Nudge plus networks
Policymakers need to acknowledge the part that social networks play in changing people's behaviour, argues Paul Ormerod, even if it does make their lives more complicated.
Over the past 60 years, many of the functions previously within the domain of the third or private sectors have gradually been embraced by the state. Meanwhile, generations of policymakers have been raised to have a mechanistic view of the world.
This is a comforting environment in which to operate - seemingly predictable and controllable - but this approach has not delivered anything like the success imagined by the founding fathers of the welfare state. While the inefficient use of public resources plays its part, the principal cause of the failure of what we might call the social democratic model is not the size of the state but the intellectual framework in which it operates.
The world is more complex and less controllable than 'rational' planners believe. There are two main reasons for this. First, as behavioural economics tells us, agents - be they individuals, institutions or governments - do not necessarily behave rationally; their responses when confronted by new information or a different set of incentives may be hard to anticipate.
Second, as the study of networks shows, our tastes and preferences can be altered directly by the behaviour of others and can change over time. Natural selection is now believed to favour social learning strategies that specify when and whom to copy. It seems that humans are particularly adept at this.
By treating individuals like so many Robinson Crusoes taking independent, autonomous decisions, traditional economic theory fails to account for the social dimension of human activity.
We have made great strides in developing our scientific knowledge about behavioural economics and network effects over the past couple of decades. But how far has this actually shaped our approach to public policy?
In opposition, leading Conservatives urged their MPs to use their summer break to read Nudge, Cass Sunstein's and Richard Thaler's bestselling 2008 book on behavioural economics. Two years on and the coalition government has set up a ‘behavioural insight team', having apparently embraced nudge approaches as part of its policy armoury.
Much more significant is the lack of attention given to networks. This is despite a rapidly growing body of research telling us more about their considerable potential to complicate, disrupt and deliver policy intentions. Meanwhile, new technologies, global communications and challenges such as climate change suggest that network effects are becoming ever more important.
Some - such as peer influence - are likely to be relatively direct, obvious and familiar to us. Likewise, most of us are aware that the way in which people affect us varies according to context: the people who sway our choice of pension are unlikely to be the same as those whose behaviour encourages us to binge drink. The policymaker's challenge is to translate this commonsense understanding of the power of networks into workable interventions and to be able to factor in much more complex effects.
We can start with a basic model; we have a group of individuals contemplating whether or not to buy a product or undertake an action. Each has an intrinsic preference for what is on offer, so will decide to buy or act at different levels of ‘price' (which here has a much wider, multi-dimensional meaning). In a non-networked world, a policy intervention in this scenario is simple: put the price up and demand will fall.
But what happens if agents base their decisions in part on the actions of others? In this scenario, increasing price initially has no effect. Then, suddenly, for any further small increase in price, we get a bigger change in demand than would take place in the absence of network effects. The risk is that, unless network effects have been factored in, the policy may be abandoned too quickly, as increasing the price will be seen as having failed. Ignoring networks can leave the policymaker 'flying blind', relying on inaccurate evaluations and repeating (often expensive) mistakes.
To make life complicated, a decision made by an agent may cascade across a network, so that someone directly influenced by them goes on to influence someone else whom the original agent does not know, and so on. Understanding 'cascades' and the conditions under which they might arise is one of the principal challenges of network theory.
Game-changing possibilities
Duncan Watts, now director of the Human Social Dynamics Group at Yahoo!, published a brilliant article in 2002 - the practical implications of which are profound - when he was a sociology professor at Columbia. He was interested in what happens when the only thing that affects how agents choose between alternatives are the choices of others.
We can think of this model as a game with simple rules. One rule is that we connect agents to one another randomly. This does not mean that those to whom you are connected will necessarily affect your choice, but that they have the potential to do so.
This way of connecting agents may seem unrealistic but does, in fact, offer a reasonable approximation to many practical situations. Epidemics, for example, are often spread by random contact. In a strange city, you go out to eat and observe two similar restaurants near to each other, one that has plenty of people in it and one that is nearly empty. A sensible decision would be to eat at the one with more people in it. In financial markets, one trader may monitor a small number of others closely, but if the market starts to move strongly in one direction as a result of the decisions of many people entirely unknown to the trader, a sensible decision might be to follow this trend.
Another rule in Watts' game is that each agent has a choice between two alternatives: A or B. When the game starts, we assume all agents have chosen A. Watts then made the entirely realistic assumption that each agent differs in his or her intrinsic willingness to switch to B: their ‘persuadability' is weighted accordingly. The game is played many times with identical rules. The only difference lies in the agents chosen at random to switch to B at the start. A crucial point is that the same number of agents are selected to switch from A to B each time, so the initial shock to the system is exactly the same.
The result of any given 'play' of the game may be sensitive to the particular circumstances. For example, suppose the agents selected to make the initial switch from A to B were connected to agents whom it was very hard to persuade: that is, they required almost everyone who might influence them to choose B before they themselves did. The ‘cascade' - the spread across the network of people choosing B rather than A - may well stop there and then.
So, for example, the degree of optimism or pessimism that firms feel at any point in time is an important determinant of the boom and bust of the business cycle. We can think of a firm in state A as optimistic. The economy receives a small adverse shock.
A few firms switch to state B: pessimistic. How many others will abandon their optimism? If enough do so, the economy will move from boom to bust.
Well, yes and no. The same small initial disturbance can have dramatically different outcomes. Most of the time, the initial switch by a small number of agents from A to B does not spread very far. Occasionally, however, there wll be a cascade across the system and most agents will end up with B.
But identifying the initial cause of a strong effect can be very hard in networked systems. 'Black Monday' of 19 October 1987 is a pertinent example. Suddenly, and apparently inexplicably, stock markets around the world crashed. Various accounts have been given to explain what happened but, more than 20 years later, there is no consensus. What we do know is that traders on stock markets receive a large number of potential shocks in the form of new information. Each piece of new information has the potential to trigger a large cascade. Few do. For the most part, the disturbances are contained by the robustness of the network. Every so often, however, the system proves fragile.
So, systems of interconnected agents, each of whom influences the behaviour of the others, are both robust and fragile. There are many possible reasons for economic recessions, but the 'animal spirits' of Keynes - the waves of optimism and pessimism as they either spread or are contained across the economy - are a powerful factor. Network models seem essential to our understanding of recessions, and especially those involving financial crashes.
Nudge plus networks means that if you have some understanding, albeit imperfect, of the network structure and flows, only a small number of people need to be nudged, yet the number who eventually change their behaviour could be enormous. This represents a potentially huge increase in the ability of policy to affect outcomes. But do we need to know the exact structure of a network before we can begin to think about the potential impact of policy changes?
In the late 1990s, a group of epidemiologists, sociologists and physicists analysed a database of individuals and their sexual contacts. The results were published in Nature, one of the world's leading scientific journals. They found that most people have only a few sexual partners, but that a small number have hundreds or even thousands. The real originality of the paper was its finding that the structure of the pattern of the contacts closely reflected a recently discovered type of network that is described as ‘scale-free'.
Such networks are important in the natural sciences, and more of them - at least, good approximations of the scale-free pattern - have been discovered in the human world. The internet, for example, has these properties. A few sites receive a massive number of hits, while most get very few. A whole industry has grown up in American marketing circles trying to find these influential 'hubs'.
Certainly, if the network is not random but scale-free, persuading one of these hubs - agents with large numbers of connections - to change their choices will make a big difference to the eventual outcome. They have the capacity to influence many other individuals. Because they are connected to such a large number of agents, the chances, for example, of their being connected to others who are easily persuaded is very high. Such networks may be important in the spread of ideologies and beliefs, where a small number of charismatic individuals might be decisive in persuading others to adopt their views.
For policymakers, this type of network presents, once again, both an opportunity and a challenge. If the actual network of interest is fairly similar to a scale-free one, then the task of persuasion - getting people to adopt different behaviours or make different choices - is made much easier if some of the hubs can be nudged into this.
There is, of course, the problem of identifying who these might be. If the network turns out not to be scale-free, then a strategy based on the view that it is will be unlikely to prove effective. In crowd control, for example, the police or military may believe there are a few 'ringleaders' and aim to nullify them in some way. But as a strategy, this is rarely seen to work, precisely because the assumption that a scale-free network underlies crowd behaviour problems is usually wrong.
The real problem for authorities arises in cases in which the network is scale-free and the aim is to prevent a particular form of behaviour from spreading. Consider the earlier example: the potential spread of an epidemic. If agents are connected at random, inoculation or influencing people not to take up the riskier mode of behaviour can be effective. In general, not everyone needs to be inoculated in order to prevent the ‘virus' from spreading. There may be occasional local outbreaks, but if a sufficient number of people are inoculated against adopting this mode of behaviour, it will die out. This is definitely not the case for scale-free networks. If the hubs are targeted, then literally every single one has to be caught in order to suppress the spread across the system, which is a difficult task.
Another important type of network that makes life even more complicated is the ‘small-world' network. When we delve into the maths, there are considerable similarities between a scale-free and a small-world network. But their basic social structure is different. In the scale-free network, there are a few agents who have huge potential influence. The small world is much more like overlapping sets of ‘friends of friends'. The additional feature is that, while no one has a large number of connections, a few agents may have ‘long-range' connections to others who are remote from their immediate cliques. However, these individuals may be even harder to identify in practice than the hubs of a scale-free network, precisely because they themselves are not distinguished by having an unusual number of connections.
Network theory in practice
Random, scale-free, small-world; each of these networks has been shown to exist in a range of contexts. While it can certainly help to identify what structure is present, in practice we rarely have accurate information about the precise nature of people's relationships. When it comes to practical policymaking, we inevitably have to rely on approximations. The good news is that old-fashioned survey research, combined with the modern, computer-oriented methodology of agent-based modelling, often enables us to get a reasonable estimate of the type of network that is relevant in any given context.
For example, with Greg Wiltshire, I recently undertook research on binge drinking in the UK. While there is some strong evidence that social networks are important in this activity, much of the literature around the issue neglects the potential role of peer acceptance in the sudden and rapid rise in binge drinking. Our work showed that there is clearly a dramatic difference between the perceived behaviour of the friends of those identified as binge drinkers and those who are not. It would be curious, to say the least, if large numbers of young people had suddenly decided quite independently of one another to binge drink, and then had happened to congregate in friendship networks. So, while the existence of a contagion effect among friendship networks is not technically demonstrated, it seems a far more likely explanation.
The important point here is that policymakers do not need to be supplied with the full details of a network to gain information about its fundamental structure.
A different example is given by studies of the world trade web, the import/export connections between countries. Such connections are interesting not just for their intrinsic importance, but also because the spread of financial crises often follows the pattern of trade flows between countries. The more closely countries are connected by trade, the more likely it is that a currency or financial crisis will spread between them.
Direct bilateral-trade relationships can explain a small fraction of the impact that an economic shock originating in a given country can have on one that is not among its direct-trade partners. A network analysis of the world trade web can go far beyond the scope of standard international-trade indicators, which only account for bilateral-trade direct linkages. An understanding of this network does not just provide information to policymakers about how shocks might cascade, it also offers a different and powerful way of thinking about the effects of, say, introducing trade barriers. Such policies may be of interest to the developing, if not the developed, world.
There are circumstances in which the classic 'rational' economic approach still has traction. However, there is inherent uncertainty about the impact of policy in a world in which network effects are important, which no amount of cleverness can overcome. This is not a comfortable world for the policymaker. But ignoring network effects means that we carry on with the same model, spending vast amounts of money, with at best a hit-or-miss success rate, as the evidence of the past 60-odd years shows.
One possible implication to be drawn from the networked view of the world is that little or nothing should be done, on the grounds that we have little or no idea of the eventual consequences of introducing any particular policy. Far from it.
The RSA's Social Brain project has started to explore the implications of individuals being more aware of their own cognitive frailties. Early findings have indicated that giving people the tools to understand how their brains, behaviours and environments interact helps them make better decisions and tackle habits such as smoking, binge drinking and overeating. The same kinds of questions can be asked about networks. It is arguable that understanding network effects - how our connections and other people's behaviour influence us - can empower people to make different and better choices.
More broadly, this changes not just the focus and design of public policy, but also the way we think about success and failure. Understanding and using networks can make a significant contribution to tapping into civic capacity and meeting public-policy goals. But they are complex and the way they operate is unpredictable. Traditional policy interventions tend to be large-scale and expensive and aim for relatively marginal improvement in outcomes. They seek to minimise risk through systems of regulation, audit and accountability. These design features do not fit the characteristics of social network interventions, which will often fail or have unpredicted results. Occasionally, however, small interventions will have a major impact through contagion effects.
When it comes to contemporary challenges - climate change being an obvious example - we will often need to induce dramatic mass behaviour change. We will not do so using simple, incentive-based approaches, and we need to get better at harnessing the power of networks. The potential gains are enormous.
Paul Ormerod is an economist, author and director of Volterra Consulting.
This article is based on N Squared - Public policy and the power of networks, one of a series of essays commissioned by the RSA that explore the concept of 21st century enlightenment.
| Connected Communities The government faces a challenge in delivering its Big Society vision in deprived neighbourhoods, according to a report published by the RSA a year into its Connected Communities project. In one of the project's research areas - New Cross Gate in London - a quarter of the 280 people involved could not name anyone in their social network who they thought was good at bringing people together or could help them contact someone influential locally. Meanwhile, more people recognised and valued their postman than their local councillor: the report suggests that ‘familiar strangers' such as these may be an underused community resource. The report details the first year of the project, which is working with two communities at a local level: New Cross Gate and Knowle West in Bristol. By mapping existing networks, the project has aimed to develop a new way of understanding social networks. In the project's second year, the RSA will use these findings when working in partnership with local residents to co-design and test network-based interventions aimed at addressing local problems. What is already clear is that a ‘community' cannot simply be defined geographically and that a fresh approach to local regeneration, based on mapping social networks as forensically as possible, is required. This knowledge offers a new way of designing government policy, one in which small interventions have the potential to make a big impact through network effects. Find out more Read the report Connected Communities: How social networks power and sustain the Big Society by Jonathan Rowson, Steve Broome and Alasdair Jones |
Share