positive surprises; dopamine neurons show only minor, probably indirect, response to aversive and neutral stimuli.[188] In particular, Nacc is unresponsive to such stimuli.[189] Thus, contrary to what common sense might have expected, reward and punishment are not coded along a single continuum to which an integrated brain system responds. (This will prove important to understanding why adverse consequences of addiction and AG are notoriously ineffective, at least by themselves, in reconditioning afflicted people.) Dopamine neurons do not discriminate with respect to the sensory modality through which stimuli are delivered; speaking anthropocentrically, they ‘care’ only about the extent to which a stimulus is relevant to the difference between the prediction and the delivery of its associated reward.[190] (This is important to explaining why, in addicts and AGs attempting recovery, anything that has been a characteristic part of the person’s habitual environment for drug-taking or gambling is an equally strong potential trigger of relapse.) As learning proceeds, the system stops responding to administration of the reward itself and responds instead to its predictors.[191] Furthermore, dopaminergic neurons show depressed response when rewards are smaller than what learning would have led the system to predict; thus the dopamine neurochemistry converges qualitatively with the predictor-valuation model that quantitatively describes the fMRI data.[192]

Non-technical synopsis: The limbic system gets surges of dopamine when it gets positively surprised. The part of the limbic system most responsible for addiction doesn’t get less dopamine than normal when it’s negatively surprised. So, in the brain, punishment isn’t just the opposite of reward, and we shouldn’t think we can cure addicts by punishing them. It doesn’t matter to dopamine release how the body gets the surprising information, so any clue that tells the addict ‘my drug is coming’ will have the same potential effect. Eventually, the system stops even noticing the reward itself and just responds to cues that predict them. Other parts of the system besides the part mainly responsible for addiction start responding sluggishly when cues turn out to be disappointing. So the dopamine system chemically mimics the neuroeconomic model of the whole brain.

A major barrier to the understanding of behavior by common sense is that everyday psychology misleadingly separates – or, at least, fails to strongly connect – the passive, perceptual aspect of reward and the active response to rewarding prospects. This partly stems from everyday psychology’s overly direct connection of pleasure with reward; pleasure, as we have seen, is relatively passive and weak as a response motivator. But natural selection did not build massive learning systems for us so that we could contemplate complex attractions; in the brain, learning is for doing. Note also (a point to which we will return in more detail shortly) that thinking about something (as opposed to thinking about other possible things) is not (as in Hamlet) the opposite of acting. It is a kind of action, since it involves the brain’s neurochemically and neurodynamically doing something (i.e., using resources for this rather than that). We should thus not think of reward learning as akin to a kind of perception, however much it has that aspect; the reward system is, fundamentally, a mechanism for allocating neural resources and prompting behavior. VTA and SNpc directly project to motor anterior cingulate and motor pre-frontal cortex. By this pathway, dopamine signals entrain motor habits that can proceed without conscious mediation after learning. This includes both ‘bodily’ behavioral habits and ‘thought’ habits in the sense of automatic associations we make when presented with familiar stimuli.[193] The portentousness of this for addictive and compulsive phenomena, in which people feel that they have lost deliberate control, will be evident. Where the reward system is concerned, paying attention to something and doing something about it are more or less the same thing – it can’t tell the difference between caring about (e.g.) gambling and getting busy gambling.

Non-technical synopsis: The reward system closely fuses learning what to attend to and acting on what it attends to. It’s a system for drawing the organism to something so as to act on it, without any thinking in between.

The most dramatic indication of the intimate relationship between the reward / dopaminergic system and motor behavior, and one that will turn out to be profoundly important to our understanding of AG, is the range of consequences attaching to impairment of the system’s capacity to send adequate dopamine signals. Parkinson’s disease is a degenerative disorder of the dopamine system as vested in substantia nigra. Its symptoms are loss of motor control (expressed in movement inability or difficulty [akinesia], tremors and rigidity), impaired cognition (difficulties with concentration, planning and learning) and motivation.[194] Lesser compromises of the system due to lesions or chemical antagonists produce subsets of these same symptoms to varying degrees.

The way in which we have presented findings to this point may have misled the reader into thinking that neuroscience first arrived at a new understanding of reward in the brain, and then discovered by poking about in the deeper chemical details that dopamine is its agent. In fact, the dopaminergic pathways, and their effects on behavior, were isolated first – and some time ago – through animal experiments and the study of clinical populations. Why, exactly, motor control, learning, and motivation should have been linked through one central neurotransmitter was only gradually worked out, and fMRI investigations were crucial to the discovery process. In the course of this, a neuropsychologically adequate concept of reward finally, and only very recently indeed, snapped into view. Our claim to be able to understand addiction and AG rests essentially on this new enlightenment.

As recently as 2003 there was still disagreement on the interpretation of the findings. In the most complete survey of reward system science that had been done to that time, Berridge and Robinson[195] argued that dopaminergic response does not mediate reward learning in the sense of that phenomenon as it was then modeled (as “capacity to predict rewarding events based upon associative correlations”). Rather, they suggested, the system mediates “attribution of incentive salience to otherwise neutral events.” By this they mean that dopamine response to a stimulus is the basis for the organism focusing and maintaining attention (both perceptual and behavioral) on that stimulus. As Schroeder[196] explains “experiments in which the dopamine-releasing cells of the VTA projecting to the motor PFC were destroyed found that this impaired monkeys’ abilities to keep a prior intention in mind long enough to execute it after a delay.” Schroeder cites Durstewitz and colleagues[197] as suggesting “that the primary role of dopamine in the motor PFC is precisely to stabilize its goal-serving motor intentions against interference by other possible motor intentions that would be counter-productive. That is, information about expected reward is used to keep people and other animals like us focused upon our reward-directed intentions, to prevent us forgetting them or going off to do something else before carrying them out.”[198]

Non-technical synopsis: Two researchers argued a few years ago that the job of the dopamine system isn’t to learn to predict reward. Instead, they argued, it’s to pull something out of a background for attention and stay focused on it, not letting it slip back into the scenery again.

The reason it was somewhat natural for Berridge and Robinson to frame this idea as a rival to the reward-learning model of the dopaminergic system stems from the fact that reward still had not (in 1998) been fully disassociated from pleasure. Their survey[199] devoted substantial critical attention to establishing that dopamine projections to the Nacc and neostriatum are not needed for “normal hedonic evaluations, for hedonic modulation, or for learned adjustments in hedonic value”– which is quite so. They noted that incentive salience attribution corresponds roughly to what behavioral economists had dubbed ‘decision utility’, defined as “the degree to which a goal is chosen or sought”. They recognized, of course, that decision utility – ‘wanting’ – and hedonic appreciation – liking – are related. Thus they suggested that once decision utility is associated with a stimulus then “on each subsequent encounter … its capacity to support wanting is maintained or strengthened by associative reboosting of the incentive salience assigned to its representation. Reboosting happens when a wanted incentive is followed again by activation of hedonic liking”.

Non-technical synopsis: The reason the two researchers mentioned above thought that dopamine wasn’t for reward learning is that they knew it wasn’t responding (much) to pleasure – and they hadn’t quite realized how weakly related reward and pleasure are in the brain.

In retrospect this critical quarrel – as is usually the case with apparent scientific disagreements – was really a step along the road to completing consensus on the conceptual clarification of ‘reward’ and its distinctions from related phenomena. “The compatibility of specific reward learning models,” declared Berridge and Robinson, “hinges on precisely what … authors of reward learning hypotheses for dopamine function … mean by ‘expectation of reward’. If these reward learning hypotheses posit that dopamine mediates Pavolvian learning about incentive value alone (not stimulus-reward association per se, not hedonic value, not cognitive expectation of outcome) and that anticipatory neuronal responses to conditioned stimuli occur because those stimuli carry incentive motivational qualities, such as incentive salience, separately from mere association or from hedonic qualities, then those reward learning hypotheses are compatible with our incentive salience hypothesis … If, however, dopamine neurons are posited to mediate any other sense of ‘expectation of reward’ … then either the reward learning hypothesis or the incentive salience hypothesis must be wrong” .

Non-technical synopsis: The researchers pointed out that we could, philosophically, separate reward from pleasure if we want to.

Berridge’s and Robinson’s invitation to proponents of the reward-learning hypothesis to refine their concept of reward met, ultimately, with a favourable response. The first paper to state the mature model of the dopaminergic system, by McClure and colleagues,[200] essentially fused Berridge and Robinson’s incentive salience hypothesis with Berns and Montague’s predictor-valuation model (see section 3.1 above). All recent theories of the dopaminergic system, they pointed out, “agree on the singular principle that dopamine function is causally located between the identification of a potential future reward and the generation of action to pursue it”. However, the incentive salience hypothesis “leaves off after the identification of a goal.” What they mean by this is that it is a theory of the behavioral meaning of dopamine signals, but incorporates no account of learning except for the vague talk of ‘reboosting’ that Berridge and Robinson associate with it. This won’t do, since even if an initial dopamine response is taken to strongly bias selection of a goal, the system still must choose an action sequence, and choice in this context can’t be separated off from reward-valuation as expressed in relative motivational response. Temporal difference learning, as described in section 3.1, is precisely an algorithm for this, nudging the system towards actions that lead to better-than-predicted rewards. Here is where the predictor-valuation model can take up the slack: if the rate of dopamine neuron spiking encodes a prediction error signal, where incentive salience is then taken to be a function of the magnitudes of these errors, then the gap in Berridge and Robinson’s model is closed. “We propose,” say McClure et al, “that the concept of incentive salience is the expected future reward … In addition, we propose that the role of dopamine in learning to attribute such expectations is to situations that are predictive of reward … and in biasing action selection toward such situations … serves as the formal counterpart to the ideas of Berridge and Robinson … about the role of dopamine in attributing and using incentive salience”.

Non-technical synopsis: The research team who created the neuroeconomic model of the reward system pointed out that a system for pulling out objects of attention from the background and keeping the agent focused on them couldn’t work if it wasn’t tightly integrated with decision making. But once those things get integrated that just is learning about what counts as rewarding for the system. So the neuroeconomic model makes a prediction about what the dopamine neurons are up to. And if that prediction is right, the model explains the dopamine system.

As stated, here we have the current model of the reward system that has been confirmed by numerous experiments conducted over the past two years, several of which will be discussed below, and which will provide the basis for our later claim that AG will soon be a manageable health problem. The power of the model is nicely illustrated by application to a study – one directly relevant to gambling behavior – in which the interpreters failed to appreciate it. Zink and colleagues[201] compared skin conductance and BOLD responses (the name of the basic class of bloodflow variables measured under fMRI) of two groups of subjects. One group earned money by performing a task, while the other performed the same task but received equivalent expected quantities of money independently of task performance. As expected, significantly greater activity in the dopaminergic reward pathway (Nacc and caudate nucleus) was observed in the first group. Zink and colleagues interpret this as showing that dopaminergic activity mediates salience rather than expected future reward. Of course, they are quite correct about this if someone insists on interpreting ‘expected future reward’ as necessarily denoting the monetary value of the experimental outcomes. But this confuses what is rewarding to (as it were) the whole person with what is rewarding to the reward system. The point of McClure and coauthors’ mature version of the reward-predictor model is that what is rewarding to the reward system is incentive salience. Thus Zink’s ‘choice’ doesn’t arise: both interpretations are correct. For the first group of subjects in the Zink and co. experiment, the money was an incentive to action and therefore was salient – which is what ‘rewarding’ has turned out to mean in the brain. The second group (hedonically) enjoyed getting their money. Had they been previously unacquainted with what money can do, their subsequent use of it might then have been rewarding. But while choosing actions in the experimental task, their reward systems were uninterested in future hedonic prizes that were independent of subjects’ actions. It should be obvious that this is significant for understanding AG, although it will take some time to draw out the details of exactly how it is.

With the model in hand, experimental work can proceed to refine it. For example, as McClure, Berns and Montague[202] point out, the reward learning hypothesis does not distinguish between different possible dimensions of reward unpredictability (e.g.: What stimulus will occur next? When will the stimulus arrive? What quantity of stimulus will appear next?). They report an experiment that enabled them to isolate, under fMRI, predictions about what was expected by subjects within particular time frames. BOLD responses correlated with positive and negative errors against these temporally framed forecasts.[203] That is, subjects framed expectations that they’d receive rewards at certain times, and then their reward systems became most active at those times, because that’s when the comparisons between what was expected and what was received were made. The reader might suppose that this is a problem for Rachlin’s molar account of PG behavior as discussed in Chapter 1 section 7. If people’s reward systems are sensitive to prediction errors against fixed time frames, then why would gamblers be insensitive to differences between shorter and longer strings of losses between wins? However, as McClure and his coauthors note, their experimental setup was extremely simple – subjects simply waited for alternating squirts of water and fruit juice – and this might have encouraged subjects to look for simpler temporal reference functions than would be applied to more typical situations, including situations set up in gambling. The study just described is, for now, merely an illustration of the kind of experiment we will see many more of in the course of gradually refining the reward-learning / predictor-valuation model. As will be clear in section 3.6, development of effective anti-AG pharmacological therapies need not wait on these refinements, though the efficiency of therapies will no doubt be improved by them.

Indeed, another recent refinement experiment is very good news for Rachlin’s hypothesis about AG behavior. Nieuwenhuis and colleagues(2004)[204] performed fMRI examination of subjects involved in gambling scenarios where both the possible kinds of outcomes available relative to one another (large monetary gain, small gain, no gain, large loss, small loss, no loss) and the specific monetary magnitudes at stake were varied independently. They found that subjects’ reward-system responses were much more sensitive to variations caused by context shifts than to variations in available absolute magnitudes. For example, when subjects shifted from a scenario (i) in which there were possibilities of both no loss and a large loss to a scenario (ii) in which there there was a possibility of no loss but no possibility of a large loss, their reward systems were more active in scenario (i), and – the important part here – this difference was greater than the difference between activation levels when they switched from scenarios where possible losses were large to scenarios where possible losses were small. Thus the reward system is more alert to constrasts in what could happen than in magnitudes of possible changes in wealth levels relative to the status quo. The exact logical mirror of this is a person who is more interested in a change in the excitement level of a gamble than in a change of odds in her favor. This is immensely interesting to students of AG, since the experimental subjects here were randomly sampled people, not PGs. Equally interesting, from the perspective of our interest in AG and in light of Rachlin’s hypothesis, is the fact that some parts of the reward system differentially responded only to binary distinctions between outcomes (i.e., winning versus losing) and not to magnitude gradations at all. This is among the most positively suggestive neuroeconomic findings that a defender of Rachlin’s hypothesis could hope for. The sort of experiment it suggests – performing comparative fMRI examinations of normal and PG subjects while they gamble and the length of strings of losses between wins is experimentally manipulated – would directly test Rachlin’s theory of PG. If Rachlin’s theory were confirmed by such an experiment, we’d then not only have new grounds for thinking that Rachlin is right, but the model of the reward system on wh

THE BEHAVIORAL ECONOMICS AND NEUROECONOMICS OF DISORDERED GAMBLING: A POLICY-FOCUSED SURVEY OF RESEARCH

TABLE OF CONTENTS

EXECUTIVE SUMMARY

Chapter One

Chapter Two

Chapter Three

Chapter Four

1. Gambling, Addiction and Behavioral Economics

1.1. Gambling in scientific focus

1.2. Behavioral economics

1.3. Behavioral economics of substance use and addiction

1.4. The matching law and temporal discounting

1.5. Temporal discounting, substance use, and addiction

1.6. Three BE theories of addiction

1.7. Extensions of BE research to gambling

1.8. How many kinds of disordered gambling are there?

2. Behavioral Economics, Impulsivity, Addiction and Disordered Gambling

2.1. Impulsivity and addiction

2.2. Discounting: delay and probability

2.3. Impulsivity and Disordered Gamblers

2.4. Conclusion

3. The Neuroeconomics and General Neuroscience of Pathological Gambling

3.1. Neuroeconomics

3.2. The Reward System